Skip to content

By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

The Pragmatic EngineerThe Pragmatic Engineer

Software factories and code reviews with Harjot Gill

In this conversation, we step into the world of an AI-pilled CEO to explore how their startup works.

Harjot GillguestGergely Oroszhost
Nov 18, 20261h 27mWatch on YouTube

Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.

Great conversations run long. Their best ideas don't have to.

uListen is free on iPhone, iPad, Android, Chrome, and Claude.

See how

At a glance

WHAT IT’S REALLY ABOUT

How CodeRabbit uses AI agents to build a software factory

  1. Harjot Gill explains how he and CodeRabbit use AI agents to dramatically increase shipping velocity, including merging over 5,000 pull requests in a month by keeping multiple agents running in parallel.
  2. He defines “software factories” as systems where agents continuously hill-climb toward goals using layered validation loops (tests, simulations, emulators, cloud environments, and production feedback) rather than relying on manual development workflows.
  3. He argues human code review (reading diffs) is effectively dead at agentic scale, but human review remains essential for deciding what to ship and for assessing risk at the release level.
  4. He describes practical quality controls at CodeRabbit, including frequent releases with risk indicators, rollback readiness, continuous “deslopping” of tests, and a “driving license” model where developers lose direct-to-prod privileges if they introduce regressions.
  5. He discusses the operational and economic implications of agentic development—large token and CI spend, compute constraints, constant model eval work, and hiring that prioritizes agency, systems thinking, and collaborative “war room” execution.

IDEAS WORTH REMEMBERING

5 ideas

Human code review is “dead,” but human judgment shifts to release/risk review.

Harjot Gill says traditional “human code review” (reading diffs line-by-line) no longer scales once AI and agents dramatically increase change volume. He argues humans should still review at a higher level—deciding whether to ship, assessing risk/blast radius, and monitoring releases rather than individual pull requests.

A “software factory” is agent fleets running continuous validation and improvement loops across the SDLC.

Harjot describes two modes: interactive “crafting” (like shaping UI via chat) and a more hands-off “factory” mode where agents hill-climb toward goals using validation loops across local checks, emulators, real cloud environments, and production feedback. He frames frequent releases as “turns” in a strategy game where agents learn from each iteration.

With validation, higher AI-driven velocity can coincide with fewer defects.

He claims defects are decreasing as velocity rises, arguing that with strong validation layers, AI can produce less “slop” than humans. He emphasizes that quality depends on agents being able to verify work against tests, simulations, telemetry, and production signals.

Quality control becomes a trust-and-risk system, not a per-PR reading process.

Harjot describes internal mechanisms to manage risk at high shipping rates: risk labeling per PR, human review focused on releases, frequent rollbacks, and a “driving license” system where developers lose autonomy to ship if their changes cause long-lived regressions. He also notes some developers are diligent in steering agents while others are reckless, requiring intervention.

Agentic development makes CI/test strategy and flake reduction a first-class, ongoing engineering problem.

He describes “deslopping tests” as continuously removing redundant or low-value tests and shifting toward more useful integration/E2E coverage to keep CI times low. He says CI and compute costs explode with agentic development, leading them to build their own CI and to run agents whose job is to reduce flakiness and maintain the SDLC.

WORDS WORTH SAVING

5 quotes

So it's kind of funny, like the AI is driving the car while, like, the other AI is driving the company.

— Harjot Gill

So it's like, like everything is code at, at, inside CodeRabbit at this point.

— Harjot Gill

You're right. I mean, the human code review is dead, but the human review is not dead.

— Harjot Gill

Because the human taste is the bottleneck now.

— Harjot Gill

And once you have like enough strikes, we take your driving license away.

— Harjot Gill

Read the available transcript
5,000 PRs in a month via agentsCodex desktop app vs Claude Code workflowsSoftware factories and hill-climbing loopsRelease-based review and risk managementDriving license model for shipping to prodCI cost explosion and test “deslopping”Evals, model routing, and harness tuning

Ask ChatGPT about this episode

Get its key ideas, plus a summary of each chapter linked to that point in the video, right in your chat.

  1. Add uListen AI to ChatGPT.

  2. Paste this prompt into a new chat.

    Use uListen AI to give me this episode's key ideas and a chapter-by-chapter summary, with timestamps: https://www.youtube.com/watch?v=4Cjb7691GJ8

SPEAKERS

  • Harjot Gill

    guest

    CEO/founder of CodeRabbit; longtime software engineer and repeat startup founder with a PhD from the University of Pennsylvania.

  • Gergely Orosz

    host

    Software engineer and writer behind The Pragmatic Engineer, hosting interviews about engineering practices and industry trends.

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.