TB: Agentic Engineering (Willison)

Core Thesis

Agentic engineering is the practice of building systems where AI models run tools in a loop to accomplish goals. Code execution is the defining capability that separates agentic systems from conversational ones. Good agentic practice requires a personal library of verified working patterns and strict accountability for agent-generated output.

Key Takeaways

The core pattern

  • "Agents run tools in a loop to achieve a goal."
  • Code execution is the defining capability that makes agentic engineering possible — it transforms the model from a text-producer into an actor.

Hoard things you know how to do

  • Maintain a library of working code snippets to provide as LLM context ("hoard things you know how to do").
  • Verified, working examples replace expensive re-discovery and reduce hallucination. The library is the craftsman's tool chest.
  • Practical note: prefer explicit tools (e.g. curl) over agent-chosen defaults that may summarize rather than return raw content.

Anti-patterns: accountability for output

  • "If you open a PR with hundreds (or thousands) of lines of code that an agent produced for you, and you haven't done the work to ensure that code is functional yourself, you are delegating the actual work to other people."
  • "They could have prompted an agent themselves. What value are you even providing?"
  • The professional obligation is to verify agent output, not merely to produce it. Submitting unverified agent code transfers your work to reviewers.

Mental Models

  • Second-Order Thinking — the second-order effect of agent-generated PRs is that code review becomes the real engineering work, done by someone else
  • Incentives Matter — if unverified agent output is accepted, incentives shift toward volume over quality

See also

Source

Simon Willison, simonwillison.net (Agentic Engineering Patterns guide), 2026 Original guide

Backlinks