Stop prompting. Start engineering loops.
Invited by Accenture as an expert on applied AI and agentic systems, in front of an enterprise audience. The topic that evening: loop engineering, meaning AI agents that work in loops and check themselves. The thesis: the agent does not grade its own work. It writes the pull request, a second agent with a fresh context and read-only access scores it against five criteria. Same model, same prompts. The grader has nothing to defend.
Stop prompting. Start engineering loops. That was my talk in Mannheim. It was the Claude Code Meetup, hosted at Accenture, with three talks on the evening. One piece of my talk. The agent doesn't grade its own work. It writes a PR, then a second agent with a fresh context and read-only access scores it against five criteria. Same model, same prompt quality. The difference is that the grader never saw the work happen, so it has nothing to defend. Thanks to Hans-Peter Erl for the invitation and to Accenture for having us.
Loading sends your IP address and browser data to LinkedIn. LinkedIn may set cookies, associate your visit with your profile and process data in the USA. You can close the post at any time. Privacy details (German)

