Excited to share we just launched a new product, an AI coding platform (i.e., a vibe coding platform) called Converge. We’ll get into why we built Converge and how it is different from the plethora of other vibe coding platforms out there, but before we do that, we need to back up a few months.

P.S. For being a supporter of ours, here is a code to get 50% off your first month of Converge: SUBSTACK50

It all started in the summer

In the summer we were exploring pivot ideas from PromptHub. PromptHub is successful by most metrics (users, usage, retention, general love), but we were curious about building in the application layer, rather than the tooling layer. PromptHub is still up and running and being supported.

We had a bunch of ideas, but started working on a tool that was essentially a chat application built from the ground up to focus solely on B2B, with a big bet on MCP. Our main competitor would be ChatGPT Enterprise, Glean, homegrown solutions, things like that.

We had heard a lot about Convex (take note because Convex will come up again) and specifically why it was so great for building AI apps, so we gave it a shot.

We actually got started with their vibe coder, Chef.

Chef was a fun way to get started, and we eventually exported the code and picked it up in our IDEs (Cursor + Claude Code).

Working with Convex felt amazing, the real-time sync aspect of the platform is something I could never live without, and it was particularly helpful for our AI chat app.

Their components were also really helpful, in particular we used their RAG component and AI agent.

Convex components differ from npm packages because they are fully isolated backend modules with their own schema, data, and execution environment, and only access your app through explicit interfaces.

This makes them safe by default and avoids the tight coupling and side effects common with traditional libraries.

So we were working on getting an initial version up and running. We ran into some challenges getting MCP integrations to work smoothly (lot more to say there), but come mid-September, we were getting close to launch.

Then Wednesday 17, 2025 hit. I was on my honeymoon in Nice and was clicking around on my laptop while my wife was getting ready, and then I saw this.

I immediately cloned the repo and spun it up locally in a few minutes. I dropped my team the link to the announcement and got back to enjoying my honeymoon.

Forking Chef

By the time I was back it was clear we had a few more weeks of work to get a v1 of the chat app out the door. I personally was losing interest in the idea and was drawn much more to the idea of forking Chef and doing something with it.

With not much runway left, and with the need to only focus on one idea at a time, we dropped the chat app where it was and focused 100% of our attention on building on top of Chef.

Forking a repo and making it into your own product is no small task, but it’s doable (Cursor, Windsurf etc). To be honest, we didn’t have many better ideas. We ran into a lot of issues and had to rip and replace a lot of code (a post for another time).

But aren’t there a bunch of AI coding tools already?

Our intro into Convex via the chat app had us fully Convex-pilled, we were believers.

I had previously tested and used most of the major vibe coding platforms (Lovable, Replit, Bolt, etc), and to be honest, none of them were as good as Chef. The first 10 minutes using them can feel great, but they all quickly break down.

Chef has some limitations but they were much more solvable compared to the other platforms, who had deeply nested issues based around their backend architecture and database choice.

That is where the main difference between Converge and all our competitors is: Every app built in Converge gets a Convex backend which includes database, file storage, real-time updates etc, while other platforms use Supabase.

Why Convex is 10x better for AI coding

Convex keeps the entire backend, (schema, business logic, permissions, and data access) as TypeScript in the codebase.

That matters for AI because:

• the AI can safely read and modify backend logic (it’s all code)

• permissions are explicit, not hidden in SQL rules or dashboards

• there are fewer configuration surfaces to drift or break

• the system is structured in a way AI can reason about and extend

With Supabase, those same concerns are split across SQL, RLS policies, app code, and dashboards.

Convex also comes with powerful pre-built components (Stripe, Resend, cron jobs, AI agents, etc.) that are easy to add to any project.

Supabase is great when humans are building software.

Convex is better when the “developer” is an LLM.

That difference shows up very quickly once AI touches your backend.

What does this actually mean for users?

We tried to answer this empirically.

So we ran the same prompt across all major vibe-coding platforms and asked them to build:

• a Slack clone

• a Notion clone

Same prompt. Same requirements.

We scored them on a rubric that covered the major functionality requested plus basics like auth, database, etc.

I expected Converge to do well, but I didn’t expect it to be so lopsided.

Here is the final score on the rubric from the Slack app clone.

1. Converge — 13/13

2. Replit — 5/13

3. v0 — 3/13

4. Bolt — 2/13

5. Lovable — 0/13

You can see the full breakdowns (with video) here:

• I tested every vibe coder by building a Slack clone

• I tested every vibe coder by building a Notion clone

Practically, this means:

• fewer broken auth flows

• fewer “why did my data disappear” bugs

• fewer infinite prompt loops trying to fix backend state

• fewer credits burned fighting infrastructure instead of building features

You spend more time building the product.

And less time debugging the AI.

What’s next?

We’re in the early stages of seeing who Converge resonates with and proving our hypothesis that Convex is the backend of the future for AI-generated code.

I’ll be building in public and sharing as much as I can, so if you’re interested in AI coding/AI app startups, feel free to follow along here on Substack, on LinkedIn and X.

If there’s anything you’re interested in knowing more about, feel free to comment or reach out!

P.S. PromptHub is still actively maintained and growing

P.P.S. If you want to try Converge, you can use code SUBSTACK50 for 50% off your first month.