This page is outside your selected path.
How to study this if you...
Two questions on this material help you choose your next step.
Materials familiar from answers are hidden.
Highlighting is partial. Follow scrolls to matching passages and waits for the next match in between.
The book is assembled as a learning path across five maturity levels with a preparatory Module 0, but your entry points as readers are obviously different. On this page I would like to show different learning tracks: from the very first page, take what your particular role needs.
If you have no experience with the terminal, code, and Git, start with Getting started with AI agents. M0-P9 completes the first independent project. Then lessons 0.15–0.20, patterns 0.1–0.3, and practices M0-P10–M0-P16 cover code, contracts, data, application structure, access, and releases. After this preparation, read the technical Audit Log description and choose the next route for your role. Lesson 0.14 is part of the main route. If the cloud environment is unavailable, review the preparation for M0-P8, record that the cloud run did not take place, and start M0-P9 from M0-S7. M0-P17 is optional.
Quick entry to the foundations
If you already work with a project, choose a specific gap. A title or role does not replace a demonstration of the skill.
| What is difficult to explain or check | Where to start |
|---|---|
| The behavior of a small function | Lesson 0.15. Understand a small program and an agent’s change |
| Fields, errors, and interface compatibility | Lesson 0.16. How application parts agree: contracts, errors, and compatibility |
| Related records and preservation after a failure | Lesson 0.17. How shared application data works |
| Responsibilities, a shared operation, and an external dependency | Lesson 0.18. System design for a small application: parts and dependencies; Pattern 0.1. One application operation for several interfaces; Pattern 0.2. Access an external service through a gateway |
| A format change for existing data | Pattern 0.3. Change a format while preserving working data |
| Access to a specific record and secrets | Lesson 0.19. Who can read and change data |
| The actual version, release, and rollback | Lesson 0.20. Release a working version and verify it |
Senior engineer / individual contributor?
If your goal is to move personally from chat to controlled work with an agent, read the introductory materials and the running example, then modules 1–3 in full. In module 4, take lessons 13–16 and patterns 14, 15, 16, 17, 21, and 23.
In module 5, start with Lesson 21. Trust boundaries: the harness, model, and tools and pattern 27, then read Lesson 23. How to evaluate and update agent configuration and patterns 29 and 34. Complete the main route with Lesson 26. Capacity for meaningful oversight. This prepares you to check changes to your own configuration, turn recurring review feedback into checks, and understand a result before accepting it.
When a task calls for several workers, read patterns 18–20, followed by Lesson 17. Orchestrating tasks, teams, and organizations and Methodology 11. Gas Town on Gas City: organizing agent development. Lesson 24. Agent platform as an internal product and pattern 33 help when maintaining shared tools and transferring a practice; Lesson 25. AI-native organization: agents coming into contact covers work with other participants. Choose pattern 30 for model comparisons and pattern 35 for checking your own skill.
Tech lead / staff / architect
If you are establishing a shared way of working with agents in a repository and a team, read the introductory materials and the running example, then modules 1–3 in full. In module 4, complete lessons 13–16 and patterns 14–23, followed by Lesson 17. Orchestrating tasks, teams, and organizations and Methodology 11. Gas Town on Gas City: organizing agent development. Here you learn to check the workers’ shared result and resume their work after failures.
In module 5, complete the base sequence from preparing a team to meaningful human participation. It covers shared configuration checks, transferring a practice, maintaining tools, and agents working together. Add model selection and caching in patterns 30–31, and patterns 36–38, when a specific task requires them.
Engineering manager / head / CTO
If your goal is to understand what to adopt, in what order, and which result to take responsibility for, start with “How the program is structured,” the introductory lesson, and the running example. Then read module 1 as a maturity map and a shared language. In module 2, take lessons 7–8 and patterns 1–2. In module 3, read lessons 10–11 and the introductory parts of the methodologies to understand which tasks they suit and what the team will need.
In module 4, read lessons 13–16 and patterns 14–15, 21, and 23. Review the coordination approaches in patterns 18–20, then Lesson 17. Orchestrating tasks, teams, and organizations and Methodology 11. Gas Town on Gas City: organizing agent development. For the methodology, focus on roles, dependencies, and acceptance of the shared result. P106 and P107 use supplied descriptions and do not require installing Gas City.
The main module 5 route covers running an experiment, total costs, evaluating updates, transferring a practice, and responsibility for decisions. Read the platform and AI-native organization lessons to choose how work is organized and who owns it. The technical detail in patterns 27–28 and 30–31, as well as patterns 36–38, is task-dependent.
Head of AI / AI function lead
If you own shared AI tooling and its use across several teams, start with the same management foundation. In module 4, study patterns 18–20, the orchestration lesson, and methodology 11 in greater depth.
In module 5, complete the entire base sequence, including isolation, corporate sources, model selection, and caching. You need to understand who maintains the tools, how shared configurations are updated, and how each team verifies its results.
Optional patterns for any role
- Choose Pattern 36. Recurring work: heartbeat and scheduled runs when a task already works once and needs to run repeatedly on fresh data.
- Use Pattern 37. Adaptive restoration of operator participation when a review or approval is postponed and someone needs to return to the decision after a pause.
- Use Pattern 38. Agreeing on a shared dependency through agents when agents working for different participants need to agree on shared work and check the combined result.
Of course, the most complete way is going through it cover to cover, but I understand that not everyone has the time and bandwidth for that.
Your notes, a self-check, and the next step
My lesson notes 0Only visible to you
Check yourselfOptional · Does not affect access
Knowledge check
Check what stayed with you
This short check does not affect access or your read mark.
Sign in to take the check and save every attempt.
Sign in and answerChecking read state
Marked. On My learning this is a material read and a learning day.