The Evolution of Divot: How Agents Earn Their Autonomy
summary
Divot automates three concrete workflows and shows measurable wins: a feature-flag pipeline that cleans stale flags after 15 days by dispatching coding agents that open up to three PRs at a time, cutting active flags to a sustainable baseline; a PR reviewer that scores changes 1–5 on a risk/complexity rubric (built to handle ~30% low-risk “rubber stamp” PRs) and raises repo thresholds as CI and tests improve; and a Slack assistant that throttles autonomy by gating tool calls until approval rates justify automation.
AI-written summary · about
chapters (8)
- 0:00AI project failures and Divot
- 5:02Why AI solutions fail trust
- 6:51Divot as code maintainer
- 12:10Software factories maintenance
- 15:20Divot as PR reviewer
- 20:15Divot in Slack productivity
- 23:45Agentic solutions takeaways
- 24:27Q&A
related talks
17mSoftware Development Agents: What Works and What Doesn't
30mBuilding Tools with Tools: Modern AI-Driven Development
36mAgentic CI/CD: from Pull Request to Production without the Paper Cuts
51mFuture of Software Development
57mRise of the Machines - Automate Your Development
52mAre Your Engineering Teams Really Prepared for the GenAI Revolution?!
42mRevisit Old Problems with New Eyes!
46mRecalibrate - How AI shakes up software team