Everyone’s Still Prompting Their AI Wrong. The Best Engineers Quietly Switched to Loops

2026-07-06 · Source: Artificial Intelligence in Plain English - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Advanced, extended

Summary

Boris Cherny, creator of Anthropic's Claude Code, and other leading engineers have shifted from direct prompt engineering to "loop engineering" by mid-2026. This evolution, following prompt, context, and harness engineering, involves building autonomous systems that discover work, hand it to an AI agent, check results, record progress, and decide next steps. A loop comprises a trigger, discovery mechanism, a "maker" agent, a "checker" agent (crucially, separate from the maker), stop conditions, and memory between runs (e.g., `progress.md`). Tasks are "loop-ready" if they are checkable, have a bounded blast radius, are recurring, and stoppable. Production loops require both a quality brake (verifier) and a cost brake (budget guard). This paradigm redefines the engineer's role from operator to architect, emphasizing reliability engineering for AI systems. However, unattended loops risk compounding token costs, comprehension debt, and cognitive surrender.

Key takeaway

For AI Architects and ML Engineers designing autonomous systems, recognize that direct prompting is being superseded by loop engineering. You should prioritize building robust, self-managing systems that orchestrate AI agents, rather than manually supervising them. Implement independent "maker-checker" agents and dual quality/cost brakes to ensure reliability and prevent runaway expenses. Maintain human judgment and oversight, especially for defining stop conditions and reviewing output, to avoid comprehension debt and cognitive surrender.

Key insights

Shift from direct AI prompting to designing autonomous systems that manage AI agent workflows for repeatable tasks.

Principles

• The agent that writes code must never grade its own work; use a separate, skeptical "checker" agent.
• A task is "loop-ready" if it's checkable, has a bounded blast radius, is recurring, and stoppable.
• Production loops require independent quality (verifier) and cost (budget) brakes.

Method

A loop consists of five parts: a trigger, a discovery mechanism, a "maker" agent, a "checker" agent, and stop conditions, all connected by memory between runs (e.g., `progress.md`).

In practice

• Implement a `progress.md` file for memory across loop runs.
• Use GitHub Actions to run loops on a schedule.

Topics

• Loop Engineering
• Prompt Engineering
• AI Agents
• Reliability Engineering
• Software Development
• AI Safety

Best for: AI Engineer, Machine Learning Engineer, AI Architect

Related on AIssential

• Loop Engineering — Loop engineering shifts from direct agent prompting to designing autonomous systems that manage agent interactions and tasks.
• Loop Engineering (Beginner Guide): Why Top AI Builders No Longer Prompt Claude — Loop Engineering transforms AI interaction from one-off prompts to autonomous, self-correcting systems driven by defined goals.
• Loop Engineering Explained — Loop engineering shifts agent interaction from micro-prompting to designing autonomous, self-managing systems with verifiable goals.
• Stop Prompting. Start Looping. — AI agents can operate autonomously in loops, self-prompting to complete defined tasks and validate against goals.
• Loop Engineering, Clearly Explained! — Effective AI agent engineering shifts from prompt tuning to building robust systems around the core execution loop.
• Loop Engineering — Design autonomous systems ("loops") to prompt coding agents, replacing direct human interaction for iterative task completion.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence in Plain English - Medium.