From 126 review comments to zero: letting agents auto-merge PRs
calabi_and_yau · x · 2026-10-27
Martin Gratzer reflects on a year of shifting his coding-agent workflow from human-reviewed PRs to letting his agent auto-merge.
- In January he left 126 inline comments on 25 PRs; by Sep–Oct, zero comments across 159 PRs; his newest project has just 3 comments total.
- The agent still errs, but fresh-context reviewers, tests, and CI now catch mistakes before merge, so reading diffs stopped paying off.
- Work is still broken into vertical-slice Issues that ship independently or behind feature flags.
- The human moves to both ends: shaping the issue up front (sparing with the agent) and shaping the system afterward — turning observed drift into guardrails via the repo and Forge skills.
- The pipeline is ship → guard → land: low-risk changes auto-merge, others get a risk verdict and targeted review.
He insists he's still in the loop and in control — just no longer the middleman reading diffs.
More from coding & agent
- xAI releases free Grok Bot course: 18 real-team playbooks, ~3 hours, covering sales to engineering — victor_explore · 2026-10-27
- 100+ commits per day is the new bare minimum in the AI coding era — damnGruz · 2026-10-27
- Rocket launches open app platform to help vibe coders distribute AI apps — alexmacgregor__ · 2026-10-27
- Turn product photos into interactive 3D-style displays with AI video frames, no 3D modeling — aigclink · 2026-10-27
- Full walkthrough: interactive product displays from photos using GPT-6 Astra and AI video as UI — MaziyarPanahi · 2026-10-27
- DeepSeek leads four Qwen3.5-9B models to reverse engineer binaries — team mode costs more than solo — HauntingMoment · 2026-10-27