Your agent wrote it. Who’s on call?
Presented by: Dane Knecht, Madison Faulkner, Aaron Jacobson, Chris De Vylder
Originally aired on October 23 @ 2:00 PM - 2:30 PM EDT
Cloudflare shipped an OAuth library for Cloudflare Workers largely written by AI, publishing prompts alongside code for the industry to review commit by commit. This provided an early, honest look at what agent-authored production code takes to get right: a lot of review. As generation stops being the bottleneck, the constraint moves downstream — to review, CI, and production, where failures land on on-call engineers who didn’t write the code. This panel examines the data: Factory brings throughput numbers from agentic coding teams; Sentry brings failure profiles — error classes, revert rates, triage time. Is generated code’s distinct failure pattern a generation problem or a review problem?
English
