- CodeRabbit's CEO Harjot Gill argued on The Pragmatic Engineer that when agents write the code, deciding what to build is the scarce skill.
- He said he reads none of his agents' Rust code and now screens hires for agency, range and strategy-game experience.
Harjot Gill, who founded CodeRabbit and runs it as CEO, argued that when agents write the code, the scarce skill is deciding what to build. CodeRabbit sells AI code review. Gill made the case on The Pragmatic Engineer, in an episode published on 18 November 2026. He said his hiring tests already reflect it.
What was said
Gill described a way of working where people no longer read the software. His agents, meaning AI programs that write and change code on their own, produce Rust, a programming language. He said "I have zero idea how Rust works" and "I have not read a single line of it."
So what is left for humans? Gill said "the human taste is the bottleneck now, human judgment." His example was product design. An interface has no single right answer, he said. It can be built in ten different ways, and choosing among them is where taste comes in.
He applies this to hiring. The host read out a post Gill had published. It set two bars: tastefully shipping at least 1,000 pull requests (proposed code changes) in 30 days, or 500 hours or more on simulation games such as Civilization. Gill compared steering agents to those games, where each turn is a release.
Agency came first when he interviews, he said. A candidate with no side projects is a red flag. Range came second. Agents can go deep, but a person with broad experience can point them at many kinds of systems.
Why it matters
Our reading: if nobody reads the code, the check moves from how it was written to what it is for. Interviews that reward coding speed then test the part agents already do. Harder to fake is choosing among several reasonable designs and explaining why. Managers could ask candidates what they built unprompted, and what they chose not to build.
Training changes too. Gill said he would take either a senior engineer willing to unlearn old habits, or a fresh graduate with high agency who learns fast. Both choices bet on judgment and learning speed, not on knowing a language such as Rust.
The other side
This is one founder's view, and his company sells AI tools. Gill himself warned that shipping at this volume breaks things. The excerpts do not say what a higher-level human review involves, or how someone who has read none of the code would spot a flaw in it.
The host said he stayed skeptical on several points. Gill offered the strategy-game test as his own take, with no evidence that it predicts good work. The host also said he could not recommend Gill's habit of waking at night to prompt agents.
Written by the WebPulse Newsroom with AI assistance, and checked by our editorial review: every quotation was verified against the recording's transcript. How we use AI.
The conversation this talking point comes from
- The Pragmatic Engineer: Software factories and code reviews with Harjot Gill (2026-11-18)





