Cognition at $40B: what your coding agent really costs

A coding agent vendor just doubled its price tag in eleven weeks. Cognition, the company behind Devin, is reportedly in early talks to raise at a $40 billion valuation — up from the $25 billion pre-money it closed at in late May. Nothing about the product changed in that window. The revenue did. If you are budgeting for AI coding tools this year, that gap between the round and the run rate is the number to watch, because you are the one who funds it.

What actually happened

Bloomberg first reported the talks, with TechCrunch confirming the details. The mechanics:

• May 27, 2026: Cognition raised $1 billion at a $25 billion pre-money valuation, per TechCrunch
• At that round, the company reported a $492 million annualized revenue run rate, with enterprise usage growing roughly 50% month over month over the prior six months
• The new talks are reportedly tied to a run rate approaching $1 billion — about double the May figure
• Named customers include Mercedes-Benz, NASA, and Goldman Sachs

Founder Scott Wu has been consistent that Devin is not sold as a headcount replacement. The pitch is the boring work: legacy migrations, platform ports, repetitive refactors — the tickets nobody volunteers for. That is also why enterprise usage compounds monthly. Migration work is not a one-time purchase.

Why coding agent pricing matters for your business

A round is a bet on future pricing power. When a vendor's valuation runs at 40x its run rate, the investors are underwriting the assumption that seat prices go up, usage-based charges expand, or both. Devin's list pricing already moved once. It will move again — and small teams are the last in line for a negotiated rate.

Three things we do about this, and recommend you do too:

Price the work, not the seat. Track cost per merged PR and cost per completed migration, not cost per developer per month. Seat pricing hides the fact that agent spend scales with attempts. When your vendor changes its meter — and coding agents have been drifting toward usage billing all year — a per-outcome baseline is the only thing that tells you whether the new plan is worse.

Keep the agent outside your build. Devin, Claude Code, Copilot, and whatever ships next quarter all sit on top of a repo, a test suite, and a CI pipeline. Those are yours. If a coding agent's output only makes sense inside that vendor's cloud workspace, you bought lock-in, not leverage. We wrote about this when Cognition bought Poke and TierZero in four days — the vendors are buying the layers around the model precisely because that is where the switching cost lives.

Budget for the validation half. Agent-generated code still has to be reviewed and tested. The CI bill moves with it, which is why a CI cloud just raised at 10x in a year. Count both halves or your forecast is wrong by design.

Cognition's growth is real. So is the fact that somebody pays for a $40 billion valuation, and it is not the investors.

Sources: Bloomberg, TechCrunch.