Agentic Engineering Patterns: The Playbook I’d Hand to My Team

🤔 Curiosity: The Question

When I watch teams adopt Claude Code or Codex, the same thing happens:
we get speed, but we lose consistency.

So the question isn’t “which model?” anymore.
It’s:

What repeatable patterns actually make agents reliable enough to ship?

Simon Willison’s Agentic Engineering Patterns is the most practical answer I’ve seen so far.

---

📚 Retrieve: The Knowledge

What Simon’s guide actually gives you

It’s not a single essay. It’s a pattern library for working with coding agents:

Principles

• Writing code is cheap now
• Hoard things you know how to do

Anti‑patterns

• What not to do when agents are in the loop

Testing & QA

• Red/green TDD
• First run the tests

Understanding code

• Linear walkthroughs
• Interactive explanations

Annotated prompts

• A real, end‑to‑end prompt example (GIF optimizer)

Why I care (as a shipping engineer)

This library turns vague “be clear” advice into repeatable procedures.
It reads like a checklist I can hand to a production team.

Minimal operational loop (adapted)

flowchart LR
  A[Start task] --> B[Run tests first]
  B --> C[Apply smallest change]
  C --> D[Red/Green TDD]
  D --> E[Linear walkthrough]
  E --> F[Commit + log]

A lightweight “patternized” agent task

# Retrieve: enforce patterns as guardrails
TASK = {
    "goal": "Fix login race condition",
    "rules": [
        "run tests first",
        "change the smallest possible scope",
        "leave a linear walkthrough"
    ],
    "deliverables": ["test output", "diff", "walkthrough.md"]
}

---

💡 Innovation: The Insight

What I’d actually deploy

If I were running a game‑team codebase, I’d bake Simon’s patterns into three rules:

1) Tests before edits — no exceptions
2) Linear walkthroughs after edits — make future agents (and humans) faster
3) A prompt registry — “hoard things you know how to do” becomes a shared internal asset

Key Takeaways

Insight	Implication	Next Step
Patterns beat ad‑hoc prompting	Reliability goes up	Standardize checklists
QA is the real bottleneck	Test‑first wins	Require test logs
Knowledge hoarding compounds	Team velocity scales	Build a prompt vault

New Questions This Raises

• What patterns are unique to game pipelines (live‑ops, content, balance)?
• Can we automate walkthroughs as part of PR checks?
• What’s the minimum “pattern set” that moves reliability by 2×?

---

References

• Simon Willison — Agentic Engineering Patterns: https://simonwillison.net/guides/agentic-engineering-patterns/
• Intro post: https://simonwillison.net/2026/Feb/23/agentic-engineering-patterns/