READING

What the field is saying.

Picked by hand, kept short. These are the clearest posts and papers on agent loops, checks, stopping, and the harness around the model.

Start here

Simon WillisonAn LLM agent runs tools in a loop to achieve a goalThe compact field definition of an agent, and the best first link for the loop idea.↗AnthropicBuilding Effective AgentsThe canonical workflow-versus-agent framing, with the simple-loop-first advice every builder should start from.↗Simon WillisonDesigning agentic loopsA short, sharp guide to tool choice and stop conditions from one of the clearest writers in the field.↗

Go deeper

Yao et al.ReAct: Synergizing Reasoning and Acting in Language ModelsThe paper that made the think, act, observe cycle legible as the core shape of modern agents.↗Sydney Runkle (LangChain)The Art of Loop EngineeringShows how agent, verification, event, and improvement loops stack around each other in real systems.↗Addy OsmaniLoop EngineeringA practitioner-friendly explanation of designing the recurring system around the model, not just prompts.↗Simon WillisonHow coding agents workThe cleanest tour of the loop inside coding agents like Claude Code and Codex.↗AnthropicEffective harnesses for long-running agentsExplains the harness work needed when a loop runs across many sessions or context windows.↗

When you're building for real

AnthropicWriting effective tools for AI agentsTools are the loop's hands, and this is the practical guide to designing them well.↗Hamel HusainYour AI Product Needs EvalsThe best mindset piece on why checks and evals turn repeated model calls into useful products.↗CognitionA review of OpenAI's o1 and how we evaluate coding agentsA production look at how the Devin team decides whether a coding-agent loop is actually done.↗AnthropicEffective context engineering for AI agentsThe advanced companion to harness design: what the agent should see on each pass around the loop.↗