It's 9:47am at a startup you've heard of. A junior engineer opens a pull request. Two seconds later, the rabbit reads it.
The rabbit leaves seventeen comments. Some are nitpicks about variable naming. Two are real bugs. One is a security issue the engineer would have shipped to production by lunch. By the time anyone with a Slack avatar shows up to look at the PR, the worst of it is already flagged, summarized, and rendered into a tidy sequence diagram. The human reviewer skims, approves, moves on. Coffee is still warm.
That rabbit is CodeRabbit, a two-and-a-half-year-old company in San Francisco that has quietly become the safety net under the great AI coding boom. More than 10,000 organizations now hand their pull requests to it before any human looks at them. Mercury runs banking infrastructure through it. Chegg sends homework through it. Groupon, Crunchbase, and Harvey send theirs. The list is long, and growing about 20 percent a month.
Code review was the slowest part of shipping. Then AI made code cheap, and review the bottleneck.
For most of software's history, writing code was the hard part and reviewing it was the chore. Senior engineers complained about it. Junior engineers waited days for it. Compliance teams asked for evidence of it. Everyone agreed it mattered, and almost no one liked doing it.
Then GitHub Copilot shipped. Then Cursor. Then a half-dozen agentic coding tools that could write a thousand lines before lunch. Suddenly the code was free. The reading of it was not.
The math is uncomfortable. A solo developer with Claude and Cursor can produce a workweek of code in an afternoon. A reviewer cannot read a workweek of code in an afternoon. The bottleneck moved.
The dirty secret of the AI-coding boom is that most teams' review process did not get faster. It got more behind.
Two founders, one previous exit, and a hunch that the rabbit could keep up.
Harjot Gill had been here before. His first company, Netsil, an observability startup, sold to Nutanix in 2018. After that he co-founded FluxNinja and watched, from the front row, as AI tooling started rewriting how engineering teams worked. In early 2023, with co-founder Gur Singh, who had led developer teams at the healthcare payments company Alegeus, he bet on the unglamorous half of that rewrite. Not writing more code faster. Reading it faster.
The first version was, in Gill's own telling, a scrappy GitHub app that wrote markdown summaries on pull requests. It went live on the GitHub Marketplace and word spread the way it does on developer Twitter: quietly, then suddenly.
Within months CRV and Engineering Capital led a seed round. By August 2024, CRV led a $16M Series A. In September 2025, Scale Venture Partners led a $60M Series B with NVIDIA's NVentures joining, valuing the company at $550M. Total raised: about $79.6M. Total time elapsed: roughly two years.
What the rabbit actually does
CodeRabbit installs as an app on GitHub, GitLab, Azure DevOps, or Bitbucket. When a pull request opens, it does what a thoughtful senior engineer would do if a thoughtful senior engineer had unlimited time and no caffeine ceiling.
It summarizes
Every PR gets a plain-English walkthrough of what changed and why it might matter. For long PRs it generates sequence diagrams. For migrations it traces dependency impact.
It comments line by line
Not just bugs. Naming, anti-patterns, missing tests, security smells, performance regressions, edge cases the author missed at 11pm on Thursday.
It learns
You can teach it your team's coding standards in plain English. It remembers them. Next PR, it applies them.
It runs everywhere
There's a CLI for the terminal, an IDE extension for VS Code and Cursor that catches issues before the PR is even opened, and a self-hosted enterprise tier for the kind of customers whose general counsel does not love SaaS.
From a GitHub app to a $550M valuation in 30-ish months
The CodeRabbit Timeline
Cumulative funding, in dollars the rabbit can count
CodeRabbit funding by round (USD millions)
Improve code quality. Cut the wait. Keep humans in the loop.
Gill is careful, in interviews, not to claim the rabbit replaces the reviewer. He says it removes the parts of review that nobody wanted to do anyway: the first pass, the formatting, the "did you forget a test" reminder. What's left is the part humans are good at - judgment, architecture, taste.
The company also leans hard on the boring parts of being a vendor. SOC 2. GDPR. Ephemeral data storage. A self-hosted enterprise tier for banks and defense contractors. Most AI startups would rather not talk about compliance. CodeRabbit puts it on the landing page.
Why the rabbit matters next year, and the year after
If you believe the next five years of software will be written largely by AI agents - and the venture market clearly does - then the choke point is no longer typing. It's trusting. Every line a model writes has to be inspected by something before it ships, or the production incidents pile up faster than Friday afternoon postmortems can be written.
CodeRabbit is betting that the inspector will also, increasingly, be an AI. Not the same AI that wrote the code, because that defeats the purpose, but a different one, tuned for skepticism, with your team's standards loaded into its memory and a habit of asking awkward questions.
It is, in a sense, an AI hired to argue with another AI. Which is funny, until you remember that's roughly what code review already was.
It's 9:47am at a startup you've heard of. The pull request is already merged.
The junior engineer who opened it never had to wait for a senior to wake up in another time zone. The senior who eventually approved it spent four minutes on the review instead of forty. The bug the rabbit caught at line 322 never reached production, which is to say no one will ever write a blog post about it, which is to say the system worked exactly as designed.
Somewhere in San Francisco, the rabbit is reading the next one.