Don’t sleep on Chef (I can’t believe it works this well)

So, you know how there are a ton of AI tools popping up that promise to build apps for you? Tools like V0, Lovable, and even Firebase Studio can whip up some really nice-looking user interfaces and even give you decent React code. The big problem, though, is that they pretty much fall flat when it comes to the backend – all the essential stuff like user sign-ins (auth), handling file uploads, managing databases, and user accounts that make an app actually work. You usually end up having to ditch the AI tool and build that backend part yourself, which kind of defeats the purpose.

Enter Chef by Convex. This is a new AI app builder that specifically tackles that backend problem. The presenter actually admits he kind of pushed Convex (a company he's worked with before, though they didn't pay for this video) to build this because he was frustrated with the limitations of other tools. Convex's whole deal is providing a complete backend solution – auth, database, file storage, server functions, the works – all tightly integrated, especially for React/TypeScript developers. Chef uses this Convex platform under the hood to generate not just the UI, but a fully functional backend defined entirely in code.

One of the most impressive demos shown was building a Slack clone. Chef generated the whole thing – channels, messages, user authentication, and even live updates where messages appear instantly for everyone – from a single prompt. That's the kind of thing that's usually really complex to build manually, let alone get an AI to do it. The secret sauce here is Convex's architecture. When data changes in the Convex database (whether triggered by a user action or even manually editing the database), any part of the UI using that data automatically updates. This is handled by Convex's special hooks (like useQuery in React) that manage WebSocket connections behind the scenes. It makes building real-time features surprisingly simple.

Another core idea is that your entire backend configuration lives as TypeScript files inside a convex folder in your project. Your database schema, your server functions (like creating a channel or sending a message) – it's all just code. This is way different from clicking around in AWS dashboards or configuring endless services. It's more like "code as config," making the backend version-controllable and easier for teams to manage. Plus, Convex provides end-to-end type safety, meaning if you change your database schema, you'll get TypeScript errors in your frontend code if it breaks something – a huge win for catching bugs early.

How does it stack up against others? Well, Firebase Studio, despite looking beautiful, apparently struggled immensely. It couldn't even get basic auth working in the presenter's tests and was painfully slow. V0 is much better at generating polished UIs (using popular libraries like ShadCN), but you still have to manually integrate separate backend services (like Supabase for auth, Upstash for data storage). Chef's output UI is admittedly quite basic or even "ugly" right now, but the argument is that it's much easier to take Chef's working full-stack code and make it look pretty later, than it is to take V0's pretty UI and try to build a working backend for it.

Chef isn't perfect, of course. It sometimes runs into errors during generation, especially with type checking, though it's usually able to fix them itself. Integrating complex third-party services can also be tricky; an attempt to add payments using the Polar SDK failed due to type errors, highlighting that AI still struggles with external APIs. The UI it generates is very basic visually. There's also a mention of a potentially confusing "chef token" system for measuring usage, separate from the underlying AI model's tokens.

Despite these points, the overall impression is highly positive. Chef successfully generated a working "Party Planner" app with event creation, user sign-in, and persistent data. Adding features like file uploads for event banners was done with just another prompt – something considered very hard with other tools. The presenter, who was initially hesitant about Convex's all-in-one approach, now sees its huge value, especially when starting new projects with an AI builder like Chef. The key takeaway is that if you want an AI tool that can generate a real, functional full-stack application, backend included, Chef by Convex seems like the most promising option right now.