The Intent Loop
Product design when building is cheap. A method for small AI-native teams.
The product design books I learned from were written for a world where building was the expensive part. That cost structure no longer holds. This is my attempt to design a method for the world we're actually in, and to be honest about what is evidence and what is still a bet.
Four bets
Intent is the product
The written intent (purpose, decisions, evals, design system, constraints, agent contract, decision log) is the source of truth. Code is an output that can be regenerated.
Build to decide
When building is cheap, disagreements are settled by building several working versions and gathering evidence, not by debate or sketches.
Budget judgment, not effort
Human attention is the binding constraint. It is planned, spent on purpose and counted.
Two kinds of user
Every product is designed for people and for agents from the start.
Six steps. The unit of work is a decision, not a feature.
Frame the decision. Fan out several working versions with agents. Prove them with evals and users. Judge, as humans. Harden what wins. Watch it in use. Then write back what you learned into the intent, so the next pass starts from a better place.
Synthetic users may eliminate variants; only real users (and real agents) may choose one.
Inside
Part I
The Shift · chapters 1–3
The ground has moved. What the evidence says about how cheap making has become, what the classic methods assumed about cost, and which things are now scarce.
Part II
The Intent Loop · chapters 4–11
The method: the written Intent Stack, the six steps of the loop, the budget of human attention that paces it, and design for the agents that now use products alongside people. A four-person team, Harbour, works through three decisions.
Part III
Teams and Rhythm · chapters 12–13
How roles have changed, how to assign responsibilities rather than titles, the cadence that keeps the loop moving, and the common ways it fails.
Part IV
When the Product Is AI · chapters 14–16
Autonomy and trust, evals and model behaviour as design materials, memory and proactivity, and interfaces beyond the screen.
Part V
Evidence and Strategy · chapters 17–19
Field notes on sixteen public cases, separating what companies say from what others have found, and what strategy means when anyone can build your product.
Part VI
Putting It to Work · chapters 20–21
An adoption guide for a small team starting next week, and how to test the method honestly, including the results that go against it.
Free, in exchange for an email.
You'll get the full PDF now, and an occasional note when a chapter changes or the method gets tested. Nothing else. Unsubscribe is one click.
Version 0.1, September 2026. It is a design, not a result: chapter 21 sets out how it should be tested.