My Agent Shipped 3 PRs in an Evening. 40% of My Messages Were Corrections.

Short summary

A developer ran an AI agent session that shipped three PRs (3,500 lines) to a SharePoint web parts repo in 75 minutes of active time, all passing automated validation untouched. However, 40% of the developer's messages were corrections, clustered during pipeline setup and warning fixes. The key insight: failures were context retrieval problems, not reasoning failures — smarter models won't help if constraints don't surface at decision time.

• •Agent session shipped 3 PRs with 3,500 lines of code in 75 minutes of active time
• •40% of human messages were corrections, mostly during pipeline setup and warning fixes
• •Failures were context retrieval issues, not reasoning failures — fix is pinned constraints and checklists, not smarter models

Generated with AI, which can make mistakes.