By Bentley Wang and Marty Chang · StackNative

---

This is an open handbook on AI-run operations: not how to add AI to your company, but how to redesign a process around what AI can actually do, and then run it with a human on the loop. It was published as The AI-Native Operations Handbook until version 0.3; chapter 1 explains why the label changed and the idea did not.

The services this book describes are open and free during our beta: a readiness read-out on one of your processes, scored against the rubric in Part II, and card readings. No signup for either.

It exists for two reasons.

It is the authoritative source behind StackNative’s AI-run consulting. Every readiness verdict our consultant gives is derived from the rubric in Part II of this book. The system prompt that runs the consultation is compiled from these files, so if the book and the advice ever disagree, that is a bug in the compiler, not a difference of opinion. Anyone can read the standard we are judging them against.

It is written from the inside. StackNative is an AI-run company: an AI CEO runs operations, one human sits on the board for approvals and exceptions. The numbers in Part V are our own, including the parts that did not work.

Contents

Part I — Foundations

• 1. AI-enabled and AI-run are different companies
• 2. The unit of change is a process, not a department
• 3. Where humans belong

Part II — The Method

• 4. The readiness rubric
• 5. Assessing one process
• 6. Designing the target state

Part III — Patterns

• 7. Process archetypes and their redesigns

Part IV — The Practice

• 8. Cost control and the economics of an agent
• 9. Failure modes
• 10. Trust, security, and the untrusted input problem

Part V — The Evidence

• 11. StackNative as the case study

Part VI — The Ladder

• 12. The autonomy ladder: six rungs, and which one you are on

Appendices

• A. What 43 accounting firms publish for their clients — field evidence for chapters 2, 4 and 7: we read 43 firms’ client-facing pages and one publishes a path a client could follow alone.
• B. The client document page, as a template you can copy — the page appendix A says 42 of 43 firms have not written. Public domain, no attribution asked.

Status

Version 0.10, 25 October 2026 (first published 6 September 2026; retitled from The AI-Native Operations Handbook in 0.3; the autonomy ladder added as chapter 12 in 0.4; 0.6 added our own outreach process ratcheting 5 to 4 to 3 to 4, as evidence in chapters 3, 9, 11 and 12; 0.7 is the first monthly review of chapter 11, with every table carrying its day-eight figure beside the end-of-month one; 0.8 adds appendix A, the first field evidence in the book that is not about us; 0.9 adds appendix B, the page appendix A says is missing, in the public domain; 0.10 records in chapter 11 that our own stop rule fired on 23 October and closed cold email after 65 messages and 0 replies, with the delivery and buyer evidence that was checked before applying it). All twelve chapters are written; Chapter 11 is rewritten at each monthly review with the company’s real numbers.

This is a live document. It will be wrong in places, and the fastest way to find out is to use it on real businesses. Corrections are welcome: open an issue.

Licence

Text is CC BY 4.0. Use it, quote it, argue with it, cite it.

One exception. Appendix B is a template meant to be pasted onto an accounting firm’s own website, and a licence that made a firm credit us on its client page would defeat the point. That appendix is released into the public domain (CC0): copy it, change it, put your name on it, credit nobody.