Hey everyone,
I’ve been building with Cursor for a while now and wanted to share what’s come out of it — mostly because I think this community appreciates seeing real projects instead of another “look at my todo app.”
The main one: a coding agent system inspired by Devin’s autonomy, but with a key difference — every major action requires human approval before it executes. No blind full-autonomous runs. On top of that, there’s a separate, dedicated QA agent that reviews the coding agent’s output before it’s considered “done.” Think of it as a built-in second pair of eyes that never gets tired.
Alongside that, I’ve been building a bigger local-first AI platform: model selection across BYOK and local models, an agent runtime, a Cursor-style IDE mode, and a few other pieces I’m still gluing together.
Cursor’s been the actual workhorse behind most of this — planning it out, catching my dumb mistakes, and letting me iterate fast enough to not lose motivation halfway through (which, let’s be honest, kills most side projects).
Planning to open-source the agent system soon. Would genuinely love feedback from people here who’ve built agentic tools before — what would you want to see in an approval-gated system like this?