Agents, guardrails & review
Safe AI delivery
Using AI agents in real engineering work, in environments where mistakes are expensive: permissions, review and judgement.
What I believe
Safety in AI delivery is a permissions problem before it is an approval problem. Let the system decide what an agent can do, and spend human judgement where systems can’t. AI made code cheap; proving it’s right is now the job.
Start here
More articles
All articles →- If you know the steps, you don't need an agentAI agents are being added to processes that were never uncertain. A simple rule for where they belong: automate the predictable, use AI for the uncertain, keep people for the accountable. And the failure mode to design for when an agent says 'Fixed.'5 min read
- Moving my blog off WordPress with an AI agent: what I delegated, and what I didn'tI migrated this site from WordPress to a static site in one afternoon, working with an AI agent. The speed was impressive. The more useful lessons were about where the agent stopped and I had to decide.4 min read
From my reading
All notes →- AI made developers faster. Is their judgment keeping up?The hollow orchestrator: approving more work each month without getting better at judging it.Brian Houck, Scott Carey
- Speeding up code review with 'small-change'Median PR review time rose from about 33 to 77 minutes as AI raised the number of PRs.Will Larson
- The Manager's Path in the age of AIAI improves visibility, not understanding; talk to people more, not less.Camille Fournier
- Elephants, goldfish and design docs as the real artifactDesign the doc in one long AI session, then test it on a fresh one.Dave Rensin
- How AI has changed my engineering leadership roleLeads should stay hands-on with small, high-leverage safety fixes.Bjorn Roche
- LeadDev Engineering Leadership Report 2026: the player-coach is backLeaders are more hands-on, mostly because AI makes contributing easier.LeadDev