Radical Geek field guide

Radical Geek's Guide to Agentic Engineering Patterns

A practical pattern catalogue for governed agentic engineering, bounded work classes, human accountability and independently verified evidence.

For
Engineering leaders, architects and developers building with agents
Format
Practical pattern catalogue
Access
Free PDF guide
Version
2026.07.28
Cover of Radical Geek's Guide to Agentic Engineering Patterns

Why this guide

An agent can produce a convincing patch very quickly. That does not tell you whether it understood the work, chose the right boundary, tested the risky path, or left enough evidence for somebody else to trust the result.

This catalogue is drawn from building agent teams that plan, make changes, test, review, escalate and report through inspectable engineering systems. It is designed to be used while shaping a real workflow, not admired as a collection of diagrams.

It is part of the Radical Geek guide system, where these patterns implement the shared lifecycle from governed work class to accepted outcome.

Use the patterns as a working catalogue. Start with the failure mode in front of you, then add only the structure needed to make the work inspectable and repeatable.

What you will leave with

A guide built to be used.

  1. 01

    Define governed work classes with bounded inputs, authority, stopping conditions and acceptance rights.

  2. 02

    Choose patterns for planning, delegation, model routing and escalation while preserving human accountability.

  3. 03

    Introduce independent evidence and deterministic checks before work reaches an accountable reviewer.

  4. 04

    Use Proof of Operation to qualify patterns against representative work and exceptions before wider delegation.

  5. 05

    Connect agent activity to CI, evidence-gated merge, deployment and rollback.

Inside the guide

The working ground it covers.

  • Planning, decomposition and context patterns
  • Specialist agents, sub-agents and hand-offs
  • Model routing and escalation paths
  • Adversarial review and verification gates
  • CI, merge readiness and deployment evidence
  • Safe tool use and human control points