MARCH 25 UPDATE:
1) Progress: After the submission, I forked the codebase and kept building. I refined the system, added stricter constraints, and now spin up isolated virtual machines for each execution. The biggest improvement is that the system actually works RELIABLY end-to-end now.
At this point, I’m fully committed to turning this into a startup. I also applied to Y Combinator (Spring 2026), and they said the application looks promising but asked for more time to make a decision (should hear back within the next month).
2) Landscape: Honestly, when I first submitted this, I still had doubts — like, is this something people actually want? Is it even commercially viable? Fortunately, I’ve started getting external validation.
On March 23, Ramp came out with a blog post about how they built an agentic pipeline that can self-heal their codebase for RampSheet (https://x.com/ramplabs/status/2036169782764925106?s=46), and it’s currently blowing up on Twitter/X. This strengthens our belief that what I’m working on is the future.
That said, for that self-healing system to work, the Ramp team still needs to integrate external observability tools like Datadog for it to “maintain itself.” My pitch is a backend that both “builds” and “maintains” itself end-to-end. In fact, part of our roadmap on the “build” side is helping users choose their observability stack (Datadog, Sentry, etc.) and automatically implement it, which then directly enables the “maintenance” loop down the line.
Log in or sign up for Devpost to join the conversation.