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.