Greptile raises $50M Series B led by Menlo to automate code validation as agentic coding spend explodes

Greptile raises $50M Series B to validate AI-generated code

Greptile has closed a $50M Series B led by Menlo Ventures. The company's pitch is straightforward: as coding agents generate more code than engineers can manually review, someone needs to sit between the agent and the merge button. Greptile wants to be that layer.

Daksh Gupta founded the company on a contrarian read of where agentic coding was headed. The prevailing assumption a year ago was that the frontier labs would own everything, including validation. Gupta bet instead that enterprises using multiple coding agents simultaneously would want an independent validator, one that wasn't built and sold by the same company generating the code. The separation of concerns mattered.

Nvidia validated the thesis early, becoming Greptile's first large enterprise customer. Nvidia told Gupta it was buying coding agents from essentially every meaningful vendor, all of which offered some version of a review or testing product, but it didn't trust those companies to also approve code they had written. That conflict-of-interest concern, which Gupta initially expected to resonate only with a narrow slice of companies, has turned out to be nearly universal across enterprise buyers.

“We are trying to automate code validation, which to us means that we want to make it so that you can review and test any code change and for sure know it is safe to merge. Today, announced our series b. We raised $50,000,000 from a variety of investors. Menlo led the round... We heard this first from NVIDIA, which became our first large customer. They said we buy coding agents from every good coding agent company that there is... but we wanted our validation to be separate.”

The volume problem

The urgency behind the product is arithmetic. Gupta describes companies spending as much on coding agent tokens as they pay individual engineers, and doing so with almost no scrutiny. Hiring a person for $250,000 a year gets careful deliberation; adding $20,000 a month in token spend gets rubber-stamped. He expects that tolerance to evaporate quickly as CFOs start asking what the productivity gains are actually producing.

His framing on quality is specific. Users don't measure bugs per unit of code change. They measure bugs per unit of time. If agents push code volume up 10 or 20 times, the rate of user-facing failures rises proportionally, even if the per-change error rate is the same. Human tolerance for faulty software doesn't scale with deployment velocity.

A gap that's hard to explain

Gupta flags something he describes as genuinely confusing: despite widespread claims of 10x individual productivity gains from coding agents, large companies don't seem materially different from several years ago in terms of software quality or feature breadth. More startups are shipping faster, and smaller teams are doing more, but incumbent enterprises, judged by what their products actually do, appear roughly unchanged.

His working theory is that productivity gains are being absorbed into a wider surface area of projects rather than compounding into measurable output at any single product. Better observability, and eventually the kind of spend-grading Greptile is building toward, may be the only way to reconcile the gap between what companies say AI is doing for them and what it's actually producing downstream.