When we were building the AI web app builder experience for cto.new, one question kept coming up: what backend should we bet our user experience on?

A while back, we shipped a product called backengine.dev, where we rolled our own set of backend services for vibe coders (before the term was coined). Fortunately, the market has come a long way since then and there are a few complete backend-as-a-service options to choose from.

Most BaaS weren't actually built for AI app building, even if it’s now one of the sectors biggest growth drivers. They were built for human developers who can click through a dashboard, inspect a database schema, hand-roll an API layer, and spend an afternoon configuring auth. Things are a bit different when an LLM is tool calling its way around the stack.

Vibe coding a backend is hard

We know this from our experience building backengine.dev. The frontend isn’t too tricky - it’s easy to click around to check it’s what you asked for and it usually kinda works. The backend is much less tactile, more brittle and difficult to visually debug. Any part of the backend stack can go wrong where a non-technical user has basically no chance of fixing it.

We needed a backend that collapsed all of that complexity into something an AI could reliably work on and a beginner could easily interact with.

Why Convex?

Convex gives you a complete backend suitable for pretty much any kind of application. It includes a database, auth, file storage, serverless functions, scheduled tasks, and logging all in one place.

Even better, it’s all in typescript. It’s just code, so it’s easy for an AI to work with.

It’s also incredibly easy to self-host, meaning we can just run it in our user’s sandbox alongside their TanStack app. When they’re ready, users can easily deploy to Convex Cloud and take advantage of the scalability of PlanetScale.

For cto.new users, this means a web app project where everything is pre-configured and ready to go from the start with an easy path to scale. You open a chat, describe what you want, and the agent builds against Convex without you ever needing to know about it.

It’s a first class use case

The thing that gave us the most confidence is that Convex has learned the use case from first principles by building their own vibe coding tool on top of it.

Chef is their AI-powered app builder. It generates a full-stack TypeScript application, with real-time data, auth, background tasks, the works, all running on Convex. They've run hackathons around it. They've benchmarked different LLMs against their platform to figure out which models write the best Convex code. They've published prompting guides for non-developers.

They're finding out firsthand what AI agents need from a backend by building and shipping their own agent. The lessons from Chef feed directly back into the platform and it shows.

What this means for cto users

Every web app project on cto.new comes with Convex built in. Database, auth, file storage, functions, scheduled jobs, logs. It's all there in the Convex tab. You don't need to configure or deploy anything, you just start building. When you’re ready, it’ll scale with you.