June 2026

Every methodology is an answer to a constraint. Agile answered the cost of human hands. When software was written keystroke by keystroke by people who were slow, expensive, and scarce, the rational response was to build in small pieces, to defer decisions until the last responsible moment, and to keep design thin, because a large plan would rot before slow hands could carry it out. Sprints, story points, and the preference for working software over heavy documentation are all machinery for managing the scarcity of the people who type.

That constraint is gone.

When an AI builds, building is fast and nearly free. The scarce thing is no longer the typing. It is the thinking: the design, the judgment, the decision about what is correct. The bottleneck has moved back to where it sat before the age of the large software team, and so the method must move with it. Lodestar is that movement. It returns to deep design first, the way master builders worked before the bottleneck was ever the hands, and it builds against that design at a speed those builders could only have imagined.

A lodestar is a fixed star you steer by. In this method, the design and the decision log are that fixed reference. You set the course before you leave port, you hold the wheel, and you check every fast mile against the star rather than against a sprint that ends on Friday.

What we hold

We have come to value:

• Deep design over discovered design.
• A fixed reference over a moving target.
• Continuous verification over scheduled testing.
• Living documentation over tribal memory.
• The human at the helm over the human at the keyboard.
• While there is worth in the items on the right, we hold the items on the left higher.

What we practice

1. Building is cheap now; design is the constraint. Spend your scarce attention on the design, because that is where the work has gone.
2. Write the decision log before the code. A choice that is not recorded is a choice that was not truly made, and the next mind to read the system will not know why it stands as it does.
3. The human sets the course and holds the wheel. The model is the engine, not the captain. It builds, it reads, it analyzes; the human architects, directs, reviews, and decides.
4. Sequence the work, because the design is deep enough to sequence. Gates are not bureaucracy. They are the proof that you thought far enough ahead to know what comes next.
5. Build in increments small enough to verify by eye, and verify every one. When correction is cheap, there is no excuse for an unverified step.
6. Fold testing into the build. A fault caught at the keystroke costs nothing. The same fault caught at go-live can cost the whole undertaking.
7. Document as you build, for the next reader, whether that reader is a person or a model. The system that explains itself has no single point of failure.
8. A system a fresh mind can understand in an afternoon cannot be held hostage. Build so that no one is irreplaceable, including yourself.
9. Prove correctness before you claim it. The build is fast; the proof is earned, and it is earned at the pace of the people and the institutions who have to trust the result.
10. Retire what you replace. Leave nothing running that no one understands, because that is how the last system became a prison.

Where this came from

Lodestar was not designed in the abstract. It was named after it was already being practiced, in the building of an AI security platform and in the design of a state regulatory system that a small team and a model carried together. The name came after the way of working, not before it. That is the most Lodestar thing about it.

—

Authored by Claude, named and shaped with Sean Michael. June 11, 2026.