Humans on the loop, not in it: Taking agentic engineering end to end
summary
Claim: put humans on the loop, not in it by letting agents build and evolve software end-to-end while enforcing quality through CI/CD. Experiments use Claude, cloud code and AWS CLI where agents run the full CD pipeline to deploy, run journey tests, lint, mutation/fuzz tests and report DORA-style metrics. Agents repeatedly break infrastructure, loop on fixes (costly and noisy), sometimes change env instead of code, and often fail on deploys. Treating the pipeline as a pipeline-as-operability-harness surfaces operability gaps and forces repeatable sensors and guides.
AI-written summary · about
chapters (9)
- 0:00Humans on the loop intro
- 3:24Touching code defeats agents
- 9:23Humans in loop to pipelines
- 14:16Good software for agents
- 19:31Measuring deliverability and operability
- 23:45Agents vs DevOps delivery
- 26:33Pipelines as operability harnesses
- 30:48Agent deployment challenges
- 31:30Summary and future
related talks
33mHumans on the loop, not in it: Taking agentic engineering end to end
21mPlatforms for Humans and Machines: Engineering for the Age of Agents
34mThe Human-in-the-Loop is Tired
35mThe Agentic AI Engineer
46mWhen You Run Out of Requirements: What Happens When You Go All-In on Agents
40mHow To Tame Your Agents?
30mArchitecting Integrity: Responsible Software Evolution in the AI Era
25mHuman in the Loop: Governing API Evolution in the Age of Agents