Armin Ronacher · 2026-06-23 · notable
Armin Ronacher: 'The Coming Loop' — why even skeptics end up looping
Armin Ronacher argues 'harness loops' — outer systems that re-run AI agents past their natural stopping point — work well for code porting and benchmark runs, but breed defensive, dependency-creating code when pointed at real codebases.
Armin Ronacher on why 'harness loops' that re-run agents past their stopping point are coming for every team.
What is it?
Armin Ronacher's June 23, 2026 essay is a working programmer's take on the rise of agent harnesses such as Claude Code, Pi, and Fable that wrap a model in an outer control loop. The post argues that even engineers skeptical of looping agents will end up using harnesses, because the wins on certain workloads are real.
How does it work?
Armin Ronacher walks through where harness loops help — porting code between languages, hammering benchmarks, performance testing — and explains the wins as the harness redirecting the model when it would otherwise stop too early. He then walks through where they hurt: on a permanent codebase the same loop tends to produce overly defensive code, hide intent, and quietly grow cognitive dependencies on the AI system.
Why does it matter?
The Coming Loop is one of the first plain-English mental models for when to reach for an agent harness and when to push back. For teams about to wire a long-horizon agent into production code, Armin Ronacher's framing — short-horizon wins vs. long-term code-quality tax — is a much sharper way to decide than vibes.