Human out of the loop
25 ways to let your AI keep working on its own, safely and robustly. A book by Floris Knol, in English and Dutch, as a paperback and an e-book.
Your agent does in an hour what took you an afternoon. And then you sit next to it for the rest of the day: a permission prompt, a code by text message, a login screen that keeps coming back, a question asked at three in the morning that waits until morning for an answer. The human in the loop is often the slowest link.
Lindeboom Fietsen
This book follows Marijke of Lindeboom Fietsen, a fictional bicycle business with a workshop, a webshop and four agents, through 25 ways to take the human out of the loop and still stay in charge. You see every way happen in her business first, then you do it yourself.
25 ways, in five parts
From access to a fixed role: what an agent needs to keep working without you sitting next to it.
Part IAccess
- 2.Your agent's own email address
- 3.Logging in, logging out and keeping sessions
- 4.One-time codes
- 5.Secrets outside the chat
- 6.Connecting without a plugin: the API itself
- 7.Where there is no API: the browser as the connection
Part IILimits
- 8.Your safety rules as hard limits
- 9.Testing whether your agent can escape
- 10.A gate for what the agent sends
- 11.A log that records everything
- 12.Human check as a setting, not a habit
Part IIIKeeping on running
- 13.Preparing work so the agent doesn't disturb you
- 14.All the questions at once, then carry on
- 15.Keeping on running when your computer is off
- 16.Restarting, a nudge, resuming
- 17.Recovering after an outage at the provider
- 18.A queue and a board
Part IVCost
- 19.Cheaper when it keeps grinding on
- 20.Caps per day and per agent
- 21.Fewer tokens
Part VRemote control and quality
- 22.Steering from your phone
- 23.An overview of what is running
- 24.Notifications that matter
- 25.Tests your agent runs itself
- 26.From one-off job to fixed role
To finish27. An afternoon to get started
Every chapter, the same shape
- The problem you run into
- Why it holds you back
- The approach, in plain words
- A hands-on block: a short prompt for your AI agent
- The pitfalls
- A checklist, and something to do the same day
Every prompt, and what it made
Download the hands-on blocks of the book: each prompt exactly as it is printed, with what an AI agent made of it in a clean practice environment. Its files, its own report, the checks it ran, and screenshots of each.
28 examples, one folder each, with a README · ZIP, 24 MB · free
Four editions
Available from 10-10-2026, most easily at bol.com: see where to buy the book.
Who it's for
For anyone who already works with Claude or ChatGPT, dares to hand an assignment to an AI agent and wants the work to go on while they are not sitting next to it. Everything is explained in plain words.
Also by me: From Zero to AI Agents That Do The Work, on how to build AI agents and put them to work.