The author's setup

This is how the book's author works with AI. It is not a requirement for readers. The book's exercises work on the free plans of the assistants for beginners. This page simply shows how a practitioner handles demanding work. Checked: 2026-06. Access and prices change, so check them. This is not advertising or financial advice.

The author's rule is simple: use the strongest models for serious work, use agents for routine and repeated chains of work, and keep a human check throughout.

Models for demanding work

Tool	Why the author uses it	Note
Claude (Anthropic)	Long documents, reasoning, book work, and code	Paid plan; the free tier is enough to try it.
OpenAI Codex	An agent for coding and development tasks	The same access limit applies.
GPT Pro "Thinking" (OpenAI)	Difficult reasoning, analysis, and review	The same access limit applies.

These are paid subscriptions. At the time of the source page, each cost about $100 per month. Check current prices. For most readers, that is more than they need. The difference from free tools shows up on difficult tasks; on ordinary tasks, it may be small.

Tool	Why use it	Note
Hermes (Nous Research)	A personal agent with memory that can learn skills	Easier to start with. See agents now.
OpenClaw	An agent environment for tasks	More powerful, with a higher learning curve.

The author prefers Hermes as the easier first step. Both tools are for readers who want to learn how they work. Most readers can start with the built-in assistants listed in agents now.

What to take from this

Take the approach, not the subscriptions: a strong model for important work, an agent for repeated work, and a human check every time. Start with what you can access in the beginner route. Add complexity only when a task pays for it. A tool should earn its place through the work it helps you finish, not because it is fashionable.

Version: 2026-06-20.