Convex is Not Efficient for AI-First Devs

After spending several weeks investigating Convex for my stack, I’ve concluded it’s not the right choice for indie developers or small teams.

First, while Convex claims to improve development velocity, the lack of comprehensive documentation and AI specific knowledge bases means you actually end up wasting more time. It promises speed but delivers a learning curve steepened by obscurity.

Second, there's the issue of hosting. The CEO of the t3 stack has praised Convex's engineers as world class, and I don't doubt that for their public cloud offering. But if you venture into self hosting, that's where the story ends. The maintenance costs are high, and the error message stacks are often too convoluted to easily debug.

Third, when you compare it to a standard vanilla Node.js stack using something like Next.js, Drizzle, and Postgres, Convex just doesn't hold up. It's too new, and there are no established best practices. For instance, how do you integrate it with something like the Vercel AI SDK? While there are official examples, they don't seem to be used in production. This is a significant drawback compared to mature, battle tested solutions like Vercel with Supabase or Neon.

Ultimately, Convex is just too far from being a mature ecosystem for me to recommend it at this time.