March 4, 2026

Gang of Four? Meet Gang of Bot

Agentic Engineering Patterns

Are we reinventing the 90s playbook for robot coders

TLDR: Simon Willison shared a guide for getting better results from AI code helpers using test-first routines and clear steps. Commenters split between calling it obvious, saying it only shines on simple, testable tasks, and wondering if we’re reviving 90s pattern book hype—many use AI for boilerplate, not mission-critical work.

Simon Willison dropped a tidy guide to “Agentic Engineering Patterns” for coding agents—think robot interns like Claude Code and OpenAI’s Codex that write and test software. He champions simple rules: writing code is cheap now, always run tests first, and use clear walkthroughs to understand how the bot’s code works. It’s very “give the intern a checklist” energy. You can read his intro here.

Then the comments lit up. One reader waved a different playbook: StrongDM’s Dark Factory is “more immediately actionable (sorry, Simon!)” with a link to the rival guide here. Another voice of hard-earned wisdom said agent workflows shine only when the task is easy to test—like compressing files where success is a clear pass/fail. The snark brigade chimed in: “Isn’t this how everyone uses agents?” Meanwhile, the nostalgia crew asked if we’re about to revive 90s-style pattern book hype, and a weary practitioner admitted AI is great for boring boilerplate but not trusted for serious ownership, “never been very impressed with the output.”

Cue memes: readers joked about the Gang of Four pattern book era becoming Gang of Bot, with agents as eager interns who need tight tests and strict steps. The vibe: Simon’s guide is helpful and tidy, but the crowd wants proof it works beyond toy tasks—and they’re side-eyeing any return to glossy pattern playbooks.

Key Points

  • The page introduces “Agentic Engineering Patterns,” a collection of practices for coding agents.
  • It highlights two Principles: “Writing code is cheap now” and “Hoard things you know how to do.”
  • Testing and QA topics include “Red/green TDD” and “First run the tests.”
  • Understanding code topics include “Interactive explanations” and “Linear walkthroughs.”
  • An Appendix provides “Prompts I use,” and the page links to an introduction and notes sponsorship by Augment Code.

Hottest takes

"more immediately actionable (sorry, Simon!)" — ukuina
"Isn’t this pretty much how everyone uses agents?" — chillfox
"never been very impressed with the output" — benrutter
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.