Armin Ronacher is a software engineer and writer whose article "The Coming Loop" describes the next abstraction above direct coding-agent use: harness-level loops that keep tasks alive after the underlying model would normally stop. He is cautious rather than dismissive: loops already work well for porting, benchmarking, security scanning, research, and short-lived experiments, but he is uneasy about using them to write long-lived code that humans no longer understand.source: armin-ronacher-the-coming-loop-2026.md
Ronacher's core distinction is between the agent's internal tool loop and the outer harness loop. The outer loop can resume the same session, start a fresh session with modified context, route work to another machine, or ask another judge whether the task is done. In his framing, this moves the engineer's job from prompting a model to designing the queue, stopping conditions, and supervision structure around model work.source: armin-ronacher-the-coming-loop-2026.md
His concern connects directly to cognitive-surrender and ai-assisted-software-development: if loops write, review, patch, and maintain code, the codebase can become useful but organism-like, understood mainly through machine diagnosis. Ronacher argues the practical question is no longer whether engineers will use loops, but how they retain judgment, good engineering rules, and responsible human supervision while doing so.source: armin-ronacher-the-coming-loop-2026.md
Related pages: agent-loops, harness-engineering, ai-assisted-software-development, cognitive-surrender, loop-dependent-software.