Yury Molodtsov

The Age of the Polymath

Thu, 14 May 2026 20:19:00 +0000

I’ve tried working in so many areas. I used to code and wanted to become a developer. I did a bit of design. I love photography. I did venture investing. The outcome of this is a career as a generalist. And being a little bit of everything sometimes felt like a disadvantage because the world rewarded the people around who went deep on something.

It turned out the breadth was useful anyway. I could talk to anyone and translate the ideas between engineers, executives, and reporters. That’s what I’m doing daily working in comms. But I always wondered what I would have built if I’d just picked one thing.

Not anymore. Because using AI and agentic engineering to their full potential requires having at least the basics of all those skills simultaneously. The things I knew a little about — code, design, copy, marketing — suddenly composed and compounded.

“An app designed by a developer” has been a meme for a reason. You’ve likely seen them and used them. They exist because developers know how to code and build software, even if they don’t know how to design it. As long as the app was uniquely useful, people would take it anyway. Meanwhile, designers couldn’t build much on their own. PMs can’t do much without either role.

With AI, the breadth of knowledge finally pays off. What matters is having ideas and applying taste, Rick Rubin-style. You no longer have to be an expert in all of these areas. But you must come up with ideas to build, and it’s up to you how great they turn out.

In my view, vibe-coded apps can be extremely polished and pleasant to use. In fact, far better than 90% of the software you might have on your computer or in your browser. Apple’s platforms have long had a culture of small, polished apps and tools, but even those are quite rare these days. And most web apps are quite primitive, especially in the B2B space, which is why companies like Linear have been so successful.

For the best result, it definitely helps to know what NextJS, Tailwind, Postgres, and many other words mean. Agents can make all the choices for you, but if you can’t evaluate them, there could be issues.

Any time I’m building a new side project these days, I get it to a place where I’d feel great releasing it to the public, even if it’s just for myself. I just can’t do it any other way. That means a polished UI, animated interactions, fine-tuned light and dark modes, beautiful custom icons, etc. I use these apps daily, so I want them to be great.

I only stop short of releasing the app when there are specific reasons. I’m writing this post in Elsendo, a web-based note-taking app I created with shareable links. That implies both server costs and an infinite potential for illegal content from piracy to CSAM. That’s not something I want to deal with in a free app and since it’s minimal, I’m not sure many people would pay for this, although the new owner of Evernote presumably believes it’s worth $250 a year. But I wanted this exact thing for years, nobody built it and now I have it. It’s open source.

Gathering fascinating bits and links is my favorite thing ever, which is why I’ve recently built Mimir, a knowledge management app that aggregates articles, links, highlights, and more in one app. It has browser extensions and connects to X and Readwise APIs. Add Stripe, and it’d be ready to go. But it’s too heavily inspired by a few existing tools, plus I want ultimate freedom in how it will work. So it’s just for me.

This is the best part about building your own apps. Need a feature? You don’t have to wait or pester the developer, who likely has other ideas. Just build it. Some are harder than others but there’s a good chance the agent will do this for you. And running such an app for yourself isn’t a big burden. Occasionally, you see a bug and fix it. Over time, there are fewer and fewer of them. Or none.

Most of my things are public. Since the previous post on this, I launched Syndicator, my intelligent RSS reader (think of it as a modern Google Reader replacement). And there are smaller tools and utilities, like a fast natural language timezones converter or MacPod, a virtual iPod that controls music on your Mac’s screen. I love reading Hacker News sometimes but never liked their 2000-era website or any of the third-party clients, so I built Hack and Cheese.

Some of those were built in 30 minutes on a Sunday evening. I usually had to spend more time on polishing in the next days but I continue using it, so polishing, adding obvious features, and fixing bugs is a good way to spend time for me. Because I actually enjoy this process a lot!

Here are some things I learned.

***

Models. Opus and Codex have never been closer, but Opus still excels in design. It’s a pricier option in terms of limits, so you can ask Opus to plan the app and create visual mockups of the interface, then ask Codex to build on top of them. Without supervision and hand-holding, it adds too many gradients, uniform buttons, useless text labels, and other noise.

Agents. I read about people who just build things with OpenClaw and don’t understand how they do it. Yesterday, I finally pulled the trigger on building my own little Pinterest and decided I’d just give it to OpenClaw. Despite using the same GPT 5.5 model as Codex, it ignored my mockups and made extremely silly choices, like wiring the context menu buttons to a POST delete request. Claude Code or Codex would never do something like this, so use them — the harness matters a lot.

Serverless. Vercel and Netlify are great platforms for little web apps because they are essentially free until you get big (but also rapidly become very expensive afterward, so be careful). You will also have to pay for most additional features, sometimes quite basic ones, like a database. Most of my side projects are hosted there.

VPS. Don’t be afraid of the VPS. Agents can set everything up and help you manage them. I recommend Hetzner because it is roughly three times cheaper than DigitalOcean today. Be mindful that the default agent behavior can expose vulnerabilities through open ports and user-accessible APIs. And on a VPS, all those problems will be exploited. The first time I launched a server, it was taken over by a mining bot in about 10 minutes. My best practices for deployments are in agents.md now. Push your agent to think about it and close the ports.

You can also take any local agent or OpenClaw and ask it to pentest your app. Then use this report to fix things.

***

The rise of agents poses many interesting questions we can’t answer yet. For one, it definitely seems that the value of these vibe-coded apps is unproven. Some become more popular than others and get engagement, but as a few creators noted on X, competing is extremely difficult because “competitors” can copy most of your features the next day. Everyone is building a Markdown editor these days (I also have my own). Yes, it’s become easier for you, but also for everyone else. So you can only win through superior distribution, attention to detail, or by making apps that do truly complicated things nobody can repeat with two prompts.

Scalability is another question. Every other month, Vercel generates a social media story about some creator getting a crazy bill, and then the CEO has to come down personally and void it. But the apps at least scale and work. If you create something that gets millions of users as a monolith on a VPS or AWS, can agents help you scale it?

Finally, we’re seeing more and more people in all roles being pushed to contribute to the codebase. Designers, PMs, and non-technical CEOs themselves are getting into development. It will clearly become an expectation for most people in tech companies, but does it mean the wall around software development as a role will become more permeable as well, or not?

And will the next Notion or Linear start as a vibe-coded product or is it always going to be limited to people building personal projects or adding features to existing ones?

It’s an extremely interesting and exciting time. Especially for polymaths. I know the word was (is) overused in Silicon Valley and is too generous for this concept, but no better word exists. And yes, there are true polymaths out there, people who can build delightful full-stack apps with great design and copy without any AI and I’m fascinated by their ability.

But if you’ve always known just a little code, a little design, a little writing — this is a great moment for you. Because you don’t need to know how to write code, you just need to know what code can do.

We’re in the age of the polymath.

We Need a New Product Hunt

Sat, 18 Apr 2026 16:52:00 +0000

This is Product Hunt’s main page today. 4 out of 5 places are taken by extremely large companies that need little promotion: Anthropic, Google, Cloudflare and Vercel.

Product Hunt used to be a website I opened daily looking for cool new products. Some of them were just nice personal tools and some eventually grew into billion-dollar companies (I found Truebill there during their launch). But since then, it has turned into a marketing platform for big tech startups. Sometimes it’s genuine products, but more often than not, it’s mundane updates grouped under a “Pied Piper 2.0” or just the most basic features.

The only thing that matters is how much attention you can manufacture around the vote. There’s an entire industry around ProductHunt. Infinite guides, professional “hunters”, workflows to ensure you will get hundreds of upvotes right away just so bystanders could organically see you and start voting. Founders and marketing teams plan it for weeks and blast across all of their channels on the day.

So I stopped opening Product Hunt years ago because I rarely saw anything interesting anymore. A true indy project has little hope. Which is why we need a new Product Hunt.

Product Hunt started in November 2013 as a scrappy side project by Ryan Hoover. He liked browsing App Store charts in other countries and exploring AngelList to see what people were building (I get that feeling) but there was no single destination for discovering the new product launches. Product Hunt initially launched as an email list and later turned into one of the main niche tech websites and grew until AngelList acquired it in 2016 for $20 million. It is still the default launchpad for new tech products, which are accelerating even more with agentic engineering.

What changed the most is the ecosystem. Back then, the information was scarce and products were rare. ProductHunt could legitimately serve as the hub for all the day’s key product launches. It was never the only choice even then. Hacker News: Show was an early inspiration, and there were true alternatives like Betalist and many others. But none of the alternatives have been able to elevate the launch into a separate event.

Right now, we’re drowning in new companies and products. Not a day goes by that people aren’t hyping up yet another thing that is bound to replace the previous thing. Aggregating all of them is a futile exercise because what matters is not the discovery but curation. Any catalog built around voting will inevitably attract the same marketers that Product Hunt suffers from. As long as it’s about the headcount and attention, large companies with social media presence will always win. Which leads us to the question: can we even have a new Product Hunt?

One particular change that could fix this problem is tying the product to the company. Imagine that you could submit your company only once. Have another product? Feel free to post about it, but it won’t take a slot. Then neither Google nor Vercel would compete with indie developers. Makers could break through, and marketing teams won’t be able to repackage their existing products countless times. You could also go authoritarian and limit who can vote, though Product Hunt already prioritizes old, verified users over new signups, especially on launch days.

The more I think about it, the more it seems that we will never have a true replacement. But what we can have is individuals and communities curating their app collections. This removes the promotion issue altogether. It adds other issues, most importantly about access — as a founder, you can’t post your project on someone’s blog. You need to make the authors interested in some way.

We’re shifting from a centralized directory to a fragmented ecosystem of tastemakers. And the biggest problem for consumers is finding the people to follow.

I watch A Better Computer, an awesome channel by Matt Birchler, who figured out there was an almost unoccupied niche in talking about software and not gadgets on YouTube. I discovered and subscribed to MacStories because they took a very different route compared to most Apple-centric websites and focused on unique apps. I also subcribe to Internet is Beatiful, which shares hidden internet gems along with useful tools and products. There are niche subreddits on all sorts of topics related to computers.

Maybe one day, one of these platforms or another will become a new Product Hunt. But I’m doubtful for the same reasons I outlined earlier. Most likely we will just have no tech monoculture anymore apart from the Twitter timeline. The algorithm will decide whose launch everyone will see tomorrow.

AI Won’t Kill SaaS But It Might Kill The Messy Middle

Mon, 16 Mar 2026 16:50:00 +0000

AI won’t kill the SaaS models. OpenAI itself famously runs on Slack and uses countless off-the-shelf tools. There’s simply no point in rebuilding key apps your entire workflow depends on with vibe coding. You will be just wasting resources.

Especially considering how difficult it is to build production apps. After a few experiments, I’m now primarily working on two specific applications. And like all software, they’re never finished. I’m making daily commits to both.

But not all apps are essential or that demanding. I believe two sides of the spectrum are safe:

Simple tools that solve a specific problem and don’t cost much
Expensive tools that act as your system of record (chats, CRM, ERP, etc)

It’s the messy middle that has to be worried.

I used to manage a WordPress website at a previous job. Cheap to run. Countless plug-ins and vendors. Easy to add the content you planned to add. But as soon as you want to change anything else you have to spend hours or hire that agency again (great business model for them, honestly).

So when an agency built our new website in Webflow, I loved it. Not only was it easy to add content, but I could also easily create a blog, add a sign-up form, and update multiple pages as we grew.

But over the past year, I had grown increasingly annoyed with it. Webflow is not cheap, especially at the CMS layer which is the entire point of their app if you ask me. It has also gotten much slower, specifically that CMS editor, while their “Designer” is somehow still quick enough (one would think it’s the more complicated part).

So I fired up Codex, asked it to download our entire website and rebuild it on a modern and portable tech stack. I chatted about options and chose Astro and Vercel. After an hour of polishing, our new website was running. That might sound like a long time, but instead of keeping the old visuals, I asked it to rebuild it completely from scratch.

Then I spent about 4 hours updating the content, since I was now able to do so via the agent, which is just wonderful. Instead of being stuck in the CMS, I can just ask it to do things for me, like updating all the listed publications with the current metadata.

Webflow has become that messy middle. If you want a quick website, there are fully managed platforms: Substack, Beehiiv, Ghost, Shopify, etc. And if you want your own thing, you can build it with AI.

The question is, who’s going to be technical enough to use Webflow and build entire pages from scratch, but not technical enough to use Claude Code or Codex?

That market will shrink into nothing.

Building Personal Software

Wed, 04 Feb 2026 18:30:00 +0000

My feeds are filled with posts praising Claude Code. People claim they feel bad if they don’t have an agent running 24/7. Developers are asking for ways to manage more and more agents at once through apps like Conductor.

The comments are always the same: “What have you actually shipped?”

Fair question. I built several apps that I use practically every day and put the source code on my GitHub. I’ll share more on them later but first I wanted to give a bit of background.

I started programming as a teenager. Pascal, Delphi, eventually moved on PHP and spent a year building my own CMS (who didn’t back then). I loved coding. That magical feeling when the words you write make the machine do something. But I had other interests and my career moved me elsewhere so I mostly skipped the switch to modern web development with node.js and React.

AI Engineering got me back big time. Claude Code with Opus 4.5 was specifically wonderful. It’s the same magical feeling once again, except now I can code in English. Still useful to have the basic idea of how things worked and what specific technologies and frameworks can enable. I only have a glimpse of those and I can already build things I couldn’t do just a few years ago.

The barriers to programming have never been lower. That doesn’t necessarily mean you would land a Software Engineer position at a tech company tomorrow. I don’t think you will. But I can certainly see product people and designers jumping straight to code without waiting for engineers.

What you can do instead is build the software for yourself. The apps you always wanted the market wasn’t able to produce. Nobody can stop you. This is exactly what I did.

I like Safari’s Reading Mode. It’s the best one out there. Chrome never had a good one. So I built 📖 Readr, a Safari-like Reading mode for Google Chrome and Chromium-based browsers. There are several extensions that try to solve this but all look like they were built by a developer inspired by a 2004 Linux distro. A million settings and none can make them look good. Readr has zero settings and simply uses your system font and light/dark mode.

For many years I wanted to have a minimal photo blog to document my trips and projects. Flickr and 500px are very old and push you to their subscriptions. I knew it was possible to build with Hugo but I never had time to build what I actually wanted and deal with photo processing and visibility scopes. With Claude Code, I built it in one evening. It’s 🌁 Photowall, a clean and minimal Hugo theme for photographers with a unified masonry feed. Here’s my photo blog running it.

I spend most of my day in the browser. All the apps are there. But it’s a distracting environment. Sometimes I just want to have a full-screen note where I can scribble. So I built 📝 Flow, a minimal one-page Markdown-enabled notes app. It’s completely private because the data is stored directly in your browser. Don’t store anything valuable in there but it’s great for temporary things.

I like Readwise and Readwise Reader is my read-later app of choice. But I very much dislike Readwise’s web interface. It’s clunky and outdated and I can never figured out where each particular highlight came from in the first place. Thankfully, it has an API so I built my own frontend 📔 ClipSavvy. You can even use this deployment by bringing your API key from here (or just deploy it yourself).

These apps aren’t groundbreaking. Any developer would be able to build any of them with minimal time. But it’d have taken me a lot longer. And even developers confirm the barrier to engage has lowered so much they’re starting side projects and experiments they didn’t previously have time for.

***

If you’re thinking of building something for yourself, here’s the framework I came up with.

Easy to vibe code: anything without persistent data storage or authentication. Local-first tools, apps relying on third-party APIs, static sites. Something you ship but don’t have to manage.

Hard to vibe code: anything multi-user with background jobs and synced data. Yes, Supabase exists and you can put a task manager on it quite easily but a multi-user RSS reader would be difficult. You have to manage job queues, deal with bandwidth issues, manage costs, etc.

OpenClaw is a perfect example. It’s an incredible complex piece of software but it doesn’t matter since users have to run and manage it themselves.

Considering this, there are several interesting directions to explore:

Chrome extensions. Most people use Chrome. You don’t need the cloud. You can actually get organic users. There are entire businesses built on top of Chrome Web Store.
Native macOS/iOS apps. Claude Code can build them as well. You get multi-device sync with CloudKit for free. There are apps for everything but many of them feel scammy and users are annoyed with subscriptions.

The real unlock is building the specific tool nobody bothered to make because the market was too small. Now the market of one is viable.

The AI Price Hike

Sun, 07 Dec 2025 11:20:00 +0000

We should be in a recession. Interest rates jumped after COVID, the war and tariffs ruin global trade, while western countries are struggling with expensive energy, social spending, and competition from China. But the economy is growing and markets are breaking records. This is happening because of AI: it explains roughly 85% of US equity gains this year, and about half of the S&P 500 stocks are associated with AI.

Everyone comparing it to the dotcom bubble either doesn’t know about what happened in 2000 or misunderstands it. Even though the internet was early, for some reason people believed it’d change everything in a year or two. Investors poured millions into any companies with the world “internet” in their boilerplate despite little to revenue, relying on users alone. And by users I literally mean “signups” because nobody tracked Daily or Monthly Active Users back then.

Most of Mark Cuban’s fortune came from his dotcom project Audionet, a streaming platform for audio and video that he sold to Yahoo! for $5.7Bn in 1999, just in time before the party ended. You know what revenue they had? $13.5 million in the last quarter before the sale.

This is a typical S-curve of expectations: investors overestimated short-term effects but underestimated long-term ones. Because the internet did change the world and the tech companies dethroned oil corporations as world’s most expensive businesses. Pets.com launched in 1998 to delivery pet food and toys and shut down two years later— but the model is reasonable and you can easily order those things online today. They simply were too early and tried to induce the demand that wasn’t there yet.

And unlike the dotcom companies, AI labs are printing money. OpenAI and Anthropic collectively generate $25 billion in revenue. These are the fastest-growing companies in the history of the world.

Source: Deedy Das

People argue whether Nvidia is going to bust, but its $4.4T market cap is far more reasonable when you look at their revenue — the just earn a lot of money. Nvidia trades around 45 P/E, which is within benchmarks for their industry and growth rates.

These companies are benefitting from AI directly by selling inference (directly or not). If we trust Altman, he said that OpenAI is profitable on inference and only loses money on training. But what about others, the ones who have to pay for these APIs and the included margins?

If you look around, AI seems to become the biggest differentiator for tech products and yet all of them have it. Task managers and notetakers that existed for years before have suddenly become AI-first if you look at the landing pages. Many apps and platforms have pivoted their entire businesses because the new opportunities are far bigger.

Replit started in 2016 as a web-first IDE available in the cloud, primarily targeting students. Last year they had slightly over $2M in ARR. This year, Replit shifted from a coding IDE to an AI-first agentic platform for creating app and suddenly grew to $150M in ARR. It is now projecting $1 billion in revenue by the end of 2026. The demand for AI coding is so massive even minor players get a lot of interest.

Cursor, an agentic IDE based on VS Code, was launched in 2023. This November they raised a funding round valuing them at $30Bn. That’s probably three times more than the entire company of JetBrains on the back of Cursor’s enormous revenue growth. Developers just love Cursor and can’t get enough of it.

But these are apps for developers. Software development is the most obvious use case for LLMs beside general knowledge retrieval and far more lucrative. Developers cost a lot already so anything that makes them more efficient is a good leverage and businesses are more willing to spend on it. Besides, programming languages have something that natural ones like English don’t — formal verification. The code is code, either it works or doesn’t, which makes it easier to iterate and improve LLMs’ capacity to produce code.

What about others? We’re not shy of examples, since every app is integrating AI now. Google pushes Gemini in Gmail and Docs. Notion has Notion AI everywhere in the app. I’ve recently gotten AI in Slack and it tries to do minor things there and summarises the threads that took place while I was away. Grammarly acquired Superhuman and relaunched as an AI productivity suite under its brand.

For many companies, AI has been a great excuse to raise pricing. Our Notion pricing has grown by +25% per user. Google Workplace raised the price by ~$2 for everyone because we bundled Gemini. Do we use Notion AI? No, but we can’t avoid paying for it, the same for Google Workplace. I suppose the silver lining is that if you actually use Gemini a lot, all other business users basically subsidise you at the moment. That’s how it works: prices are higher across the board to cover the inference costs but if most users barely touch AI the company’s margins are also higher.

But we need to distinguish solving clear pains with AI and bolting it on top of an existing solution with the hope it will take off. Adding AI just because you can is a cargo cult. Raycast an indispensable Mac app, tried adding this to their subscription in the hope to justify it. Eventually, they gave up and let anyone add their own API keys.

People don’t want to pay ten times just to chat with a slightly different LLM in every app. For this, they can just use ChatGPT, which is OpenAI’s ultimate moat and the reason to exist. Apart from this, the real advantages of LLMs come not from chatbots but specific little features enabled by them.

That’s how you get a market where software got expensive across the board not because AI created value, but because companies used AI as cover to raise prices while their competitors were doing the same thing.

I talked to several operators across some of the largest tech scale-ups. All of them are frightened. The common belief is that you have to integrate AI if only because all of your competitors are doing so. It’s the Red Queen hypothesis where “it takes all the running you can do, to keep in the same place”. And they are not crazy, just look at Cursor again and all the competitors they were able to overtake solely because of AI.

Companies that were building products for years expect them to have a moat at least in terms of features and capabilities. Turns out, AI can replace the need for most of this functionality altogether and even if it’s not great yet, OpenAI, Google and Anthropic are burning billions on the next training run of the next SOTA model anyway. The models will only get better.

How Startups Get Viral

Wed, 19 Nov 2025 17:20:00 +0000

Tech people adore Stripe. Linear’s old branding infected GenAI so much all its landing pages look purple. Arc Browser received glowing coverages from top-tier outlets even for basic features.

Some companies get viral while their competitors struggle to break through the ice. Why?

This is a question that bothers many founders and marketers wishing to promote their companies or clients. What’s curious is it often have little to no connection to the financial success. Adyen is as big as Stripe, but only the latter is the Silicon Valley darling. There are immensely successful tech companies out there, entire unicorns whose founders have only a few hundred followers.

Arc is the perfect example of a viral product. The reason I know is because I’ve become a user and a fan, but before this, me and my team were leading communications for one of their competitors, Sidekick, which was acquired by Perplexity.

We were against not the wind, but a storm. Had to spend weeks coming up with dozens of ideas and pitching hundreds of reporters just to get something meaningful (and we had some great stories in the end). Meanwhile, Arc was getting covered by The Verge even for minor updates. There are many stories written about their brand and what made it work. It’s a combination of the top-down strategy, dedicated pros thinking about storytelling, which funnily enough was all about centering people to create a relatable brand. But also flawless execution, with every simple feature made with a degree of care few others would go to. All of this led to news stories like titled “Arc is the best web browser to come out in the last decade”.

At the time, Arc didn’t generate any revenue. The viral tech companies stand out not just for their financial success (many more companies are financially successful than viral), but for deeply resonant brand strategies and product philosophies coupled with excellent execution that their competitors often lack.

But Arc had a lot of users! First early adopters started talking, then influencers, then I saw it slowly spreading outside the US. All this demand served as a validation for reporters, who felt they were reporting on the phenomenon, not merely covering a feature launch. There are many companies whose products people find OK and few that they truly love. With Sidekick, we had 15,000 of them, and even though a significant share were already paying, it only constricted the number of people who could rave about us. Meanwhile, Arc’s fans were everywhere.

This is the first requirement for virality and the most obvious: users. But you need more than users, you need hardcore fans and you need them to be publicly vocal. I’d say that you need to have a concentrated set of users: better to capture a single tribe but completely, then expand afterward. Startup founders and VCs are one example of a good audience, because they can pay and have a lot of clout. That’s why apps like Superhuman and or Granola were so successful in capturing them. And suddenly everyone uses the “thing” and you’re behind, which creates FOMO for others. Raving users are the first pillar of virality.

Arc was built by the company with a highly unusual name: The Browser Company. Far from the usual nouns (sometimes with skipped or added letter), that’s how people name their businesses in late 1800s, not 2020s. It was truly unique and burnable, because anybody trying to follow them would instantly become a copycat (Jordi Hays had a great post on this). It’s quirky, it’s fun and it hints at their main goal: building a new redefined browser, the first company launched to do so since Mozilla.

Each viral company clearly communicates its purpose and values, which makes their brands memorable and engendered loyalty among users, investors, and even talent. And the brand is one thing competitors can’t copy. This is the second pillar: great storytelling. Stripe isn’t a card acquiring platform, they’re growing the GDP of the Internet. Lovable’s tagline is “building the last piece of software”. You don’t necessarily have to have grand views. Linear calls itself a purpose-built tool for planning and building products. Straight and to the point, just like everything is about them. Every startup is unique.

And most of these companies need the enemy. It can be a single product, a company, or an entire faceless industry. For Linear it was Jira. For Stripe it was the the very challenge of accepting payments on the web. With Arc, Josh Miller gave talks on why you need a better browser than Chrome.

Along with brands like Stripe and Linear, Arc proves that viral startups are obsessed with product taste, a rigorous attention to form and function that goes far beyond what’s common in traditional enterprise software. Early adopters and powers users definitely care for delightful software, which is why they share praises and screenshots publicly, which fuels the growth further.

If your product solves the pain but hard to use, you will not get public praises, and then you won’t be able to launch a flywheel of generating new users and referrals. Superhuman famously built a major part of their brand around the need to get all interactions down to 100ms or less, which is supposedly the border where people notice delays. That extreme love to the product among its small but powerful subset of users picked everyone’s curiosity, acquired tastemakers, and later converted into the demand from truly profitable market of enterprise customers.

Superhuman also came up with the PMF metric that perfectly defines delightful and useful products: asking how disappointed would the users be if they could no longer use the app (good if over 40% of them are very disappointed). Although, it can be slightly misleading in case of unique monopolistic products you can’t replace (don’t like using it but don’t want it gone anyway). So now we have the third pillar: delightful product.

Don’t all companies care about their user experience and try to bring delight? Everyone wishes they’d do this, few have the capacity. The hardest thing, of course, is that you can’t copy this by hiring an agency. It’s something the company must have internally and the direction should come from the founders themselves. Any friction ruins the experience. You might even have a good product, but if it’s considered “heavy” or “laggy” you will never get the laurels.

The best viral companies hire and empower people with a natural sense of what makes software elegant, intuitive, and delightful. It’s hard to find such people and even harder to hire them in the first place, because they have many opportunities open. Usually they were involved in building similarly great products before, even if those didn’t work out. Cursor, a $30Bn AI coding IDE, specifically hires great people over great candidates:

At Cursor, the recruiting process looks like this: post the name of someone really, really good in the #hiring-ideas channel in Slack, swarm that person with attention, conduct team interviews (wide range of “process” here), and if the desire is mutual, they start on Monday.

A product that end-users can easily touch helps a lot. Indeed, Adyen has grown its processing volume and revenue in lock step with Stripe, but focuses on larger merchants, meanwhile any developers can sign up for Stripe and connect it to their personal blog if they wished to do so.

The real truth is none of these metrics can be faked. Either you have raving users or you don’t. You technically could care a lot about the story you tell, but your ability to execute this is what’ll be the blocker. The same goes for the delightful product: a desire to make one doesn’t necessarily translate in your ability.

What else?

There’s a fourth quality at the intersection of the first three: built-in shareability. Viral startups play status games and embed virality loops directly into their product in various ways. Sometimes it’s very obvious, like with Superhuman, which had two specific things. First, it literally added “Sent via Superhuman” to people’s signature by default. Anyone could do this, but what matters is who were the people sending emails with such a signature: founders of sought-after startups, powerful executives, popular venture capitalists. Second, it allowed all of their users to refer their contacts to Superhuman directly, one by one, which fueled both FOMO and their feeling of importance.

Sometimes it’s just the opposite. The first AI notetakers, like Otter and Fireflies, used the same model of sending a bot to join your calls. Instant virality, since they join all your calls and other see them. Granola differentiated itself specifically by not doing this. At the same time, many of its users (ideally all of them but that wouldn’t be truthful) disclosed they were recording the calls to their contacts, which made those curious and interested. Alex Konrad covered Granola and said:

Granola is the first AI note-taking app that I’ve actually enjoyed using – and didn’t feel like a gimmick or more work. It’s still small, but gaining super fans in VC and founder circles.

Shareability might come in a myriad ways. A YouTuber producing endless content about their Notion dashboard is sharing and promoting the product itself. But the most straightforward way is when you can either share some content with others via a link or ideally collaborate with them — Figma built their entire product on top of this idea.

So, here are the must-have pillars for viral products:

Raving users
Great storytelling
Delightful product
Built-in shareability

These seem to be necessary, but aren’t sufficient. Companies that are already hugely successful have their advantages baked-in. But the real question is at what point do they receive it? How do you actually start from scratch if you want to build a viral startup?

I don’t think anybody can have a clear answer of this. Most of these companies weren’t too obvious shortly after inception as well. Lovable now has $200M in ARR and looking to raise funding at a $6Bn valuation, but this very idea is the team’s third pivot. And Stripe famously had only 50 users two years after launch.

Patrick Collison: “If you’re working on a startup that’s a bad idea, it’s going to feel like slow-going. But if you’re working on a startup that’s a good idea, it may feel like slow-going too”.

And the end of the day, it was a lot of work done by very smart people willing to go through this mud of early stage.

All viral companies are united by the fifth quality: they are unreasonable about something. One or a few very particular things at least. You need cornerstones, the ideas you uniquely believe in and ready to commit fully. Cornerstones help to define everything you do and bring unique taste to it. They bring believers and fans, candidates and users. This is the magic that people see in products and the reason why they love them. Because of this, viral companies tend to go the extra mile and do things that surprise people.

Stripe realized that their actual users were developers and committed to being a developer-first company. All other payment processing services focused on the business; when a deal was made, developers would be required to integrate regardless. But Stripe decided they’d have the best API and the best docs, so that developers on the ground might either make a decision themselves or support Stripe in front of the stakeholders.

This is also why Stripe launched Stripe Atlas, their company registration tool, and Stripe Press, their own publishing. Both support their mission of growing the GDP for the Internet (and generate more clients for them along the way), and are generally quite unusual for any tech company and specifically for their area of work. But they did it anyway, because this is what makes them a magnet and a mission worth associating with.

A good method for understanding whether something is unreasonable is trying to imagine this happening at a regular corporation. Most corporations are like cakes: they are built from countless levels of hierarchy and approvals. Can you spend an unreasonable amount of time on polishing a feature to perfection? Or can you approve and put up a billboard in 8 hours like Ramp can? As a startup, you main advantage is speed and the quality of execution, along with the hope that you won’t lose it as you grow. The most successful viral companies are the ones that preserve it for many years.

Superhuman’s first deck looked very different from all other startup presentations. It was basically a collection of screenshots, a vision board of their taste, with no charts or forecasts at all, all supported by their core belief that all interactions should take 100ms or less. The founders repeated this talking point so often others started believing it too — and associated their product with taste and speed.

Lovable itself decided to focus on transparency and speed of iteration for their community engagement. They shared casual behind-the-scenes updates and stories of both successes and failures. They turned their day-to-day work into social media content. When a user reported a bug, they’d rush to fix it and report about this publicly and mention the time that it took for everyone to see. Lovable wanted its users to cheer the process.

An early stage startup often doesn’t have too many things to be proud of. But you can always take pride in being unreasonable about something. About your customers, your developers, your design, your speed or something else. Figure out that cornerstone and turn it into artifacts you can share through product features, emails and social media posts. Show how much you’re doing for your customers or how much money you’re saving for them. Show your best designs and explain what unreasonable process you took to end up with them. Show how fast you’re willing to move.

Be unreasonable about something and people will remember this.

Retro Tech Became the New Luxury

Thu, 13 Nov 2025 23:20:00 +0000

YouTube is full of videos with people trying to replace their iPhone with half a dozen devices. They carry an iPod for music, a dedicated camera to take photos, a Kindle for reading, a notebook for thoughts, sometimes even a flip phone for calls.

Companies are selling minimal phones, like the $599 Light Phone that can do only a few things and won’t let you get you an Uber. The market for old point-and-shoot film cameras is so hot that Pentax and Lomo are now making brand-new film cameras and some people are buying up old digital ones. Kylie Jenner and Zendaya made Contax T2 so popular it costs $1000 now.

A used iPod Classic in good condition can sell for $200 on eBay and there are now dedicated companies that repair and retrofit them on-demand, like Elite Obsolete and Retrospekt. Batteries, screens, buttons, cases: all third-party parts are easy to find. The only limiting factor is the motherboards, so they depend on how many iPods Apple originally made.

There’s an old line: “There are only two ways to make money — bundling and unbundling.” The iPhone was the ultimate bundle, merging half-a-dozen devices into a single one and became so ubiquitous there’s now a counter-culture reaction to it. This return to separate, single-purpose devices is part retro nostalgia, part rebellion against being constantly online.

There’s an entire industry emerging around digital minimalism, complete with manufacturers, celebrities, influencers and cults. Some of them hardcore cultists and you instantly believe them when you see them. Some are geeks who like using a particular gadget.

Why do people seem to want it so much? There’s an article describing and naming a particular phenomenon anemoia:

Anemoia is the feeling of nostalgia for a time or place that one has never experienced

Most Gen Z never owned a digital point-and-shoot. Younger millennials didn’t grow up with vinyl. They’re experiencing nostalgia for a world they never inhabited. It makes them perfect customers. Because anemoia doesn’t just make you want the experience, it makes you want the object that represents it. The influencers aren’t lying when they say their iPod reignited something. It probably did. But they’re selling the object as the solution to a feeling that actually predates the object.

But I did own a point-and-shoot and only had a primitive cellphone for most of my childhood. What I didn’t have is an iPod, so I bought one myself a month ago to try it out.

And you know? It’s an awesome device and very fun to play with. The interface is fast enough, but also beautifully skeuomorphic and amazingly logical. It has all the features and functions you need to keeping and listening to your music library. Its design looks like it belongs in a digital museum (in a good way).

I already had a small MP3 library and bought more through iTunes (still the easiest way, and the files are DRM-free). People often recommend Bandcamp, but nobody except indie artists use it.

I get the sentiment. iPod is almost like vinyl, which is another trend of the last decade (sales started growing again after 2020). You have a limited library, so you end up listening more carefully and rediscovering the albums you already had. And you stop skipping every 15 seconds, because it’s not too convenient. Unlike vinyl though, the iPod gives you much better sound quality and much cheaper to run. But here’s the thing: it still spends most of its time on the shelf.

The truth is, Spotify can offer something similar if you just use it more deliberately. You can get all these benefits without the performance, the price, or the dependency on influencer trends.

Create one playlist for your absolute favorites and don’t treat “Liked Songs” as a library; it’s just a collection bin. Then, create playlists for favorite artists and specific moods, like training or reading. There are buttons at the top allowing you to filter between Playlists, Artists and Albums in your Library persistently. Just use this.

And if you want ownership, you can actually keep a library of MP3 files and listen to them on your phone. You don’t need an iPod for this! You can either upload your files to Spotify or Apple Music, or listen locally through apps like Evermusic, Flacbox, or Plexamp.

The entire market of digital minimalism exists because people feel powerless. But selling someone control is just repackaging the same problem, whether it’s a $599 Light Phone or an influencer’s perfectly curated iPod moment. You’re still buying your way to peace through things.

How Big is TBPN?

Thu, 30 Oct 2025 12:35:00 +0000

Zuckerbeg. Nadella. Benioff. An exclusive announcement from Microsoft.

TBPN has been blasting from all corners of tech Twitter lately. It’s a technology daily show hosted by John Coogan and Jordi Hays, streaming live every weekday for about three hours. Sometimes I wonder who has that much free time to listen to it.

Both hosts take it immensely seriously and first got the accolades of tech Twitter and now gathering Magnificient 7 CEOs one by one. They also secured top-tier sponsors like Figma, Brex, Linear, and Public. It all started just over a year ago and has grown tremendously since.

I have no doubt their imagery and atmosphere, lifted from sports channels but uniquely applied to the tech space, plays its role, but you probably need quite an audience to warrant the time of a Big Tech CEO, right? So, how big they actually are?

What I find fascinating is that despite being video-first, TBPN is fairly small on YouTube, with only 35k followers; most of their videos gather a couple thousand views, with one exception at 360k.

Twitter is surely a big driver for TBPN, especially because it can handle both long-form videos with chapters and short clips. In fact, I remember Coogan as one of the first people I noticed were using those a lot to post videos on Twitter some time ago (on his personal handle). TBPN has over 120k followers on Twitter now, with average views around 5000-50000 per tweet that contains a video.

So, if it’s not YouTube, where are their views, if anywhere? I imagine there are two possible avenues. One is it’s mostly in clips, short-form pieces of their longer videos. Also, TBPN is released as a podcast, so many of their listeneres might be there, on Spotify and Apple Podcasts, along with all third-party clients like Overcast, Pocket Casts and Castro, which rely on Apple Podcasts’s library.

While they do post clips on X, I doubt other social media platform carry a lot of weight with clips. TBPN has 13k followers on Instagram and only 1772 on TikTok. Many of their own clips on the latter get thousands of likes but none seem to breakthrough beyond this. Searches for their clips from third-party accounts don’t show a drastically different picture. So it’s not the clips.

Unfortunately, there’s simply no way to get reliable listeners stats for podcasts — that’s how this ecosystem works. Spotify doesn’t share its figures, Apple can’t even collect it beyond their own app, so only the hosts know their downloads figures (but not neccesarily the subscribers). But we can get estimates and their position in the top charts. Listen Notes, a podcast monitoring tool, estimates that TBPN is in the top 1.5% most popular shows out of 3,659,109 podcasts globally. Here are some well-known benchmarks:

Joe Rogan: top 0.01%
Dwarkesh: top 0.5%
All-in Podcast: top 0.05%
Acquired: top 0.05%

To give you a reference, another podcast at around 1.5% is Sam Harris.

And if we check the Tech category in Top Shows on Apple Podcasts in the US, TBPN is #32rd. Which is quite impressive for a show that only exists for a year, even though the benchmarks podcasts are in top 10. Especially if we remember the quality of guests TBPN is able to attract.

This is why most podcasters I talked to recommend ignoring YouTube unless you get really big. It’s almost better not to have it at all than disclosing your numbers. A podcast with 20,000 listeners is impressive, while a YouTube channel with 20,000 views and 5000 subsribers looks subpar, simply because there are so many big creators there, even though their real audience might be the same. But TBPN was started as a TV program, so they didn’t have this luxury. Instead, they committed and produced content that was barely seen for a long time.

And the reasons they could is because it’s like both hosts were created for this. John Coogan co-founded Soylent, where he was the CTO, and then Lucy, where he served as CMO. He also had a pretty successful YouTube channel of his own with 459k subscribers and popular videos on multiple tech companies and startups, as well as tech, like this one about the origins of DARPA. His co-host Jordi Hays created built and led a performance-focused YouTube ad network and invested in multiple startups. Together, they seem to have a perfect combination of knowledge and network to pull this off (which is what they’re doing at the moment).

Now the format. The New York Times described TBPN as “SportsCenter meets LinkedIn”. One related trait is the enthusiams both John and Jordi show to the technology industry as a whole, similar to how sports pundits enjoy the very sport they’re covering.

F1 pundits critisize the drivers and the FIA. But they love the sport. The same goes for football, basketball, etc. This doesn’t mean it’s all rainbows and stuff. In addition to bashing the gong for major funding rounds and product launches, TBPN doesn’t shy away from discussing the impact of AI generated music on artists or overhiring in tech. But you can feel they love this industry.

And this is why tech CEOs are probably happy to go on the show, even if it might not be truly big yet. It’s big enough in a certain tech afficioado circles who help drive the conversations.

The lack of such love for the industry has become one of the key critisisms of the media and reporting ecosystem in the last 5-6 years or so. As Seth Godin used to say: “The media wants overnight successes (so they have someone to tear down)". Matt Yglesias publicly commented that The New York Times made a “weird editorial decision” regarding its tech industry coverage, suggesting there was a top-down directive for reporters to approach tech stories with a more critical, investigative tone.

The industry felt this and many people adopted the “going direct” strategy, while others poured millions into building out their own media operations to fix this. Like Future, the outlet launched (and closed) by Andreessen Horowitz. But all of them were trying to come at this with a wrong atittude. Either by hiring people editorial experience and giving them unviable directions or trying to expand a usual corporate podcast into something big.

Instead, the hosts of TBPN started with a good and unique media product.

The Broken Promises of Substack

Thu, 21 Aug 2025 16:35:00 +0000

Earlier in July, Substack announced a $100M funding round, bringing its valuation to $1.1 billion.

Substack has built an entirely new media ecosystem by enabling reporters and influencers to monetize their audience longing for long-term writing. People share links and discuss pieces from Eric Newcomer’s Newcomer, Bari Weiss’s The Free Press, Richard Rushfield’s The Ankler, Gergely Orosz’s The Pragmatic Engineer, Lenny Rachitsky’s Lenny’s Newsletter, Matthew Yglesias’s Slow Boring and thousands of other blogs. Many of these projects have grown into mini-media companies with staff, essentially rebuilding the publications their founders once left.

Substack now has more than 50 creators who are making millions of dollars per year on the platforms. In total, they boast 500,000 creators, more than 5 million paid subscribers, and have about $45 million in annual recurring revenue. Their domain got more traffic than than The Wall Street Journal and CBS News in June 2025.

But building this ecosystem at scale has forced Substack to abandon some of its original promises.

Portability of Audience

Substack’s original pitch was independence. Writers owned their mailing lists, controlled their billing via Stripe, and could leave at any time.

Chris Best, Substack’s co-founder, said on the interview with The Verge in Dec 2020:

Another way we do that is we mean it when we say we’re helping writers go independent, and they own their content and they own their contact point with their audience, which means that you can leave. If you’re a writer and you build up your following and your subscriber base on Substack, you can take it away. <…> Start a blog, an email newsletter, have people subscribe directly to somebody you trust. You own your content, you own your audience.

This changed in March 2022, when Substack introduced its app that consolidates subscriptions in one place rather than dispersing them separately via email. Unlike the usual email subscribers, the followers who signed up through the app can’t be exported, creating a powerful lock-in. For many writers, it could be up to a third of their audience. Ben Thompson, whose Stratechery newsletter directly inspired Substack, wrote then that Substack has gone from being a “Faceless Publisher” behind the scenes to trying to put “the Substack brand front-and-center,” building up its app as a destination on the backs of writers.

Just a few days ago, Substack also launched in-app payments for all writers on iOS. It is true, that they can only launch it for everyone or noone, but this creates yet another lock-in. Unlike with Stripe, you can’t export your paid iOS subscribers.

No Algorithms

Substack positioned itself as the anti-platform. No algorithmic feeds, just direct connections between readers and writers.

On the same interview, Chris said:

If you don’t like the stuff that you’re seeing there, you have this really good remedy, which is [to] hit the unsubscribe button. A lot of the worst problems that content moderation addresses on other platforms is the spread of content that is bad, because basically your algorithmic feed is serving as an editor, whether you think of it that way or not.

Back in 2021, Chris wrote on his own substack:

For writers, this means being able to control their relationships with their audience instead of being mediated by fickle corporations whose algorithms decide what gets the most attention.

But in 2023, Substack launched Substack Notes, it Twitter replacement, which uses an engagement-driven algorithm that prioritizes likes, replies, and reposts to boost visibility. More recently, it expanded even more in the social network aspect, launching vertical videos and a $20M “creator accelerator fund” to attract TikTokers.

No Ads

Originally, Substack positioned heavily against ads. Here’s a comment made by one of their employees in 2022:

We aren’t in the attention economy here. No ads. We only make money when writers make money and decide to turn on paid subscriptions.

Then, in 2025 Hamish McKenzie, another co-founder of Substack, told DigiDay:

“For us, advertising is an interesting business. Maybe some way off in the distant future.”

Content Moderation

Substack’s founders long argued that they solved moderation by design. But as soon as Substack built recommendations, it inherited the same problems as every social network.

That interview Chris gave to Nilay Patel is an extremel useful benchmark to track changes in their thinking and approach:

A lot of the worst problems that content moderation addresses on other platforms is the spread of content that is bad, because basically your algorithmic feed is serving as an editor, whether you think of it that way or not. And I’m getting bad crap in my Facebook feed because some uncle of mine liked it or because it’s getting engagement. That problem doesn’t exist on Substack in the same way. I think we can and should do discovery, and it’s just important for us to do it in a way that takes advantage of the model that we have, and of the fundamental promise of Substack. <…> Because the whole value of Substack is the direct relationship between readers and writers. If we did discovery in some cheap way that might be like, what would work best on YouTube, it would be easy for us to blindly violate that and kill the thing that makes Substack good.

In July 2025, Substack sent a push alert to some users promoting a Nazi blog called “NatSocToday”, which featured a swastika logo and content pushing Holocaust denial. This is the direct result of Substack engaging with recommendations in order to help authors grow their newsletters. If you write on Substack, you might get more subscribers, but your brand will be used to promote other authors as well, even if you might disagree with them.

Substack has stumbled into Nazi-related controversy before (not a phrase you’d want to hear about yourself!). In 2024, criticism of its content-moderation policies prompted publications like Platformer to leave. In my view, that case was a bit overblown and I generally agree with the Pirate Wires’s explainer. Ultimately, Substack had to remove a number of authors involved in this.

What is the appropriate layer of the Internet stack for moderation and censorship is an important question, the one Substack’s founders like to discuss a lot, but it’s a different question here. The only reason for this latest accident is Substack giving recommendations in the first place.

In practice, Substack now functions much like any other social network. But unlike Facebook or Twitter, simply hosting an author on Substack carries a kind of implicit endorsement. That aura makes its laissez-faire moderation stance harder to defend: the company insists on First Amendment absolutism, yet its recommendation systems inevitably entangle it in the politics of promotion.

My goal here is not to attack Substack but highlight the compromises they made while implementing their original vision. It’s the endless dilemma for social products and as we see subscriptions alone don’t ultimately solve it.

Carcinization is a process where various crustaceans repeatedly evolve into crabs. There must be a similar “law” that all content platforms evolve into algorithm-driven social networks with powerful recommendation engines — or perish.

Earlier this month, Ghost, an open-source publishing platform, reported that publishers earned over $100M in subscription through their platform. Impressive, but it took them 12 years to get here and their entire model for paid communities was introduced in 2019 directly inspired by the growth of Substack.

So why do authors still go on Substack?

Because they see the value in this. On August 21, 2025, Mika Solana announced Pirate Wires are integrating back with Substack in an email:

I do recommend downloading the Substack app if you don’t already have it. The app will grant you a lot of new Pirate Wires content. All of our writers will be pretty active on Notes (social posts on Substack) moving forward, and I’ll be going live myself (a kind of video chat, sometimes solo, sometimes with my team, sometimes with other popular writers on Substack) on and off all week. I’ll also be posting my little heart out.

I’ve learned a lot about media these past few years. I grew my subscriber base with Substack in the early days as a single writer, and I grew my subscriber base without Substack, on my own, with a full team. The latter was more difficult, as I knew it would be, but I traded the Substack audience for complete creative control of my company.

It was a no brainer decision. Substack tools are incredible. Their CMS alone would be a compelling reason to build with them, but their entire publishing backend is beautiful, and their growth network is unlike anything else in the business. I’ve watched people leave Substack in a huff and immediately die (couldn’t be me). I’ve watched people grow on Substack and somehow not understand the degree to which the Substack growth engine was the reason. And now that I can reenter on precisely my own terms? It’s a new era for the Pirate Nation, tbh.

Alternatives

Substack’s greatest risk has always been writers leaving. A 10% fee (plus Stripe’s usual fee) in exchange for someone handling all the tech, subscription and mailing blasts sounds great at first, but as authors grow big and get to a couple of hundred thousands a year in income, they start looking more and more at this annoying little line in their P/L.

With Pirate Wires, Solana treated Substack as the as a launchpad before spinning out. Some other writers left citing the company’s moderation policy, like Casey Newton with Platformer, or because they couldn’t turn it into a reliable income stream. Every, originally known as The Everything Bundle, evolved into a full-fledged media company that needed its own technology.

For most authors, Substack is still the best bet.

Now, setting up your own blog is easy; I covered the options before. You could even run a free static website, but the two main hurdles are newsletters and paid subscription. As the world largely abandoned RSS (the “pull” model), email newsletters have become the most reliable way to connect to your readers directly, but it’s the “push” model that requires far more work. Sending and delivering mass email reliably is difficult and you will probably have to pay someone else to do so. Subscriptions push the complexity to yet another level.

Substack is still a very good option because you, as an author, don’t have to pay anything (except for your own domain name). What if you want more control though? At the end of it, all Substacks look like Substack, which has been another point of critisism.

Substack’s brand has become an asset for writers. They practically turned substack in a new noun (which heavily annoys many people). And it helps conversions, because readers already know what Substack it, how to subscribe and why they might want to pay for it. The funnel is also highly optimized for moving visitors to free subscribers and then paid subscribers flow. The trade-off is uniformity: every substack looks like every other substack, which can make individual blogs feel less distinct.

Substack has two very direct alternatives in the form of Beehiiv and Ghost. Beehiiv was built by the technical people from Morning Brew and provides vast customization capabilities with custom websites, custom newsletters, analytics, and API access. In return, you have to pay subscription for access to advanced features, but it provides a lot even at the free tier so you can pay $0 until you hit 2,500 subscribers.

The beauty of Ghost is that it’s a free open-source software that you can deploy yourself. But there’s the catch with actually sending newsletters, which is still challenging. If you care about it, the best option is to go for their platform offering, which immediately starts at $18 a month. But the starter tier doesn’t even support custom themes or over 1000 subscribers, so it’s more like $35 for the “Publisher” tier. Good for publishers, a bit pricey for regular bloggers.

Another less known is Buttondown, which I’m using to send out fresh posts from my blog. It’s a bit more geeky and basic, but has some nice features, including paid subscriptions, and until you get to 1000 subscribers it costs $9 a month.

If you are an aspiring writer aiming to build a large audience and sell content online, all of these might work, but Substack might be the easiest option to start because of its zero price and awareness. Still, even a couple of paid subscribers will nullify the costs of any other platform for you. That’s why Substack remains sticky: the entry cost is zero, the growth engine is real, the aura of legitimacy is powerful. But they will lock you in. If you build a large thriving newsletter on Substack, it’s a great thing in itself, but it will be quite hard for you leave. Both Beehiiv and Ghost are much more independent alternatives.

That doesn’t necessarily mean you will be successful. Most blogs on Substack earn nothing. Most blogs on the entire web earn nothing. But if you aren’t sure or don’t want to commit for all eternity, Substack and Beehiiv are probably the easiest place to start for non-technical people. Because email is how most people consume text content these days.

Substack promised independence, but has evolved into another platform playing the same game as everyone else. The crab always becomes a crab.

How Much Spotify Pays Artists

Sat, 16 Aug 2025 12:20:00 +0000

Few things get people so worked up as Spotify’s payouts to artists. Every time this topic comes up on Twitter or Threads, you can be sure there will be hundreds of angry replies from people you can’t argue with.

So I wrote this post.

On Spotify, a single stream generates on average between $0.003 and $0.005 in royalties that get paid to the rights holders. Is that small?

It’s important to note these per-stream figures are not fixed prices set in a contract, but rather averages resulting from a revenue sharing model. Many streaming platforms, including Spotify, operate on a pro-rata pool system: roughly 70% of all streaming revenue gets paid out to music rights holders, divided proportionally by each rights holder’s share of total streams. This topline figure heavily depends on the earnings, which are affected by differen pricing tiers, bundles, regional pricing, etc.

For most of its history, Spotify has been paying out just around 74% of all of its revenue as royalties to right holders. As you can see from the chart, 2025 is the first year to see a sudden drop just to 69%. This didn’t happen because Spotify altered its terms with labels (they wish they could do so). It’s because the platform has been investing in other types of content, specifically podcasts and audiobooks, where they operate on different terms. The Q1 2024 has been widely recognized as the turning point for consistent profitability for the company. Before this, Spotify did show profitable quarters here and there, but never in a sustainable way.

For comparison, Apple Music’s stated headline rate (what they’re paying to right holders) is 52%. As you can see, Apple actually keeps more of their music revenue to itself compared to Spotify (who actually do some shadier things we’ll get in later). In 2021, they publicly stated Apple paid $0.01 per stream to rights holders, although independent analysts in 2023 put Apple’s average around $0.008 per stream, roughly double Spotify’s rate. How does this happen?

It’s because the actual revenue per user is different. Streaming services have different plans for individuals and families. They enact regional pricing across different countries. For instance, a month of Spotify costs just ₹139/month in India (~$1.6). This isn’t unique to Spotify, but what is unique though is the ad-supported plan, which enables the company to have 696 million users, who, on average monetize worse than the users of Apple Music or Tidal. Finally, there are discounts. Recently, Apple Music offered me three months at the price of one in an attempt to reengage me as a user. Apple will not pay out of its pocket fo any streams I do during this.

The price per stream for Spotify has actually declined from earlier years as well. For example, an analysis of an indie label’s data showed Spotify’s effective per-stream rate dropped from ~$0.0052 in 2014 to ~$0.00397 by 2017. By 2019 it had stabilized around $0.0033–$0.0035 per play, and it remains in the ~$0.003–0.005 range as of the early 2020s.

Other Streaming Services

VIRPP has gathered the data for other streaming services. Tidal offers one of the highest rates (around $0.012–$0.013 per stream as of 2023), while Amazon Music is around $0.004 and YouTube Music is very low – roughly $0.002 or less per stream. In fact, YouTube (including Content ID streams) accounted for over 50% of music streams in 2019 but only about 6% of streaming revenue – an enormous “value gap.” This implies YouTube’s effective per-stream payout is just a tiny fraction of a cent (on the order of $0.0002–$0.0007 in many cases).

In practical terms, 1 million streams on Spotify gives you around $3,000–$5,000 in royalties, whereas 1 million streams on Apple Music might pay around $8,000–$10,000. What is important is that because of Spotify’s larger audience, artists are more likely to have a larger number of streams there. And also, the royalties are paid out to right holders, which can be the label, and then it comes down to the relationship that the artist has with the label.

What is very important here is to understand that all parties, including streaming services, labels, and artists have the same incentive. The more money Spotify makes on music, the more it has to pay back, while spending the remainder on its operations. Therefore, Spotify is incentivized to maximize its total revenue. This is why they have regional pricing and the ad-enabled tier — because the company believes they’re making more money this way.

Can’t they just charge more? Streaming services constantly check the elasticity of the demand, but any time Spotify raises its subscription by $1 you get a myriad of news stories and a public outcry. Revealed preferences: people “want” artists to make more right until it’s out of their pockets. And as we saw earlier, Spotify is barely profitable and can’t pay out more.

Streaming Is Good

Streaming has massively grown the overall recorded music market by solving the piracy problem. Labels have won the most. Their previous business model was failing, but Spotify saved them by providing and establishing an alternative. As the result, in 2024 the global recorded music revenue finally surpassed the peak of the CD era in 1999, reaching $29.6 billion globally. And streaming contributes 84% of recorded music revenue in the US and about 67% globally. Since its inception, Spotify paid out over $60Bn to right holders, including the artists.

So how much do bands and artists actually get from Spotify and other streaming services? Spotify’s own data shows that in 2024, 110,000 artists earned more than $5,000 in a year, about 71,000 artists globally earned over $10,000 USD. And of those, just around 1,500 artists generated over $1 million in Spotify payouts.

It’d be correct to say that only a very small percentage of artists see substantial revenue from Spotify and other streaming services. And if the artist is signed to a label, they might only see 15–20% of that amount as their take-home. So tours and merchandise are the primary revenue source for many musicians. But… it has always been this way.

Artists Before Streaming

Let’s compare how a band might have earned money solely from record sales in the pre-streaming era (say, the CD heyday of the 1990s).

If an artist was signed to a record label, their contract would specify a royalty rate for album sales. A typical album royalty was around 10%–20% of the album’s price. Mind you, that was the wholesale price, not retail. For a $10 album sold, the band would get $1. If that album sold 100,000 copies, which would make it a gold record in countries like the UK, the band would theoretically earn around $150,000. Most bands never sold anywhere near that kind of volume! And crucially, they wouldn’t see a dime until the label “recouped” all the money it spent on recording, marketing, and tour support. Many bands never recouped and thus made $0 from their record deals.

The Internet enables artists to grow independently and distribute their music without labels. You still need a distributor to upload your albums to most streaming providers, but many offer a flat fee and let you keep all the royalties. Now an artist with a laptop can produce, publish and promote their music without leaving their apartment. The barriers to releasing music have essentially disappeared. The final boss is the Internet’s greatest villain — attention.

According to Spotify, as of early 2025, there are nearly 12 million artists who have at least one song on Spotify. A huge chunk of these are likely hobbyists or very early in their careers. Spotify itself notes about 8 million of those artists have fewer than 10 tracks released and identifies roughly ~200–250k artists as the actively “professional” musicians. It’s hard to find a truly comparable number, but considering the challenges of recording music and putting it into stores, there surely were far fewer of them. The National Endowment for the Arts in the US reported 167,000 of musicians and composers in 1990.

There’s a headline I used to see too often, which derides YouTube for how few creators actually earn a full-time income from the platform. Titles like “Most YouTubers earn little to nothing” are written to invoke anger and despair. Well, YouTube reportedly has over 65 million YouTube creators, which potentially might even include myself, since I had an uploaded video (I deleted it). Well, I’d love my blog to have as many readers as The New York Times, but I’m definitely not entitled to this. And yet, between 300,000 and 500,000 people worldwide earn a full-time income from YouTube.

Now, there are more bands getting something than in the “good old days” when, if you didn’t land a record deal, you essentially got nothing from record sales. But it does create a long tail of artists who only get a little. One 2020 study found that the top 1% of artists get the lion’s share of streams (over 60% of all streams). This is because most people listen to Taylor Swifts and Drakes of the world and not your garage band.

Today, instead of a fan paying $10 for an album once, the fan’s listening contributes to many small payments for each stream. To match the revenue of one album sale, a song typically needs hundreds of streams. RIAA has a rule for counting album sales along streaming: it’s 1,500 streams for 1 album unit.

Let’s figure this thing out. 1,500 streams multiplied by $0.004 is $6. This is more than what you’d get from a CD, but unless you go directly, the label would get 50% to 80%, and you as an artist would get ~$1.2. One advantage of the streaming model is superfans continue generating revenue for artists even if they don’t release any albums, simply by playing music on repeat.

Labels Are Kings

Spotify can’t change the terms between labels and musicians. The reason why the music and film industries are so different is in music we have an oligopoly of major labels with an extremely useful back catalog. Now, a major company can start a new movie streaming platform filled exclusively with its own content, just like Apple TV+ did, and people might pay for it. But very few people want a music streaming app that doesn’t have Taylor Swift, or Drake, or Beatles, or somebody else they like.

And Spotify does some shadier things occasionally. Apple specifically attacked them for offering different payout terms to major and indie labels. And Spotify has a system that promotes artists’ music in exchange for cutting their royalty rates. No business is simple.

But Spotify likely provides more money overall to any artist than other streaming services purely based on its scale and you shouldn’t feel bad for using it.

***

Updated on December 31, 2025.

Many artists gripe about streaming payouts but few are ready to spill the beans on how much they actually make. I tried hard to find any metrics from the musicians’ side when I was writing this article, but couldn’t.

Now, a band called Los Campesinos shared their exact payouts (thank you!). Here are their figures from July 2024 to June 2025.

As you can see, the vast majority of both their streams and revenue come from Spotify. Alone, it brought them 3 times as much as Apple Music.

Unfortunately, of the major streaming platforms, Spotify pays significantly less per stream than anywhere else. If everyone who streamed All Hell on Spotify had done so using Tidal instead, we would have received an extra £31,847.38

This is why the per stream payout is a meaningless metric. The reason why Spotify dominates among their listeners is both their cultural prevalence and their ad-enabled plan. Just like Spotify can’t convert all of these users to a subscription, Tidal or another service wouldn’t be able to do also.

This November, Bloomberg published an entire story on how Apple Music is losing new listeners because it doesn’t have a free tier.

What Spotify does is maximizing its total revenue, from which the payouts to artists are made in the first place. So, the incentives are aligned.

The Limits of the Network State

Sun, 10 Aug 2025 14:25:50 +0000

The Network State is a concept popularized by the entrepreneur and investor Balaji Srinivasan. These network states begin as highly-aligned online communities and graduate to diplomatic recognition by existing countries, rather than seizing territory by force.

You can read in his own words:

A key concept is to go cloud first, land last — but not land never — by starting with an online community and then materializing it into the physical world.

I can understand the idea behind these words. The countries and states we have aren’t always great. They are often silly, short-sighted and outright dangerous. They hold their population hostage as politicians ride the memes to gain more power and execute things without understanding the unintented consequences. Unfortunately, I doubt we can go beyond them, even though I’d love to share the view.

Balaji’s ideas are very similar to the cyberpunk visions of the late 20th century, specifically the concepts explored in Neal Stephenson’s Snow Crash and Diamond Age. Both show post–nation-state governance models where corporations, cultural groups, or clans function as sovereign entities with their own laws, police, and territories. Phyles in Diamond Age are large, voluntary, and culturally aligned networks of people spread across the globe.

I can’t see a path for Balaji’s network states to become something more than a club or an online forum and gain sovereignty. Fukuyama’s End of History is an extremely misunderstood book: it doesn’t say the world won’t change; it implies that we will never see novel state concepts beyond liberal democracies (although countries can break down to authoritarianism again).

Sovereign states have three key monopolies they will never give up or share: identity, taxation, and the use of force. We will dive into each aspect separately, but let’s start with sovereignty itself.

There are 193 member states in the United Nations. The bar to be recognized as a sovereign state is exceptionally high. The advantages are numerous and there are few things worse than being born in an unrecognized state.

According to Balaji, a startup society with sufficient scale should eventually be able to negotiate for diplomatic recognition from at least one pre-existing government, and from there, gradually increase sovereignty, slowly becoming a true network state.

A few countries are recognized by only a small group of UN members: Abkhazia, South Ossetia, Northern Cyprus, and the Sahrawi Arab Democratic Republic. And these are some of the most chaotic, dangerous and poor societies in the world. Their recognition is primarily a political act, usually conducted by one or a few major countries in the region who have their own interest at heart (Russia for the first two, Turkey for Northern Cyprus, etc). Getting recognition by any state is extremely hard, but being accepted by a failed state or a poor Pacific Island nation wouldn’t help.

Truly sovereign states simply don’t want to lose the exclusivity of their status and unique capabilities.

Identity

Srinivasan envisions one’s citizenship for the network states will be established through cryptographic identifiers (for example, Ethereum Name Service or World Chain) rather than birth certificates or passports. But this is nothing more than relying on a library card.

If you can’t use the document to enter another state, it’s not an identity. Going through the border control requires a passport, and not just a piece of paper with a biometric chip, but a passport recognized by other countries.

The only non-governments that have the capability to issue internationally recognized passports are the Sovereign Military Order of Malta and the Vatican. As you can imagine, they do this very sparingly. The Vatican is a micro-state whose passports are issued to clergy, officials, and diplomats of the Holy See. The Maltese Order is actually recognized as a sovereign subject of international law for historical reasons. It has diplomatic relations with over 100 countries, making its passports valid for official travel in many places. However, these passports are not available to the general public and are strictly limited to official use. ⁠No other non-governmental organizations or entities have internationally recognized passports.

You can get things like Palau Digital Residency, which a blockchain-based digital identity program offered by the Republic of Palau, but you can enter zero countries with this. That’s why it’s just a “library card”.

Certain countries issue so-called alien passports to refugees. A Belarusian citizen who escaped persecution can obtain one if they live in the Netherlands. However, according to the law, they will still be Belarusian citizens and will not get any additional rights (e.g., visa-free travel to the US).

Governments have a monopoly on identity.

Taxation

Balaji believes that since the crypto-based economy is inherently borderless and can operate outside legacy banking systems, traditional tax authorities will have less visibility and control over network state commerce.

In Stephenson’s books, the invention of crypto currencies led to the collapse of centralized states, which could no longer collect taxes. We have crypto currency, and the most capable tax entities know what to do with it. The US treats crypto as property, taxing gains and enforcing strict reporting rules. IRS enforcement is increasing, with mandatory reporting and penalties for non-compliance: from 2025, brokers must report digital asset sales to the IRS.

Balaji even foresees that network states will effectively arbitrate between jurisdictions to minimize burdens and maximize freedom for their citizens. Since a network state’s people are distributed across many countries, the community can choose to establish physical hubs in places with favorable laws and low taxes. Moreover, the network state is supposed to be funded by the members, who might pool funds to buy land or build communal infrastructure.

Since the governments have sovereignty, they are not going to relieve the network state of taxes. Anything you want to collect or raise has to come on top of the government taxes. The only countries willing to accommodate such daring requests are likely failed states.

And this is where we meet the final issue.

Use of force

Max Weber, a German sociologist, was the first to define the state as the only human community that claims the monopoly of the legitimate use of physical force within a given territory.

In the early stages, network states are supposed to lean on existing states for physical security. Srinivasan is candid that a startup society won’t be handling violent crime or defense immediately:

If governance is limited to the digital realm… how does a startup society deal with physical criminals? The short answer is that for a long time, it doesn’t – it leaves that to the surrounding legacy society.

An early network state has no physical coercive power, including an army or police force. Which is why being co-located with a potentially poor failed state is quite dangerous. Either you will suffer from organized crime, or sooner or later, they will try to take over the network state. At the same time, capable countries, whether it’s the United States or Singapore, won’t tolerate shedding sovereignty.

Of course, you might want to go beyond existing states and their land simply. Create new land.

Land becomes elastic. As Will Rogers once said, “buy land, they ain’t making any more of it.” Or are they? Seasteaders and the artificial islands built in Dubai show that land supply is perhaps more elastic than we think. We also know you can build cruise ships. So it’s possible that we could start reopening the frontier physically as well, not just digitally.

Dubai isn’t a great example. Palm Jumeirah and The World Islands are famous artificial land projects that are only possible with infinite oil money and for astonishment. Many countries dredge sand to raise their shores and protect them against erosion. And some (like China) indeed create artificial islands with the sole purpose of using them to claim sovereignty over a larger sea territory.

Countries can claim up to 12 nautical miles of territorial sea and 200 nautical miles of Exclusive Economic Zone from their land, including islands. China has built and expanded artificial islands to assert its “nine-dash line” claim, which covers most of the South China Sea.

If you want to create an island nobody can claim, you will probably have to do this in the Pacific Ocean.

I Can't Stop Using Dia Browser

Sun, 03 Aug 2025 13:57:39 +0000

I was a heavy user of Arc and abandoned it only after it was “sunset”. The app still works, of course, but I used to experience a few glitches and understood they would never be solved anymore. Still, I get their decision — Arc just wasn’t mainstream enough.

I understand it even better after using Dia. It has become my primary browser and I’d be extremely dissapointed if it went away, which is a very good PMF score.

The early reviews weren’t great. Arc has become popular because of its unique and polished UI, and then the Browser Company threw it all away and started from scratch. I get the frustration. But a niche browser never had the chance, while Dia might.

Unlike Arc, Dia simply feels like a more polished Chrome. A much better Chrome. Frankly, that just shows how little Google cares about its products and their UX. A team of 120 employees has managed to build an alternative UI that is a joy to use. Dia doesn’t have many advanced features yet, such as Tab Groups or PWA support, but everything included already feels better. And Chrome’s tab groups have always been quite rough and inferior to Safari’s anyway. Plus, with the recent update, Dia again possesses my favorite Arc’s feature in vertical tabs.

What really separates Dia is the AI. The browser is built around it. And I can definitely see why the team went this way. Once you start using it, it’s really difficult to go back. And I imagine the overall familiarity makes it easy for most people to switch from Chrome.

For some reason, none of the major AI chatbots really interact with your browser. And if you send a link to ChatGPT to summarize the webpage, there’s a good chance it’d just hallucinate the content instead. With Dia, you can always use any particular set of open webpages as context for the LLM. And this is wonderful. You can summarize, ask questions, or get additional context from the LLM itself or ask it to search the web for you. Just click Cmd-E and you have Chat on the side.

You can instantly ask whether the thing you’re planning to buy is actually good and find promo codes to get it cheaper. And it works! Don’t trust the reviews on the website, just ask and the AI will give a summary from top-tier review websites and meticulous fans on Reddit. You can extract data from complex HTML layouts and get a simple list of bullet points if you need it.

You can do more complex things and turn this into a make-shift Google NotebookLM. I opened a dozen of articles and blog posts about the state of AI coding, then asked it to summarize them into a quick single industry report. The result was quite close to what I’d mention, yet it only took me a few seconds (the rest is actually knowing what context to provide).

Finally, I notice that I really enjoy using these models, whether it’s from OpenAI or Gemini, via the API and with Dia’s prompt instead of their first-party chatbots. After I “personalized” Dia by giving it several people whose writing I enjoy and mentioned I prefer outlines and bullet points, it gave me very well-written straightforward answers. It’s not hilariously sycophantic like ChatGPT that you can’t really wean from this behavior. And kindly offers “links” to ask about adjacent topics if you want to dig deeper.

When you type something in the omnibox, Dia chooses whether to send the request to Google or its AI implementation, mostly based on the length and the syntax of the request. And this is great, because some things are just easier to check with Google. You can always override and choose yourself.

The thing that surprised me the most is how often I end going to Dia to ask question in Chat. It makes sense, because I spent most of my days in the browser. But what I really like is how ephemeral these chats are. I realized I don’t like that ChatGPT keeps all of your requests in history unless you manually enable a Temporary chat each time. Most of the things I search aren’t that interesting or important, but ChatGPT is hoarding them all regardless. Sometimes I research a medical condition, but sometimes I just need a quick lentils stew recipe or a fitting English metaphor.

Dia keeps a few recent chats in history, but ultimately they’re ephemeral and very low-friction things, and that’s why I end up using its AI so much. I also can’t imagine a browser without it.

Of course, Dia isn’t the only option. Google Chrome added Gemini, but only in the US, and according to the reviews it’s pretty basic.

There’s also Comet from Perplexity, which has a more agentic AI that can actually interact with the web pages, go through links and perform actions. But it’s not entirely reliable yet and there’s a chance people will get discouraged if they hit a brick wall too often and stop trusting it altogether. Comet also pushes all searches to Perplexity, which is good for some things, but not all of them. No surprise Google is scared and pushes AI so hard.

Finally, OpenAI’s agent can also open links and interact with webpages now. There are persistent rumors about them building their own browser as well.

We are in the new era of browser competition, and most importantly, it’s finally not about coming up with yet-another-engine — Chrome has officially won. Instead, all these companies are tuning the browser’s user interface and experience, the area that stayed under the radar and stagnated in the past decade. And I couldn’t be more excited about this.

Highlights from Apple in China

Thu, 31 Jul 2025 19:45:07 +0000

I have finished Apple in China by Patrick McGee, a fantastic book dissecting the history of manufacturing and breaking up some common myths around it.

There’s an uncanny quality about adversarial books and media stories like this. Patrick has talked to dozens of people who worked at Apple or their partners. But there’s always a bias in terms of who is willing to talk — mostly the people who didn’t leave the company on good terms. Jeff Williams, Apple’s former COO, isn’t joining that meeting invite. But the people who had issues with the management, disagreed with their “ignorant” strategies or believed they were wronged would. They might also use this as an opportunity to remake their story, so we have to trust the writer to build a complete picture from multiple sources. Throughout the book I was getting just a touch of that feeling, but I believe Patrick McGee did a very good job overall.

We know our devices are made in China. We saw the label on the back and maybe have heard about companies like Foxconn and workers operating these human “conveyor belts”. And while, admittedly, it is a part of the picture, Patrick argues that we’re missing the forest for the trees.

Apple hasn’t moved to China because the workforce there is cheap. In fact, it was one of the last companies to offshore it manufacturing. Following the return of Steve Jobs, Apple continued using US-based manufacturers but struggled so much with the iMac it had to look outward. First to Japan and Taiwan, and later to China, as local companies propped up by local governments and lax regulations enabled to stimulate growth fiercely competed for their business. And as a result, China built an ecosystem of manufacturers and suppliers where you can do and build practically anything you can dream of.

As Patrick says:

The prevailing Western narrative about Apple in China is remarkably narrow. The go-to story of the past two decades has been about the tedium of assembling Apple products—a tale of low wages, underage employees, sixteen-hour workdays, suicides at Foxconn, and accusations of forced Uighur labor. This narrative isn’t wrong, but it misses the biggest piece of the puzzle: It’s not merely that Apple has exploited Chinese workers, it’s that Beijing has allowed Apple to exploit its workers, so that China can in turn exploit Apple.

You see, China was trying to catch up in the technology and manufacturing and after their acceptance into the WTO looked for semi-legal ways to import the technological know-how. And according to the book, Apple became the perfect partner.

Apple had essentially cracked the code on how to manufacture the world’s best products without doing any of the manufacturing itself. It wasn’t really “outsourcing” in the normal sense—that would imply it was sending blueprints to companies capable of taking the orders and executing. Instead, Apple was routinely sending its top engineers, designers, procurement specialists, and lawyers from the United States into hundreds of factories across the country, where they’d import machinery, train armies of workers, coordinate the delivery of intermediate goods, and scrutinize suppliers to ensure compliance.

And the most ambitious claim this book makes is that Apple uplifted the entire China’s manufacturing by bringing its talented engineers, educating manufacturers and pushing suppliers to their limit to raise quality.

China wouldn’t be China today without Apple. Its investments in the country have been spectacular, rivaling nation-building efforts in cost, man-hours, and impact. Apple itself estimates that since 2008 it has trained at least 28 million workers—more people than the entire labor force of California. China brilliantly played its long-term interests against Apple’s short-term needs. In 1999, none of Apple’s products were made in mainland China; by 2009, virtually all were.

In fact, the book devotes a lot of space telling the story of Apple’s struggles in China because their approach was too unusual. China had come up with a simple system of requiring foreign companies to set up joint ventures with the idea that local partners would master the necessary skills to start building the tech themselves.

The party’s whole reform program was meant to lure in capital and Western businesses as a way of learning, so China could reverse-engineer the technology, replicate it, and then replace it.

But unlike Samsung and all other corporations, Apple didn’t have a joint venture nor did it own any factories itself. Instead, it was sending engineers that would train workers and bought billions of dollars of equipment that was used by them.

Years before Uber would become the largest taxi provider without owning a single vehicle or Airbnb would grow into the largest accommodation provider without owning any real estate, Apple was discovering how to be the world’s largest manufacturer without owning any factories. <…> It was working so intimately with suppliers that it had come up with a new offshoring model altogether. It married the best of both worlds, imposing a zealous level of control over its manufacturing processes, but with the lower costs and added flexibility of not actually running a factory.

And Apple is brutal in their demands to both quality and costs. The company is known to extract all the margins out of its suppliers for itself. Analysts report Apple earns ~80% of all smartphone profits, even though its shipment share is in the teens (~19%). Some suppliers were so eager to work with Apple they offered to do order at cost. Why?

Suppliers would put up with this seemingly inequitable deal because they got something less tangible but more valuable than profits. They got engineering help from Apple’s best—tuition-free, on-the-ground training—for multiple hours a day, day after day, for weeks and months leading up to a product launch. “So the reason Apple gets Chinese suppliers to work for them, for zero profits, is because the Apple ops engineer, following Tim Cook’s orders, is sleeping on a mat in their factory and helping them make that line efficient,” Guthrie says.

Apple wasn’t just coming to factories to place an order. Most of these were much smaller organizations that had to be expanded drastically. And managers who knew how to work the provinces’ leadership and get manpower and land would go all-in.

When Foxconn wanted to expand from cables to building the iPod, they reversed-engineer the device from available components and demonstrated it to Apple to prove they can do this. Apple’s employees were astonished as the device was almost indistinguishable from the real thing.

There was no way Apple could send the specs to some factory and wait for the parts to be built; instead, it sent teams of engineers to Japan, Korea, Taiwan, and China to find hungry vendors it could work with to co-create the processes. “There were a few truly groundbreaking mass production processes we were involved with, where we really had to go around to find the best people in the entire world—the peak of what humans have developed for some of these technologies,” says a product manager.

Apple seemingly understood that anything they build in China would be copied, just like it happened with all other companies. So their strategy became to always stay at the bleeding edge of what was possible. And then, their inventions become the industry norm. The book retells the famous anecdote about the invention of scratch-resistant glass for the iPhone.

Jobs reached out to Wendell Weeks, CEO of Corning, a glassmaker in upstate New York, saying he needed the hardest glass they could make. Weeks told Jobs about Gorilla Glass, something Corning had developed for fighter-jet cockpits back in the 1960s. They’d never found a market for it and abandoned the project. Now, this glass is practically in every flagship phone.

The volume of devices shipped by Apple is so big and grows so quickly, the suppliers who were able to scale along ended up selling to them almost exclusively.

Overdependence on Apple could create trouble, because Cupertino had a propensity to shift directions. So if, say, a company providing a key component was 80 percent dependent on iPhone revenue, and Apple made a design change obviating the need for that component, that company might collapse. In fact, many did; this happened repeatedly, and it caused Apple problems, including negative headlines in the media and bad will from suppliers. So Apple learned to find a sweet spot: to be the most important client for its suppliers, so it had leverage, but not so much that the supplier was overly reliant. The upshot of this policy, as a senior executive put it, was that “whatever rate we’re growing at, they have to grow that fast with someone else.” So as iPhone shipments soared, Apple encouraged its China-based suppliers to feed the Android market. As a result, Apple gave birth to the Chinese smartphone industry.

At first, China looked Apple with suspicion and frustration, simply because of how secretive and unusual their approach was. There was no joint venture! So Apple gathered a team and flew Tim Cook to give CCP a presentation where he explained what they were doing and commited to spending over $260Bn until 2021. Not even as some grand gesture, merely multiplying what Apple was spending annually already.

The suppliers and manufacturers of Apple devices leveraged the same know-how to help create Huawei, Xiaomi, Oppo, and Vivo, who collectively dominate the Chinese market and are actively expanding outward.

The ascent of China’s smartphone sector spoke volumes about Apple’s dependence on the very capabilities it had orchestrated, underscoring how China had gone from a land of cheap labor to one of sophisticated automation.

This statement might sound ambitious, but without having a single first-party factory in China, it was still one of the most frequent destination for the company’s engineers and managers.

Apple had been sending so many engineers to China on temporary trips that Cupertino convinced United Airlines to begin direct flights from San Francisco to Chengdu, three times a week, arguing that Apple would regularly buy enough of the thirty-six first-class seats to make it profitable. The 6,857-mile flight became United’s longest nonstop flight.

And when COVID-19 shut down airlines, Apple sent their engineers by jet.

United halted nonstop flights from San Francisco to Shanghai from March to October 2020. So Apple scrambled the jets. In the spring, Cupertino began sending engineers to Shanghai on private planes departing from San Jose, with a pit stop in Alaska. “Each jet could hold thirteen people, but we only sat six,” says a person familiar with the flights. “We wanted room and, you know, we’re Apple.”

The dramatic shift of high-tech manufacturing capabilities happened in just about three decades. And the result were so drastic, so when Apple tried to build Mac Pro in the US to placate Trump, it had to fly engineers from Foxconn to set it up

…the project made it only because Apple leveraged relationships with suppliers in the one country where it knew the talent existed. “We flew people from China to get it fixed,” one of the engineers says. “People working for Foxconn.” The irony is hard to overstate.

The last part of the book is about the peculiar position Apple now finds itself, stuck between the geopolitical ambitions of Xi’s China and the new reversal of domestic policies in the US, along with added risks around Taiwan and TSMC.

The company that once positioned itself as a rebel now forced to give up control over iCloud in China to local “partners” along with the encryption keys, remove VPNs and The New York Times from the AppStore and avoid critisizing China in any Apple TV+ content.

In 2017, explaining why corporate executives should be more up-front about their values and “lead accordingly,” Cook had told journalist Megan Murphy that “silence is the ultimate consent.” He went on: If you see something going on that’s not right, the most powerful form of consent is to say nothing.

Tim Cook has been an extremely succesful and ruthless CEO who made Apple one of the world’s richest companies. But he also made it hightly dependent on China during the worst deterioration of local political freedoms and their relationships with the United States. We will see what comes out of it.

The Myths of Venture Capital

Tue, 27 May 2025 15:45:14 +0000

Josh Miller, CEO of The Browser Company in his latest post on finally sunsetting Arc:

So when people ask how venture capital influenced us — or why we didn’t just charge for Arc and run a profitable business — I get it. They’re fair questions. But to me, they miss the forest for the trees. If the goal was to build a small, profitable company with a great team and loyal customers, we wouldn’t have chosen to try and build the successor to the web browser – the most ubiquitous piece of software there is.

I used Arc since the beta and I loved it. Arc was the first credible and ambitious attempt to reinvent web browsing that was actually able to get some traction. Unfortunately, it didn’t get enough.

Davie Pierce covered this for the story on the original announcement of Dia:

A strange thing has happened over the last couple of years, Miller says. Arc has grown fast — users quadrupled this year alone — but it has also become clear that Arc is never going to be a truly mainstream product. It’s too complicated, too different, too hard to get into. “It’s just too much novelty and change,” Miller says, “to get to the number of people we really want to get to.” User interviews and data have convinced the company that this is a power-user tool, and always will be. On the other hand, the people who use Arc tend to love Arc. They love the sidebar, they love having spaces and profiles, they love all the customization options. Generally speaking, those users have also settled into Arc — Miller says they don’t want new features as much as they just want their browser to be faster, smoother, more secure. And fair enough! So The Browser Company faced a situation many companies encounter: they had a well-liked product that was never going to be a game-changer. Rather than try to build the next thing into the current thing, and risk both alienating the people who like it and never reaching the people who don’t, the company decided to just build something new.

Many people believe that the venture capital and the “endless сhase for growth” is the culprit here. After all, The Browser Company raised $50M at a $550M valuation in 2024. I’d say these people don’t truly understand how startups and venture capital work.

Less than 0.1% of new businesses in the US receive venture capital funding. Venture capital is intended for companies that would never be able to get a bank loan. It’s an exception, not a rule, reserved only for the riskiest bets, because 90% of startups fail within 10 years. The one and only reason some crazy people are willing to fund them is they might find the next Google or at least an Uber and earn billions. The zero-marginal-cost nature of the tech industry this makes possible: once a product works, scaling it is far easier than scaling something tied to physical infrastructure (Walmart took 63 years to reach a $700B market cap).

Most startups in any given VC portfolio fail. But you don’t know which ones will, so investors place multiple bets, evaluating what it would take for each company to become a unicorn.

Fred Wilson, General Partner at Union Square Ventures, one of the best VC firms in the game, wrote in 2016:

Our first USV fund, our 2004 vintage, has turned out to be the single best VC fund that I have ever been involved in. We made 21 investments. We made money on twelve of those investments. We lost money on nine of them. And we lost our entire investment on most of those nine failed investments. The reason that fund performed so well has pretty much nothing to do with the losses. It was all about five investments in which we made 115x, 82x, 68x, 30x, and 21x. It wasn’t like we were swinging for the fences in that fund. Every single one of those 21 investments seemed like an intelligent investment decision at the time we made it. But many of them didn’t work. We lost all or almost all of our money on over 40% of our investments in that fund.

Mind me, this is the best fund of one of the best investors out there. The majority of funds don’t even beat the S&P 500. Carta published their data on the VC performance at the end of 2024:

For the 2017 vintage, for instance, median IRR fell from 16.8% as of Q4 2021 to 12.0% at the end of Q4 2024.

During that same period, the S&P grew by 2.5x—so you’d be slightly ahead if you’d gone with public markets (with full liquidity along the way).

Indeed, taking venture capital means committing to building a fast-growing product with enormous market potential. It’s not the only way to build a company, and nobody is forced to take it. In fact, anyone who’s raised funding can tell you how hard it is to convince fund managers that your startup is the next big thing.

But it might be one of the few viable paths to building a specific kind of company: one working on a truly novel product with untested technology and unclear product-market fit. One that needs to hire a full team and pay them salaries before generating revenue.

Like the Browser Company.

The same people who lament businesses for looking to grow their revenue probably care a lot about their paycheck coming on time each month. And this team has over 130 people in employment.

Arc was an extremely ambitious endeavour. The team went to rebuild the entire user interface from scratch, something neither Microsoft nor Brave were brave enough to do with their Chromium forks. But this unique interface built for the modern age is exactly what fueled Arc’s growth and recognition. Meanwhile, charging a subscription simply to use the product would kill it in the cradle since most of the growth was happening because of the word-of-mouth and all other browsers from big tech companies are free. Unfortunately, that also means they don’t care much about them.

The middle ground simply wasn’t possible.

***

There are many small myths around venture capital. People made wild statements based on the fact that people invest in companies, as if having Founders Fund on your company’s captable means Peter Thiel is personally making decisions for you.

Or you might point to Zed, an open-source Arc’s copycat as a counterexample to this story. Well, I’ve tried it out and I have to say that lifting 99% of Arc’s design still didn’t fix the natural roughness of consumer open source.

There are plenty of companies, even in software, that never raised venture capital, ranging from JetBrains to Basecamp, and a myriad of others you and I have never heard about. There are even unique companies that switched from a venture path to a cashflow-generating business like Buffer, who bought out their investors, and Gumroad, whose founder essentially got a gift from the investors who wrote the company off. Most successful venture-backed companies like Webflow or Figma raise funding on their own terms just to have a war chest.

Venture capital didn’t kill Arc. It gave us the chance to see what Arc could be. That’s more than most products ever get.

Why I Don’t Like CarPlay

Tue, 06 May 2025 15:54:13 +0000

Get into the car, take out the phone, put in into a flimsy piece of plastic that is somehow attached to the vents, connect the charger. Navigate to your destination.

That was the life before CarPlay. Now 79% of buyers aren’t even going to look at a car that doesn’t have it.

Don’t get me wrong, CarPlay is ifinitely better than what we had before. Car manufacturers operate in a very different paradigm than smartphone manufacturers. They source individual blocks and parts years in advance, so in the end your car problaby has a dozen of small computers running different areas, and multimedia was never their strong suit.

But after trying out Tesla, I have to say I prefer their built-in navigator and multimedia experience to CarPlay. Rivian is probably the same, along with other electric-first brands.

CarPlay has the lowest denominator problem. Since Apple doesn’t know what kind of screen the car will have, they have to adapt. In addition to this, Apple is extremely cautious and paternalistic. Because of this, the experience is worst than it could have been.

Here’s a simplest scenario that works great in a Tesla. Look at the map, pinch to zoom out, find a place you want to go but can’t be bothered to remember the address. Long tap, go. I do it all the time when I want to drive to a random place in a city or when I’m traveling in another country and maybe looking for the next sightseeing point. Or when you realize that the road is blocked and you have to force the nav to go a weird route.

You can’t really do that with CarPlay. All navigation apps, whether it’s Apple Maps, Google Maps or Waze, have to operate accoridng to their invisible rules. There’s no pinch-to-zoom because your car’s screen can be shitty, you have to tap a button to change the scale, adjust it and get back. You can’t just long tap on a specific point and drive there—never works reliably for me. And worst of all, even if you’re just at your driveway, you can’t simply take out your phone and do the thing there—your nav will be stucked on turn-by-turn directions, because Apple doesn’t want you to look at the phone. Unless you know the name of the place or an address, you’re screwed.

There’s more. Yes, CarPlay lets you have your music or podcast app, even if you use a third-party one like Castro. But you can’t scroll the song or the podcast episode. Again, because Apple believes you shouldn’t. The apps themselves are pretty bad. Go try and open Spotify simply to play Liked Songs or another recent playlist. It’s not obvious at all. Finally, renting a CarPlay-enabled car in Scotland reminded me how much of your phone’s battery is burned on this, especially with wireless CarPlay. The phone dies in a few hours.

Now, the unique advantage of CarPlay is that you indeed have your phone’s envirionment with all the apps. The world is not limited to the US and Western Europe and other countries have their own maps and music services, which probably have iOS apps. Still, Tesla locks you in with their providers. For a few months it didn’t know they turned a crossing nearby into a roundabout. And I have to use either Apple Podcasts or Spotify to listen to my shows. Well, actually I don’t, because I can continue the same show I had on my morning dog walk by simply tapping the Bluetooth button—everything switches automatically and almost instantly. Still, it’s less native (and I can no longer scroll the episode here either). The audio was a problem we solved years ago and I don’t get why people act so annyoned by having to connect to Bluetooth. The phone is already connected the second I get in anyway.

But the entire experience with built-in multimedia so much smoother I’m willing to lose some things. Plus, electric vehicles in general require good navigation specifically as the app has to know your car’s charge and energy consumption to build the route accordingly and plan for charges. Google Maps won’t cut it. With combustion, most of the time you don’t need to think about gas station, you only do this when you realize you’re low on gas already.

I’m not actively opposed to CarPlay. It would have been interesting if Tesla and Rivian used it. Although I doubt Tesla would be willing to drop their Mobile Connectivity subscription. The $10 I pay every month enable the built-in “phone” inside the car to stream music and even videos when you’re stopped. But to their credit, basic navigation and remote control are free, which can’t be said about many legacy manufacturers like Kia.

WordPress Doesn't Matter for the Future of Web

Mon, 14 Oct 2024 14:53:32 +0000

In case you haven’t been following, there’s a huge drama in the WordPress community. Matt Mullenweg, the creator of WordPress, launched a crusade against WP Engine, a company selling managed WordPress hosting and the competitor of his own entities. You can catch up on the story here.

But the funniest part of this WordPress drama is that WordPress is irrelevant.

Yes, it still powers around 40% of websites. But it’s not the future of the web. You *can* be both big and irrelevant. IBM created PCs as we know them and is currently worth $214Bn, yet nobody cares much about them. In fact, this scandal is probably the most attention WordPress received from the public in the last few years.

WordPress usage has plateaued. It’s not going to die (although Matt is surely trying a lot), but it’s unlikely to grow. Unfortunately, this is exactly when people realize it’s all a zero-sum game, and the fighting starts. Automattic probably believes that taking users away from WP Engine is a more reliable growth channel than trying to acquire new ones.

Technology products don’t get replaced by new, slightly better products. More often than not, they’re replaced by a completely different paradigm that makes certain use cases and applications radically simpler while throwing away aspects previously deemed important.

Back in the day, there were countless content management systems (CMS). PHP Nuke and its countless forks, Drupal, Joomla, and many others. As an aspiring developer, I coded my own (but I had no idea how to make other people use it). WordPress did actually win. But then WordPress wasn’t replaced by an improved WordPress. Instead, fully managed proprietary solutions took over.

You want to sell things online? Shopify. You need a landing page? Webflow or Squarespace. You want a blog? Substack or Beehiiv.

Yes, there are downsides. You’re locked in, and you might have to pay fees on your revenue. These platforms are less customizable, and the lock-in enables them to raise prices over time.

But they are also dramatically easier to launch and operate. Doesn’t matter if you’re an individual or a small business, the ability to start with a template and edit things without coding experience is amazing.

So, who uses WordPress? Organizations that prefer to host things themselves. There’s an army of WordPress developers and agencies who can build anything for them, and that alone creates the demand. If you want to self-host, there are no comparable solutions out there, exactly because people rarely develop for the outdated paradigm. People don’t trust WordPress the project and are asking around what are the possible alternatives with the same philosophy. There are practically none. One exception is Ghost, which is an open-source blog engine. But even Ghost moved toward a far more lucrative space of newsletters and paid communities, and it’s not a fully-fledged competitor to WordPress.

At my previous job, we had an agency build us a website on WordPress. And while we could edit texts on the landing page and add logos and news entries, adjusting anything else was near impossible. My current company has a Webflow website, and I have already rebuilt half of it and added an entire blog. We pay $340 annually for that privilege. Frankly, not much more expensive than premium WordPress hosters.

I know serious tech companies with complex software products and an army of developers who use Webflow for their landing pages. This way, they can keep developers focused on their actual product and let marketers and designers edit the pages as much as they need to.

Why I’m Excited About Meta Orion

Sat, 05 Oct 2024 14:32:11 +0000

I have read dozens of excited reviews of Orion like these pieces from The Verge and Ben Thompson. I have also seen a comparable number of people surprised by this wave of affection, like John Siracusa in the most recent ATP episode.

And I can understand this. Orion can be an impressive demo, but it’s still a demo. A prototype. A concept car. It’s reasonable to assume that Apple might have a similar pair of glasses in their labs somewhere and faced the same cost issues. “Apple doesn’t release prototypes”.

But I think we should stop giving Apple the benefit of the doubt. It’s their job to convince us. We’re excited about Orion precisely because Meta showed them, and Apple didn’t. I honestly do not care if they have something cooking. Yes, they have a good track record overall. But I don’t believe in a second we should buy in on their secrecy. They do it for their benefit, not ours. Otherwise, it becomes a cult with blogger shamans trying to interpret the divine signs.

Also, Apple Vision Pro is exactly this – an expensive prototype. And probably a failed one. Yes, this might not age well, like the early stories about the death of iPhone. Apple can turn this ship around. But it’s telling that even the people most excited about Apple Vision Pro acknowledge they haven’t touched it in months. It’d be perfect for an airplane, it’d be perfect for a workplace, and yet practically nobody bothers to carry them on their next flight. Who’s the Federico Viticci of Apple Vision Pro, i.e., the person willing to go to unimaginable lengths to use the device they love so much? I don’t know, I haven’t seen one.

Apple focused on a great passthrough and came out with a compromised device that is too heavy and expensive and is struggling to engage either users or developers. Meanwhile, Meta now has a $300 device with Oculus Quest 3S. It has passthrough that enables you to see around, not bump into furniture or play AR games. Nobody is going outside in these headsets anyway. But in the end, Meta owns the VR market and has an army of users and developers building all kinds of experiences. They have extremely popular Meta RayBan glasses that I actually see people wearing.

Now, each of these developers understands that the experiences they build might end up on unobtrusive AR glasses in a couple of years. Meanwhile, nobody is pressured to build apps for Apple Vision Pro, which still has neither Netflix nor YouTube and not much immersive content from Apple itself.

Meta showed us a convincing vision of the future, and they seem perfectly positioned to get there.

Social Media Platforms Have Killed Links

Thu, 12 Sep 2024 15:09:55 +0000

Within cells interlinked

Hyperlinks are the foundation of the internet. People say that Tim Berners-Lee invented the internet as we know it when he connected hypertext with TCP/IP. And yet, in 2024, hyperlinks are second-tier citizens at best, as the most popular social platforms don’t support them or penalize you for trying.

Facebook started deprioritizing links in 2017. Zuckerberg confirmed this later by announcing a broader effort to prioritize “meaningful social interactions” on users’ news feeds. The stated goal was to focus on posts from friends, family, and groups over content from businesses, media outlets, and publishers, particularly those containing outbound links. Twitter joined the party in late 2020, as they also wanted to reduce their reliance on external content.

This shift forever changed the internet.

Platforms dislike links because they are, by definition, links to other websites. You click a link, and your attention goes elsewhere. But that’s how the internet works. Links are the foundation, and people use links for all kinds of reasons. A desire to start a discussion about an article or a video is one of the incentives to post on social in the first place.

Meta and Twitter imagine a future where all content is directly syndicated through their properties. Even if they don’t support this kind of content natively. Have you tried using Twitter Articles? They’re primitive, can take 10 seconds to load, and in the end, the engagement figures are not too different from an ordinary post with a link. Twitter videos can’t replace YouTube in their dreams. Most importantly, as an ordinary user, you can’t just reupload someone else’s video or a news article to Twitter to share it natively.

So you paste a link, and nobody sees it because social platforms push users towards the algorithmic feeds they control. As a result, people post less or nothing at all.

It doesn’t mean there are fewer posts. But more and more content is instead coming from other accounts that leverage the native tools and mechanics to be favored by the algorithm. And I don’t think their content is particularly good or tasteful.

One interesting signal in social products is people using weird hacks to overcome the platform’s limitations. In early Instagram, all photos had to be square. But people wanted wide and portrait instead, so they used apps to add bars on the side until Instagram enabled this natively.

“Link is in the first comment” became such a hack with the hope that the network won’t supress it, even though the conversion would drop drastically anyway. In fact, threads on Twitter are another example of a hack that eventually became a part of the product. Another alternative people used was posting a screenshot of a note instead. They wanted the ability to post long-form content, at least on certain occasions. By the way, Twitter of 2024 definitely suppresses the reach of images that contain lots of text, you can easily see this from the view counts.

It’s almost like we need a new social network for links. Unfortunately, we did have one for a while, and it died. I’m talking about Artifact, an AI-enabled news app launched by the founders of Instagram, operated as a social network with nothing but links. I loved the app, but there weren’t enough people enamored by this concept.

The real question is whether we go next.

What Makes Telegram Special

Tue, 06 Aug 2024 11:12:53 +0000

Telegram is a secretive social media platform with 900 million users. Their employees aren’t allowed to talk about their jobs. It only has one product manager in its founder. And they got to 900 million users without turning into Facebook but also aren’t earning as much.

Telegram is not really a secure messenger app. It’s one of the largest social media platforms. But certain product choices and the location of its audiences make it almost invisible to people based in the United States or Western Europe.

When you first open it, Telegram looks similar to WhatsApp. WhatsApp’s founders famously focused on keeping it simple and clean. For them this meant not adding anything else: “no ads, no games, no gimmicks”. So they added nothing to the app, even years after Meta bought it. About a year or two ago I noticed WhatsApp started adding very specific features that were obviously lifted from Telegram, like channels. But still much slower.

Telegram can get incredibly complex. On top of DMs and group chats, it can host super groups of up 200,000 people. Anyone can create a channel, a publicly-facing broadcast similar to a Facebook page. The largest one now has 55 million followers. You can save anything in your own personal space, and you can organize chats by folders. In addition to channels, Telegram pioneered other UI paradigms, like video circles, which are vastly superior to voice notes (lifted by WhatsApp again).

But despite all this complexity, Telegram hasn’t turned into the infamous Facebook main page. All these features are carefully hidden. Pavel Durov says that he is the only product manager and personally oversees how the app looks. It certainly looks like it. Despite being a social platform, there’s no unified feed, even though it’d provide the best possible ad space.

Compared to Signal or any Meta’s messaging app Telegram provides the most seamless and polished experience. You can instantly switch between conversations. You can quickly find anything shared inside these conversations, including images, files and links. And there are specific searches for all of these categories. Back in 2017, I often had to look up some old files and images in Facebook Messenger, and it was atrocious. Most importantly, you can easily use it on multiple devices with great desktop apps (there are two, for some reason), and everything is synced through the cloud.

Telegram desktop app

Meta provides you with half-baked web-based apps. And with WhatsApp, you have to relink it to the phone constantly. But Telegram can only do this because the messages aren’t encrypted end-to-end. Telegram started as a “secure messaging app” (they launched quite a bit before Signal). But they aren’t using this tagline anymore since it’s not a differentiator, and it’s not true. With Telegram, you have to trust the team not to peek and keep their servers safe and secure. You can create dedicated E2E chats, but only manually, and their functionality is as limited as you’d expect; they’re tied to a particular device and don’t sync at all.

But most users don’t care. This trade-off makes Telegram easy to use, and the team focuses much more on social features anyway. In regions where Telegram is popular, channels replaced newsletters as a form of revolt against algorithmic feeds. They’ve recently revamped user profiles to make them more public-facing and let you pin your channel right there. You can write long texts, share images, and not worry about links being suppressed. On the other hand, you will never get any algorithmic boost. You must proactively look for these channels or get a link elsewhere. In a world where social platforms demand specific content and try to commoditize all content, Telegram feels like a remnant of the old web.

Telegram did add Stories, and I think that was one of the rare misses by the team. Having someone as your Telegram contact is a weaker signal than following on Instagram. Imagine discussing your upcoming partnerships while your stories show how you left the bar at 4 am. Considering the “messaging” part with groups and DMs, they should have made stories visible to “Close Friends” only and asked you to choose the list manually. Then, it could have worked.

Telegram has a notoriously lenient approach to moderation, so you can see accounts from both sides of the Russia-Ukraine war. But unlike modern Twitter, Telegram doesn’t shove it in your face, so even Zelenskiy is happy to use it. While each channel has a URL, the reading interface on the web is extremely basic, and aggressively pushes you to go to the app instead. They definitely could have improved it but chose not to. Telegram is its own kingdom. That’s one of the reasons you rarely see links to its content. Social media apps often use built-in browsers to keep you there, but Telegram goes beyond and provides reader views for most websites, especially news media. No reason to leave the app. Last month they even added its own multi-tasking to Telegram, where you can collapse the browser and continue talking to people or reading channels.

Telegram is the app people use for everything. Talk to their partner. Set up a dinner party. Discuss work projects. And read the news, of course. You can create a group with dedicated threads that look like a Discord/Slack server. I know companies with a few thousand employees that use Telegram as their internal comms tool. It has downsides, of course, as you don’t control your employee’s accounts, and they can mix and match their personal conversation or easily forward messages. Companies write bots to accommodate (like deleting the person from all chats, but it’s still clunky).

Topics in chats

With Telegram, you need a phone to set up your account, but you don’t have to share it. Instead of giving out your phone with iMessage, Signal or WhatsApp or sharing your personal Instagram page, you can send a nickname. This made Telegram especially popular among crypto people who prefer to stay semi-anonymous. It is the default communication app for crypto projects, both for their internal ops, connecting to other businesses and community management. They could use no other single service for all of these things instead.

How does Telegram make money? It’s complicated. I’ve seen Durov’s previous project, VK, and have used Telegram since its early days. I think this team is opposed to ads and what it takes to make them work on some philosophical level, and they don’t want to learn. That’s part of the reason Telegram feels nice. There are ads, but they’re pretty unobtrusive since there’s no global feed, and they operate mostly by context with minimal targeting. Telegram also doesn’t have much data on you, so I can’t imagine that CPM is too high. You see these ads in channels, and most of them are for other Telegram channels.

Telegram also has a Premium subscription at $5/month. The subscription removes ads and raises some artificial limits in the interface (the number of chats in a folder); it also gives you a unique check mark and animated emojis. Some people buy it because they like it, and people who use Telegram heavily in their work buy it because it’s just easier than fighting the app. Back in January, Pavel reported 5 million paid subscribers. If we don’t take churn into account, that’s an annual runrate of $300 million. Very impressive, but I’m not sure if it still covers the expenses alone.

Running Telegram takes a lot of money. Partly because they store everything in the cloud, so your mother potentially doesn’t have to figure out how to move her WhatsApp messages from her Android phone to a new iPhone (saying from a personal experience). In 2017, Telegram raised $1.7 billion from investors and announced they would launch their own blockchain TON. The SEC considered their tokens to be unregistered security, so Telegram backed off. They spun off TON, made it open-source, and integrated with it later.

Crypto is a big part of Telegram right now (but you might never see it like other complications). Both Twitter and Instagram canceled their NFT avatars. But in Telegram there’s a crypto wallet, there are web-based mini-apps, and a very basic appstore. The Hamster Kombat game was launched on this platform and gathered 300 million users.

There aren’t too many useful apps yet, but it’s interesting to watch it develop. For years, people were looking for the WeChat of the West. I don’t believe there will be such an app, for many reasons, but Telegram is the closest thing, and it’s curious both use the web stack for these mini-apps.

We don’t know much about the team. Telegram says they have only a few dozen employees working on the core app. The fact that they were able to get to 1/4th of Meta’s userbase is incredibly impressive. Still, we can only rely on their words, as nobody else can share anything about their work. And if you read any stories about Elon’s dreams for X, Telegram might be much further along this path. Telegram Premium probably has more paid users than X Premium. But that’s not all.

Telegram's LinkedIn page

Pavel Durov is obviously unhappy with Apple and Google. He’s been at them for years. iOS and Android are critical platforms for Telegram, as the app completely depends on their rules and whims, from content moderation to limitations on what and how they can sell inside the app. You can buy Stars to pay for goods and services inside the Telegram ecosystem, but Apple takes 30%.

But inside this crypto wallet (which isn’t available in the US as many crypto things), you can store USDT, which is the next best thing to a digital dollar. And it’d take a flick of a switch to let users pay for content and services with USDT that they can buy anywhere. This would simultaneously solve Telegram’s problem and push its audience toward crypto.

Telegram is in an unstable equilibrium. They need to perform a careful balancing app with SEC, which already destroyed their original plan and forced them to return most funds, and with Apple, which can destroy their app as they did with Fortnite on iOS. I believe they’re betting they can do this, and their lives will be made easier with the rising pressure on Apple from regulators and the growing adoption of crypto itself. And Telegram might have the best chance to ride both waves.

How Twitter Changed Since Elon's Takeover

Sun, 28 Jul 2024 16:33:40 +0000

We’ve all seen these posts about how Twitter fired 80% of its employees, and then nothing happened. Seems it ultimately did affect their revenue:

Internal documents obtained by The New York Times show that, in the second quarter of this year, X earned $114 million in revenue in the United States, a 25 percent decline from the first quarter and a 53 percent decline from the previous year. The company aims to reach $190 million in U.S. revenue during the third quarter, bolstered by advertising associated with the Olympics, football and political campaigns, the documents said — but that target would still set the company’s quarterly earnings at 25 percent less than they were last year.

These people might argue that these two items aren’t correlated but I doubt it. Twitter didn’t have a great advertising engine, so most of its revenue came from generic brand advertising, and companies have brand safety teams for a reason. Elon, a “libertarian capitalist”, still seems to have a problem with the free market when it’s against him.

I think you can believe that Twitter was grossly over-staffed, badly managed and dysfunctional before Elon bought it and ALSO believe that he fired far too many people in the wrong way without fixing anything much. Benedict Evans

I really loved Twitter. I think it changed a lot, mostly for the worse. But because Elon has the worst fans and the worst critics, it’s extremely difficult to find non-polarized opinions on this. Being an eternal centric, I wanted to gather all the details. So stay with me if you’re interested.

The Good

Let’s start with good changes. Twitter was an anemic company. It always seemed like they were terrified to change anything in their product. Probably, because they didn’t understand why it worked and were too terrified to break it (well). And even when they came up with something new, like the predecessor of Twitter Articles, they’d make it available to ten people they knew in San Francisco for two years.

Elon came in and made them ship all those things to the point where the audience largely attributes Community Notes to him. And this is almost reasonable. Community Notes is a fascinating tech product for the social media era. The main problem with fake content is that it spreads much more easily than any rebuttal. Community Notes effectively chase down and clarify the original posts and even notify the users who interacted with them earlier.

There were all these stupid ideas about marking and signing AI content. It’s not like people want to verify if something they’re seeing is true. It’s confirmation bias: they seek something that confirms their pre-existing beliefs. You don’t need AI to create an engaging fake, like that photo of Kamala Harris with Epstein. And the problem with forcing OpenAI or Midjourney to stamp their content is that the next day there will be a Github repo for removing that signature. In addition to a dozen open-source alternative image generators, you can download from anywhere.

Community Notes solve this problem and allow citizens to keep corporations and politicians in check. You can’t just post obvious falsehoods on large accounts anymore (well, it’s still as complicated as everything in this world).

Under Elon, Twitter rolled out a reasonable paid subscription (although I think three tiers are too complicated). They released a bunch of things they built before and even some completely new ones, like X Hiring. Even little things, like finally putting your pinned lists at the top of the feed on the desktop.

After losing most of its advertisers, Twitter has seemingly revamped its ads team. One representative even reached out to my company. However, most ads I see are still from Twitter natives, like content creators promoting their accounts and newsletters. I’m not sure if there’s a lot of money in this.

Twitter now has a powerful and engaging algorithm that should excel at onboarding new users, something they ultimately struggled with before. Many of my friends tried Twitter but didn’t understand the point of it. Twitter would only work if you follow the right people. Now it’s basically automatic and on par with Meta’s tools.

Their goal was obviously to keep users on the website, and in this context, shutting down Revue and suggesting authors use Twitter Articles instead makes sense. The feed on the mobile app also changed; now, if you open a video, you can just swipe to the next one, bringing it closer to TikTok.

This is also why they prohibited alternative clients. I know many hate them for this, but purely from the business perspective, this change made sense. The old Twitter knew they had to do this but preferred the tactic of boiling the frog. The new Twitter comms on this were far from ideal, but they pulled through.

Finally, monetization created obvious incentives to set up and run accounts that posted various content on Twitter. Some of them good, some of them… well, there’s the next section.

The Bad

Twitter botched down their update to the blue check program. Look, this thing was flawed. Twitter employees would sell you a blue check for a couple of grand if you knew where to ask. At the same time, I met companies that struggled with impersonators but couldn’t get verified. The blue check was supposed to mean you’re who you are, but it became a power status symbol instead. Sometimes, when people abuse your product’s limitations, it’s a sign that you need to change it.

However, the logical solution would have been to keep the existing accounts and then offer all who are willing to get a blue check if they subscribe to Twitter and provide their ID for verification. There are KYC-as-a-service companies that do this for fintech startups, so it’s pretty easy. Don’t want to show your documents? Nobody has to have a blue check.

Instead, we got the current mess. Through several iterations, Twitter still got to the point where they awarded the blue check to notable accounts. But the important change is that it can be a DogTurd69, a famous “thought leader”. At first, blue check became coded pro-Elon and even right-wing. Now they don’t mean anything, except that your replies are boosted to the top.

You probably have seen all these companies with gold checks. Don’t believe they paid $12 000 for a year. Twitter gave these to the most notable companies, probably with the expectation that they would start paying next year. We’ll see how many do (or if Twitter itself changes course). Proving the affiliation of specific people (like writers or media reporters) with a badge sounds like a great idea. However, most marketers I talked to believe it’s just too expensive for the little value it provides now.

In addition to losing third-party clients, Twitter has become more closed overall. You can’t browse it without logging in. Often, you can’t even reliably link to it on other platforms, and there are services like vxtwitter whose entire goal is to provide rich previews. In the process, they closed down the API and killed even some of their own tools, like the Card Validator, provided by all social platforms. Finally, outside links are getting suppressed to a degree you shouldn’t even bother posting them. I’m pretty sure they now even track links in the first reply from the author, at least based on my own tests. They did this before Elon, but now doubled down.

Surprisingly, all of this didn’t help beat the bots, which was one of Elon’s key goals as he was getting into it. Just try tweeting the word “metamask,” and you will see a bunch of blue-check accounts trying to scam anyone reading it. It is becoming increasingly difficult to distinguish these bots from regular users since they look just the same, and LLMs are getting very good.

Old Twitter’s algorithm focused on ranking the people you follow with an occasional mixin. The new algorithm is much closer to TikTok. Even if your account is empty, it will create a pretty engaging feed. The problem is this engagement often comes from the worst aspects of human behavior. I’ve never been keen on the idea that social media thrives on this alone, but this particular algorithm might. The Twitter of today is toxic.

Since firing their entire trust and safety team, Twitter has been far more hands-off in moderation. People flocking to Mastodon would refer to the app as “filled with Nazis”, and at first might have been an exaggeration. But now I constantly see racists, incels, and right in my feed, even though I try to have zero interaction with them or their tweets. They hate minorities and ask if women are sentient. I don’t want to be around these people and don’t think free speech is the panacea in this case. The problem is not they’re not banned, but they’re clearly getting amplified, since I don’t follow them.

Everyone who has moderated a forum knows you must do this very diligently if you have even a small audience. If you don’t set the standard and have new people coming all the time, it’ll fall to the lowest common denominator. Is it possible at the scale of hundreds of millions? Well, Twitter could do this before, and not because it banned all conversations. But it certainly wasn’t promoting fringe ideas to everyone’s feeds. Easy to see why brands might not be happy with it.

The new algorithm is very polarized. It gathers the most engaging feeds and leaves everything else to rot. If you’re not getting into the For You feed, 90% of your followers are unlikely to ever see your content. Elon made it very obvious by putting the view counts up. So, authors are incentivized to go to the extremes.

Let’s get to the original thought. Is Twitter operating smoothly? Certain minor things have been broken for months for me. For instance, if I click on any tweet in my bookmark, the page jumps to the top. There are many places across the product that work like this now.

The Ugly

I spent the last weekend without my phone, avoiding all the feeds. When I opened Twitter afterward, it just felt so unnecessary.

There were several reasons I liked the old Twitter. It was the most open social network where you could actually post links and easily copy text. There was an ecosystem of services and bots built on top of it. And it also was the most real-time network for news junkies like myself.

I don’t think I’ve ever befriended someone in real life after following them on Instagram, but with Twitter, it did happen. I think it’s because you can get acquainted with someone much more easily by reading their thoughts instead of looking at their sunset photos.

Well, Twitter was actually a great platform to share photos as well! People were always stingy on likes, but the engagement was comparable to or better than Instagram. And the resolution and quality were much higher. I’d rather share updates about my trip on Twitter than try to jam it into a short description under a gallery of 10 photos that nobody will scroll through.

As for being real-time, I think it still holds this role, and alternatives haven’t caught up (and some probably don’t even want this).

That text-centric nature has always limited Twitter’s appeal to most people. It was never a truly big platform. However, it also self-selected the users who were interested in this—the people I wanted to follow, read, interact with, and potentially learn from.

Twitter is basically a large forum where you own all of your replies. I was an active user of many thematic forums that just died, and every time, it was almost like I lost a part of my identity. Twitter allowed you to keep this identity between “forums” and discussions. But this aspect is only as important as the network itself.

Now, most people who were on Twitter are still there, but many have left, scattering to Mastodon, Threads, BlueSky, and others. Of these, only Threads seem like a reliable alternative for me. Mastodon by itself still feels like something made by geeks and for geeks, and the lack of any algorithms and recommendations really limits your community. I’m not talking about “growing your followers”, it’s hard to simply find interesting people to follow. But, if Meta makes Threads fully federated, there will be a unified community where you can follow people from celebrities and your friends to geeky developers.

We will see what happens to Twitter, but I doubt it will have a big turnaround.

I guess growing up means watching the things you love die.