Blacksmith Became a Unicorn Because AI Is Writing More Code Than Your Tests Can Hold

On 21 October 2026, Blacksmith raised a $66 million Series C led by Sequoia, at a $1 billion valuation. The company runs the build-and-test path teams use when code changes. The round is a comment on volume: agents are opening more changes than the old pipeline was built to check.

The numbers they published

CEO Aditya Jayaprakash said the company exists to make continuous integration less painful, and that AI-written code is pushing teams past the number of checks they can run. He framed the real question as whether that code can be trusted. Blacksmith said annual recurring revenue had grown 17 times and that it had run more than 90 million CI jobs, with customers including Mercury and Granola. The Series C landed two months after a $45 million Series B at a $550 million valuation. A year earlier the company was valued at $60 million. TechCrunch: https://techcrunch.com/2026/10/21/blacksmith-hits-unicorn-valuation-after-sequoia-leads-series-c/

Faster tests are not a substitute for the right tests

• Every agent change goes through the same gate as a human change. A green badge from the agent is not a review.
• Cover the path that loses money. Login, price, stock, payment, and permissions beat a large snapshot suite that never fails.
• Cap what an agent may merge. Docs and copy can be loose. Billing and auth stay with a person.
• Watch duration and flakes. A suite that is slow or randomly red will be skipped the week the agent gets busy.

If you are already letting a coding agent draft features, pair it with the release habits in the custom software development process. The 17 October note on Replit’s default model is the other half: the model will change, and the tests are what you keep.

FAQ

Do we need Blacksmith specifically?

You need a pipeline that finishes, fails for real reasons, and blocks a merge. Blacksmith is one vendor in a crowded category. The requirement is the gate, not the logo. This is not investment advice.

Can agents write the tests too?

They can draft them. A person still has to confirm the test would fail if the bug returned. An agent that writes the code and the only test is grading its own homework.