Keystone

Keystone

Keystone is a Y Combinator W25 startup building an AI agent that autonomously fixes bugs in production code. Founded by Pablo Hansen, the company has raised a $5.2M seed round and reportedly turned down a seven-figure acquisition offer to keep building. Its customer list already includes Lovable, one of the most prominent AI app builders on the market — a meaningful signal that the product holds up under the pressure of real, shipped code.

What it does

Keystone positions itself as the seed of an "autonomous engineering organization." Instead of an inline coding assistant that helps a human developer at their desk, Keystone is an agent that takes ownership of bug-fix work end to end: it reads error reports and stack traces, navigates a codebase, reproduces the issue, writes a fix, runs tests, and opens a pull request. The pitch is that bug triage and resolution — work that is repetitive, time-consuming, and generally hated by engineering teams — is exactly the kind of work an agent should be doing while humans focus on architecture and product.

Because the agent operates on production code, the bar for accuracy is higher than in a typical AI coding tool. The team's emphasis is on agents that understand the surrounding system, not just the diff in front of them.

Why it matters

Keystone is part of a broader push at the frontier of AI software engineering: moving past the "copilot" framing — where AI suggests, and humans accept or reject — into a genuinely autonomous mode where AI agents own units of work. Its customer base, which includes other AI-coding companies, suggests the agents are credible enough to trust in the loop of teams that themselves understand AI coding deeply.

Funding and traction

Coming out of YC's W25 batch, Keystone was highlighted by TechCrunch among the most sought-after startups at the demo day where it was first publicly showcased. The $5.2M seed round, the rejected acquisition offer, and the early customer list put it firmly in the conversation about which AI coding agents will define the next wave of developer infrastructure. By definition, what Keystone ships is AI-generated code — for its users, that code is the product.