Giant Solutions

The book / in progress

Working With Coding Agents

Tools change every three months. This book teaches what survives the change: clear intent going in, trustworthy checks coming out.

By George Gens

Who it is for

Developers who already use AI and want it to hold up.

  • Your AI setup goes stale every few months, and you rebuild it each time.
  • You cannot tell whether the code the agent wrote is actually correct.
  • Your token bill keeps growing and the agent is still slow.

From the book

“Multiply a clean codebase with good tests and you get a lot of good software fast. Multiply a mess and you get a bigger mess, faster.”
Chapter 1
“Files can be re-read. Chat history can't.”
Chapter 3
“Trust is earned per workflow, not granted all at once.”
Chapter 5
“Ten agents producing work one person must review is inventory piled in front of the slow machine.”
Chapter 15

Contents

Five parts, eighteen questions

Part 1 · Why should I care?

  1. 1I change my mind about AI every three months
  2. 2Will AI take my job, or give me more work?Free below

Part 2 · How do I work with one agent?

  1. 3What is an agent, and how does it actually work?
  2. 4What is a harness, and which one should I use?
  3. 5I want to use AI, but I don't know where to start
  4. 6The agent started coding before I knew what I wanted
  5. 7What is an AGENTS.md, and what makes a good one?
  6. 8The agent re-reads my whole project for every task
  7. 9What is a skill, and do I need plugins and MCPs?

Part 3 · How do I know the agent did it right?

  1. 10I can't tell whether the agent's code is correct
  2. 11My agent keeps writing junk tests
  3. 12How do I assign reviewers to agent work?

Part 4 · How do I let it run further?

  1. 13How do I stop an autonomous agent from doing damage?
  2. 14My token bill is too high and the agent is still slow
  3. 15I have ten agents and no idea what any of them is doing
  4. 16How do I keep agents running while I sleep?

Part 5 · How do I take this beyond my own code?

  1. 17My company can't decide whether to allow AI
  2. 18How can I make money with these skills?

Free chapter / 2

Will AI take my job, or give me more work?

What is the Jevons paradox?

In 1865 William Stanley Jevons noticed something odd: as steam engines got more efficient with coal, Britain burned more coal, not less. Cheaper energy made new uses worth it, and demand exploded. Software is having its coal moment. When building something costs a tenth of what it used to, all the projects that were “not worth it” suddenly are: the internal tool, the one-off migration, the app for a niche of a niche.

What happens when typing gets cheap?

Writing code was never the whole job, but it was the visible, expensive part. Now an agent types faster than you, never gets bored, and works at 3 a.m. So the backlog of “someday” ideas becomes “this afternoon.” Companies won’t build the same software with fewer people; they’ll build far more software. The catch: someone still has to decide what’s worth building and whether it actually works.

What is scarce now?

Eliyahu Goldratt’s The Goal: a system moves at the speed of its constraint. Making any other part faster only piles work up in front of that constraint. Typing is no longer the constraint. Deciding what to build, and knowing the result works, is.

That’s the thread through the rest of this book. Every chapter is really about one of those two scarce things — clear intent going in, trustworthy verification coming out.

Want to know when it is out?

Send a short note through the contact form and you will hear when the book is ready.

Want this set up for your team?

The book is what I do in client work: find the bottleneck, change one workflow, and add the checks that make it safe.