Skip to content

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

YC Root AccessYC Root Access

Greptile: AI That Actually Understands Codebases

Greptile is building AI that understands entire codebases. Each week, Greptile reviews 5–8 million lines of code for customers and surfaces tens of thousands of bugs that would otherwise slip through human review.

Brad FlorahostDaksh Guptaguest
Sep 22, 20256mWatch on YouTube ↗

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

At a glance

WHAT IT’S REALLY ABOUT

Greptile uses AI to review codebases, catching bugs faster

  1. Greptile automates large portions of pull request review to save senior engineers time and catch more defects than human-only processes.
  2. The product is positioned as “humans + AI,” where AI handles tedious mistake-finding while humans focus on mentorship, architectural direction, and long-term codebase evolution.
  3. Greptile reports reviewing 5–8 million lines of code weekly and surfacing roughly 20,000 bugs per week across customer PRs.
  4. The company’s “why now” hinges on LLMs being especially strong at reading code, and on rising societal dependence on resilient software systems.
  5. The team evolved from codebase Q&A (understanding code) to bug detection (persistent value), after experiencing firsthand how hard large unfamiliar codebases are to navigate.

IDEAS WORTH REMEMBERING

5 ideas

Use AI to compress review time, not to remove reviewers.

Greptile’s value proposition is speeding up and improving code review quality, while keeping humans responsible for judgment calls, mentorship, and design direction.

Code review is more than bug-finding—retain the human “bigger picture.”

Daksh frames reviews as an opportunity for mentorship and forward-looking architecture, areas where today’s AI lacks the “clairvoyance” of senior engineers.

LLMs’ near-term advantage is code reading, not code generation.

Greptile bets on models’ strongest current capability—understanding and analyzing code—rather than relying on them to author large amounts of new code correctly.

If you can teach a model the codebase, bug detection becomes the highest-leverage use.

The team discovered that codebase understanding is a means to an end: preventing defects before changes land, which creates ongoing value beyond onboarding or Q&A.

Large-codebase comprehension solves a pain that teams repeatedly feel.

They started from a real scenario—losing the developer who wrote the frontend—highlighting how knowledge gaps and scale make engineering hard even for capable teams.

WORDS WORTH SAVING

5 quotes

"We're building AI that understands large codebases."

— Daksh Gupta

"Code reviews... serve a purpose greater than just detecting bugs. It's a opportunity for mentorship."

— Daksh Gupta

"This past week, we revealed twenty thousand bugs across the pull requests."

— Daksh Gupta

"They're... better at reading code than they are at writing it."

— Daksh Gupta

"You just really should focus on building products that your customers love."

— Daksh Gupta

Read the available transcript
AI-assisted pull request reviewHuman-in-the-loop code reviewBug detection at scale (20k/week)LLMs stronger at reading than writing codeCodebase Q&A and context buildingDeveloper adoption skepticism vs enthusiasmFounder focus and avoiding distractions

SPEAKERS

  • Brad Flora

    host

    Host of YC Root Access and interviewer for the episode.

  • Daksh Gupta

    guest

    Founder of Greptile (YC Winter 2024), building AI to understand and review large codebases.

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.

iOSAndroidClaudeChatGPTChrome