“This is my first job.” But you wouldn’t know it when you meet Daksh Gupta, co-founder and CEO of Greptile, who is part of the first generation of founders building companies in an AI-native world. Before GenAI, getting a real feel for a market took years in the seat and dozens of conversations with people who had seen it before. Today, founders have access to an extraordinary amount of accumulated knowledge almost instantly. But access to knowledge isn’t the same as knowing what to do with it. The very best founders stand out for their judgment and discernment: what to pay attention to, what to filter out, and which lessons from earlier companies actually apply to what they’re building. What impressed us about Daksh is that he does the homework. He studies the companies that came before him, seeks out leaders who have lived through similar inflection points, and then thinks hard about what those lessons mean for Greptile.
In one of our first conversations, Erica Ruliffson Schultz told Daksh a story from her New Relic days. When AWS launched its own monitoring service, people assumed it was the end of New Relic and Datadog. It wasn’t. Enterprises running on multiple clouds wanted one monitoring layer that no cloud provider controlled, and the independent vendors flourished. Daksh picked up the parallel before she finished: enterprises running multiple coding agents will want one independent validation layer across all of them, for the same reason. Then he added a second parallel she hadn’t expressly made. CloudWatch outages were correlated with AWS outages, just as a model’s blind spots in reviewing code are correlated with its blind spots in writing it. Greptile’s own data bears this out - the models most likely to introduce a given class of bug are also the least likely to catch it.
He tested the lesson against what he was seeing, found where it held, and made it his own.
Pair that kind of learning velocity with experienced operators who have lived those lessons first hand, and you get a founder who can move incredibly fast without mistaking speed for judgment.
Today, Greptile announced its $50M Series B led by Menlo Ventures, with participation from YC Continuity and Benchmark, and we’re thrilled Operator Collective is part of the round.
What does Greptile do?
Greptile builds AI agents that review and test code before it ships. Every time a developer, or a coding agent, submits a change, Greptile reads it against a map of the company's full codebase, along with the team's docs, tickets in tools like Linear and Jira, and the coding standards it has learned over time. Then it leaves comments directly on the lines that look wrong, flagging logic errors, performance problems, and security holes.
The company was founded in 2023 by Daksh, CTO Soohoon Choi, and Vaishant Kameswaran, three Georgia Tech computer science grads who met through the school's CREATE-X startup program before joining Y Combinator's Winter 2024 batch. Customers include NVIDIA, Coinbase, and Scale AI.
Why should you pay attention?
AI is writing a lot of code now. According to Greptile's own tracking, pull requests opened by coding agents went from under 1% of all pull requests to nearly 28% between February 2025 and April 2026. (A pull request is essentially a developer's way of saying, "here's my change, please check it before it goes live.")
That's a lot of new code, and every line still needs a second set of eyes before it reaches customers. Human reviewers can't keep up with agents that work around the clock.
There's also the question of who does the checking. Most engineering teams use more than one coding agent: Claude Code for one project, Cursor or Codex for another, Factory’s Droids or Devin for a third. Asking each tool to review its own work is a bit like letting students grade their own homework. Companies increasingly need one consistent, independent reviewer that checks everything, whichever human or agent wrote it. And because that choice gets made for the whole engineering org, it tends to stick once it's in place.
How does it actually work?
Greptile starts by building a map of how a codebase fits together: which pieces of code call which, which services depend on each other, and where problems have shown up before. That map updates itself as the code changes, so reviews stay current without anyone maintaining it by hand.
The product also learns the bug patterns specific to each customer. If a company's login service tends to break in the same few ways, for example, Greptile remembers that and gets sharper (and cheaper to run) on every review that follows. Across its customer base, Greptile has built up a dataset of more than 10 million labeled bugs across 2 million code repositories, with every new review adding to it.
In July, the company launched T-Rex. (Yes, a company with "reptile" in its name named its newest product after a dinosaur.) Reading code catches a lot, but some bugs only show up when the software actually runs. T-Rex spins up a copy of a customer's software in a sandbox (a sealed-off test environment), installs what it needs, writes and runs its own tests, clicks through the app like a real user would, and stress-tests the backend. In its first week in beta, T-Rex went from zero to testing 25,000 code changes a day.
Next up is extending T-Rex to handle larger and more complex enterprise setups, opening it up to many more of Greptile's existing customers.
Why are we invested?
Daksh, Soohoon, and Vaishant built something engineering teams go out of their way to adopt. Greptile charges more than many alternatives, and yet engineering teams at companies like Brex, Mintlify, and Reducto have switched to it from cheaper or homegrown tools and stayed. [Optional growth/retention stat, pending company approval, e.g. ARR growth or net revenue retention.]
More importantly, the team has a clear view of where this can go. The observability parallel Daksh drew with Erica is not just an analogy to him, it informs the roadmap. Greptile wants to be the independent layer engineering teams rely on no matter which coding agent wrote the code, starting with review and testing and expanding into the infrastructure companies need to ship agent-written software safely, from code hosting to security scanning.
Which is the other reason we invested. Daksh doesn’t confuse being AI-native with already knowing the answers. After one 30-minute meeting with Lynn Carter, our talent and growth partner, he described it as some of the densest learning he’d had on a problem he was actively working through. That is the pattern: he seeks out people who have built and scaled companies before, asks specific questions, absorbs what’s useful, and applies it to a market that is changing incredibly quickly.
Greptile is also a natural fit for our Collective Venture Model™. Many of our 250+ Operator LPs lead engineering, product, and security teams that are bringing coding agents into their organizations right now. They're the people deciding what gets shipped and how it gets checked. And they've been there, done that when it comes to many of the challenges that companies face as they scale.
What's next?
Greptile is putting this funding toward hiring across engineering and go-to-market, and toward extending T-Rex to more complex enterprise environments. If your team is shipping AI-written code and wants a consistent check on all of it, visit greptile.com to get started, or reach out to our team for an introduction.
