AI-Generated Code Is Already Competing With Human Code — Daksh Gupta, Greptile
Source
AI Engineer
Author
AI Engineer
Date
Key takeaways · AI-distilled
Greptile co-founder Daksh Gupta says more than a quarter of the pull requests Greptile reviewed in April showed signs of being written largely or entirely by AI agents, up from under 1% a year earlier. The sample is over a million PRs a month.
Gupta compared AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → and human PRs on four measures: revert rate, revert rate by PR size, severity of flagged issues and review rounds needed to reach a mergeable PR. Agent code landed in the same range on all four, and humans were more likely to introduce P0 bugs.
Greptile identifies agent-written PRs from author fields, co-author footers and branch prefixes, according to the talk's methodology segment.
Gupta argues manual review cannot keep up: the median Greptile user makes fewer than 50 commits a month, the 99th percentile close to 1,000. He proposes validating whether a change breaks the user contract, makes a future break likelier, or misses the author's intent.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Why it matters
Shows agent-written PRs now match human code on revert rate and severity but fail differently by tool (Claude skews toward SQL injection, Cursor toward N+1 queries, Devin has fewer auth bypasses), reshaping what code review should check for.
videoAn AI Research Agent That Runs Your Experiments — Tim Sweeney, Weights & Biases
videoBeating RL With Reflection: GEPA and Optimize Anything — Lakshya A. Agrawal, GEPA
videoHow We Built an Agent That Improves Itself — Zubin Aysola, Weights & Biases
videoAutoresearch Made Our Models 3x Faster — Tejas Bhakta, Morph
