DevTools are winning. Adoption is up. Usage is up. Developers are shipping faster than ever.
So why are so many dev tooling companies struggling?
Because the funnel is gone.
Docs used to be the funnel
For years, dev tools growth followed a simple pattern. A developer hit a problem. They Googled it. They landed on docs. They learned the tool. And somewhere along the way, they discovered a paid product.
Docs were education and marketing at the same time.
Even then, conversion was hard. Most users showed up anonymously through npm installs, CLIs, or CI pipelines. Downloads and GitHub stars measured popularity, not revenue. Community sentiment became the strongest signal of success.
It worked, barely.
AI deleted the middle step
Then AI changed how developers learn.
Instead of reading docs, developers ask an assistant: "How do I do X with Y?" They get working code. They get best practices. They get answers instantly.
No website visit. No docs page. No conversion moment.
AI did not optimize the funnel. It removed it.
Usage went up. Revenue did not.
From the outside, everything looks fine. Framework adoption grows. Communities expand. The tool feels more popular than ever.
Internally, the numbers tell a different story. Usage is decoupled from revenue. The value is real, but no longer priced.
This is the dev tooling paradox of the AI era.
Open source pays the price
Most dev tooling, especially frameworks and infrastructure, are open source. That was always a tradeoff, but it worked when a small percentage of users eventually paid for something.
As that path disappears, the cost stays behind.
Teams shrink. Roadmaps slow down. The first things to go are the passion projects and quality-of-life improvements. Then maintenance suffers. Bugs linger. Security updates slip.
To users, it looks like abandonware. To maintainers, often the founders, it is a moment of existential math.
AI is replacing the product too
The problem is bigger than marketing. AI is starting to replace what dev tooling companies used to sell.
Take Tailwind. Premium component packs and UI accelerators were once an obvious paid add-on. Now a developer can prompt an assistant for a Tailwind hero section and get something usable instantly.
That output is not magic. It is a remix of open source. Projects like shadcn made high-quality components free. AI turned searching, adapting, and copying into one step.
When good enough is instant, paid becomes optional.
The quiet risk
Open source has always struggled with monetization. AI accelerates that problem. Not with a crash. With erosion.
Maintainers lose time and resources. Dependencies receive fewer updates. Critical infrastructure quietly degrades.
Companies keep building on it anyway.
The real impact will only be visible later, when the tools modern software depends on no longer have the people or funding to maintain them.
The funnel did not fail. It disappeared.
Who survives this?
Not everyone loses equally. Some dev tooling companies have built-in moats that AI cannot easily replicate or bypass.
Database and state tools. Supabase, PlanetScale, Neon, Turso. Once your data lives somewhere, migration is painful. The switching cost is real and grows with every row. AI can help you write queries, but it cannot move your production database without significant effort.
Auth providers. Clerk, Auth0, WorkOS. Authentication touches everything. User sessions, tokens, permissions—ripping out auth is a multi-week project even with AI assistance. The integration depth creates genuine lock-in.
Infrastructure platforms. Vercel, Railway, Render. Your deployment config, environment variables, CI/CD pipelines, custom domains—all of it accumulates. Moving means rebuilding operational knowledge, not just code.
Observability and monitoring. Datadog, Sentry, LogRocket. Historical data, custom dashboards, alert configurations, team workflows. The value compounds over time in ways that are hard to replicate elsewhere.
Payment processors. Stripe remains sticky not because the API is magical, but because payment infrastructure touches compliance, user trust, and financial operations. You do not casually swap payment providers.
The pattern is clear: tools that store state, accumulate configuration, or sit at the center of operational workflows survive. They have gravity.
Who does not survive?
Everything that can be generated fresh has no moat.
UI component libraries. shadcn already proved that beautiful components can be free and copy-pasteable. AI makes this even easier. Why pay for a component kit when you can prompt one?
Code formatters and linters. Useful, but entirely reproducible. Prettier, ESLint—AI knows their configs by heart and can generate them instantly.
Static site generators. The output is just files. AI can scaffold a Next.js, Astro, or Hugo site in seconds. There is nothing to lock in.
CLI utilities. Helpful, but trivially replaceable. AI can write a CLI tool from scratch faster than you can find the right npm package.
Boilerplate and starter kits. The entire category is under threat. Why buy a SaaS starter when AI can generate one tailored to your exact stack?
These tools are not bad. They are just infinitely reproducible now.
The split
If I had to put a number on it, here is my gut feeling:
~25% of dev tooling companies have meaningful lock-in. They store your data, your config, your operational state. Switching costs are real. These companies will survive and likely consolidate.
~75% do not. They are libraries, utilities, frameworks, or convenience layers that AI can regenerate or replace. Revenue for these will continue eroding unless they find a different moat.
The brutal truth: most dev tooling companies fall into the second category.
What this means
If you are building or running a dev tooling company, the question is simple: does your product accumulate something that AI cannot trivially recreate?
If the answer is yes, you have time. If the answer is no, the clock is ticking.
The funnel is not coming back. The only path forward is building something that compounds—data, configuration, workflows, trust. Something that gets stickier the longer someone uses it.
Everything else is a feature AI will eventually absorb.