The newest AI unicorn doesn't build models, agents, or coding assistants. It builds the thing that checks the code the agents wrote.
Blacksmith, a continuous integration platform for building, testing, merging, and validating code changes, announced Wednesday that it raised a $66 million Series C led by Sequoia, valuing the company at $1 billion. The round landed just two months after a $45 million Series B at $550 million — and a year ago Blacksmith was worth $60 million. It has now raised $125 million total, with annual recurring revenue up 17x and more than 90 million CI jobs run. Customers include Mercury and Granola.
Co-founder and CEO Aditya Jayaprakash framed the pitch plainly: "most companies are struggling to keep up with the volume of CI they need to run," especially as AI writes more of the world's software. "As agents write more of the world's software, the question that matters is whether it can be trusted."
That's the shift this round is really about. The bottleneck moved downstream. Writing code is cheap now — Anthropic already rebuilt its version control to survive 30,000 merges a day as agents flooded its repos. The expensive part is knowing whether any of it works, and not shipping something an agent hallucinated at 3am.
This is the pattern in AI infrastructure this year: the money flows to the layer where the failure would be most catastrophic. Anthropic can't watch its own agents, so it pulled their internet. OpenAI's employees allegedly exfiltrate device secrets. And now a CI company is worth a billion dollars because every enterprise CTO has the same quiet fear — that the pull requests landing in their repos were written by something that doesn't understand the codebase.
Blacksmith's 17x revenue growth in a year suggests the fear is already priced into budgets. The watch-next question: whether verification becomes a real platform or gets absorbed. GitHub, GitLab, and the cloud hyperscalers all have CI products, and none of them are standing still. Blacksmith's window is the trust gap — and it's only open as long as the code outruns the checkers.
— via TechCrunch