The useful thing about a bottleneck is that it tells you where to look. Harjot Gill has been looking for nearly two decades. At the University of Pennsylvania, he studied the unseen traffic of distributed systems. At Netsil, he helped companies understand the tangle that arrived after cloud software became easy to deploy. At CodeRabbit, he is now staring at another queue: code can be generated in minutes, but a human team still has to understand it, challenge it and decide whether it belongs.
Gill’s career reads less like a sequence of fashionable pivots than a long argument with software’s consequences. Make deployment simple and observability becomes difficult. Make code generation simple and review becomes difficult. The work never vanishes. It moves to the next desk, usually late on a Friday afternoon.
His phrase for this is compact: “History rhymes.” During the cloud transition, deployment accelerated and the mess drifted downstream into monitoring and management. In the current AI transition, production has accelerated again. This time the mess is accumulating around validation, explanation and trust.
Research with commercial consequences
Gill arrived at Penn after studying electronics and electrical communication at Punjab Engineering College. His graduate work sat at the intersection of computer networks, streaming analytics, parallel systems and declarative languages. It produced papers with titles that make excellent tests of a dinner companion’s patience, including work on high-performance network traffic analytics and application-aware anonymity.
The important detail is not the density of the titles. It is that the research escaped the lab. While working as a teaching assistant, Gill met a future co-founder. The pair spun Netsil out of academic work and aimed it at the confusion of cloud applications, Docker and Kubernetes. Gill became co-founder and CEO. In 2014, he was the principal investigator on a $180,000 National Science Foundation small-business grant to commercialize the research.
Netsil reconstructed and mapped activity across complex services, giving operators a way to see systems that had become too distributed for old instruments. Nutanix acquired the company in March 2018. Gill joined Nutanix, eventually serving as senior director of technology. The transaction gave him an exit, but the more interesting inheritance was a pattern: a technical wave makes creation cheaper, then sends the bill to comprehension.
Netsil commercializes systems research from Penn.
Nutanix acquires Netsil; Gill moves inside a larger cloud software company.
Gill co-founds FluxNinja for cloud-native reliability management.
An internal AI reviewer becomes CodeRabbit.
CodeRabbit raises a $143 million Series C and broadens its brief.
The rabbit appears inside another company
Gill’s next startup, FluxNinja, began in 2021 around reliability for cloud-native applications. Then GPT-3.5 and GitHub Copilot arrived. He was managing a remote team spread across Eastern Europe, India and the United States, where code review was already a constraint. If machines were about to increase the supply of code, the review queue would not politely remain the same length.
The team built an internal AI reviewer. It was practical, close to the workflow and designed to act without waiting for someone to open a chat window. When a pull request appeared, the system could inspect it in the background and return comments where developers already worked. That internal utility found an audience outside the company and became CodeRabbit.
The early geographic signal was wonderfully inconvenient for anyone with a neat launch plan. CodeRabbit went viral in Japan before it did elsewhere. Gill has said the organic response arrived without a conventional crop of design partners. Strangers were trying it, which is a more persuasive form of market research than friends nodding over a pitch deck. He followed the pull.
How a review thinks
CodeRabbit’s work begins when a pull request opens. It clones the repository, builds a live code graph and runs static analysis. Smaller models assemble context and prune irrelevant material. More capable reasoning models then inspect the condensed evidence for bugs and risky changes. The division of labor is economic as well as technical: expensive intelligence should not spend its afternoon reading boilerplate.
The review loop
opens
static analysis
decides
Gill argues that human review remains consequential. Senior engineers should spend less time catching syntax mistakes and more time on architecture, planning and guardrails. His one-line summary has the snap of a teacher catching a suspicious exam paper: “A coding agent cannot grade its own homework.” Independence matters when the author is prolific, tireless and occasionally wrong with great confidence.
This view also explains his hiring rubric. Gill says CodeRabbit looks for systems-design knowledge, passion and “good taste.” Taste sounds soft beside a compiler, but it is brutally practical. It is the ability to choose the cleaner architecture, recognize needless complexity, ask the productive question and reject code that works while making the system worse.
Open source as gift, proof and laboratory
CodeRabbit made its reviews free for public repositories. Gill frames that decision partly as a return to maintainers, whose work is essential and often unpaid. It also created an unusually aligned distribution system. Corporate developers could meet CodeRabbit while contributing to software they already used. The product demonstrated itself in public. Every interaction supplied feedback about how real teams expressed standards and evaluated suggestions.
After its 2025 Series B, CodeRabbit pledged $1 million in cash to open-source maintainers and projects, later reporting more than $1.2 million delivered. Following the 2026 Series C, the company announced a commitment of more than $10 million over twelve months, counting cash and the at-cost expense of keeping its review and agent capabilities free for public projects. There is commerce in the loop, certainly, but also symmetry: the ecosystem that proves the product receives something useful before a sales conversation begins.
The company grows past the pull request
The financing numbers became progressively less rabbit-sized. CodeRabbit raised a $16 million Series A in 2024, a $60 million Series B in 2025, and a $143 million Series C at a $1.5 billion valuation in August 2026. The latest round was co-led by Atomico and Smash Capital. At the time of the round, Gill said the company had grown alongside a three-to-fivefold increase in code generated by developers and that adjacent bottlenecks in delivery and collaboration were coming into view.
The product brief is widening accordingly. CodeRabbit now describes an idea it calls Agentic Change Management: a control layer for software changes made by people and machines. Gill has discussed extending the system across Slack, Jira and Linear, where engineering intent, planning, incidents and the bits of institutional memory commonly known as “whatever Priya remembers” tend to live.
Inside the company, that future has already altered how people work. Gill describes small cross-functional groups called war rooms, mixing product managers, designers, developers and developer relations. As AI handles more low-level implementation, people outside traditional engineering lanes can prototype closer to the code. Fewer hand-offs are required. The name is dramatic; the underlying move is domestic. Put everyone responsible for the decision close enough to argue about it.
Gill’s public descriptions of the engineering team also reveal a useful tension. He is willing to imagine familiar tools disappearing, comparing traditional code editors to steering wheels in self-driving cars, yet he keeps returning authority to people. In his account, senior engineers become directors. They establish the destination, frame hypotheses and trace failures across a system. Automation changes their altitude more than their obligation. The keyboard may recede; accountability does not.
That emphasis on collective memory matters because software failures are rarely caused by syntax alone. A change can compile, pass tests and still violate a decision made six months earlier in a meeting that nobody documented. Gill wants CodeRabbit to accumulate instructions and interactions at the team level, so conventions survive beyond one developer’s local chat. The ambition is as organizational as it is computational: make a codebase’s history available at the moment somebody proposes its future.
There is also a founder’s restraint buried inside the technical machinery. CodeRabbit does not offer a model picker simply because model names attract attention. The company routes work among an ensemble, using smaller models for context and reserving expensive reasoning for the final inspection. Customers buy the result instead of being asked to manage the kitchen. It is an operator’s decision, and a reminder that a product can use a crowded technology without turning its interface into a catalogue of the crowd.
That sentence supplies the continuity in Gill’s work. Netsil helped teams understand what had already happened across distributed software. CodeRabbit helps them understand what is about to happen if a change merges. One looks at the living system, the other at its proposed future. Both are tools for recovering context after technology has made output cheap.
The useful lesson is portable. When a new tool promises to remove labor, inspect the next room. Someone there may be receiving a larger pile of work. Gill’s career has been an exercise in finding that person, drawing the system around them and building before the rest of the market notices the queue.