An ebook for developers
Automate Your Work Once and for All
Turn repeated development work into a process your agents can follow and your checks can verify.
Get the ebook Read a short excerpt firstChoose your store
Available in three stores
Each store shows its own price and availability for your country.
What it is
Give the agent a map.
Give the result a test.
A practical guide for developers who want to turn repeated work into a documented process. It walks through the pieces of an AI-assisted workflow, from gathering project context to checking the result before delivery.
- Explain your project once
Map your codebase and write skills: step-by-step procedures an agent can follow.
- Connect the tools
Use MCP (Model Context Protocol), an interface that lets agents call tools and receive results.
- Check before delivery
Define acceptance checks that can fail, rather than trusting a convincing completion message.
What’s inside
Thirteen chapters and three appendices
- The work that became inspectable
- Give the model the evidence
- MCP, the universal adapter
- Choose a tool you can verify
- Skills: teach it once
- Your first automation: a description you can check
- Giving it the map
- Cheap work and expensive work
- Done means a specific check passed
- Work that can run without continuous watching
- Ticket in, pull request out
- A complete exercise without writing code
- Your first week
- Appendix A: The companion and the printed exercises
- Appendix B: Prompt patterns that hold up
- Appendix C: Glossary
Includes diagrams, terminal examples and complete exercises. The examples use Python, Node.js, Git, Jira and GitHub.
Excerpt
From Chapter 1
The useful question is not how much work an agent can produce while I make coffee. It is what I can prove when I return.
The companion’s verification command tested its formatters and small checkers. It also checked whether the pipeline script was valid shell syntax. Those are useful tests. They did not exercise the complete sequence from ticket to delivery.
A syntax check can accept a script with all of those behaviours. A green result does not widen itself to cover things the command never ran.
Chapter 1, “The work that became inspectable”. The opening paragraph and two later paragraphs.
Before you buy
Who it is for, and who it is not for
It is for you if…
- you are comfortable in a terminal and use Git;
- you repeat a task often enough to write it down;
- you want checks you can run, not a promise.
It is not for you if…
- you want a guaranteed number of hours saved: time saved depends on the task, the setup and the review required;
- you want software that runs out of the box: this is a book, and live Jira or GitHub integrations need their own accounts and setup.
Companion code
Verify it yourself
The book describes a companion reference implementation with its own tests. Its local verification suite runs offline: no live model, Jira account or GitHub repository is used. From the reference-implementation folder, with Python 3, Node.js, Git and Bash on macOS or Linux:
bash verify.shIt prints REFERENCE-IMPL VERIFIED only when every step passes. That message concerns the local tests, not live integrations or time saved.
Chapters 3, 6 and 12 also contain complete standalone exercises you can run from the book alone. The source archive is separate from the ebook file, and a retailer may distribute only the ebook; Appendix A explains the details.
Questions
Is the book in English?
Yes. The book and this page are in English.
Which store should I choose?
Use the store where you prefer to keep your books. Google Play Books, Apple Books and Amazon Kindle are all available, and each shows its own price.
Does this page track me?
No. There are no cookies, analytics or advertising pixels. The only thing stored is your light or dark preference, in your own browser. Read the privacy note.
Start with one repeated task.
Keep the checks visible, and expand only after the result is reliable.