2026 · Field notes

How we engineer

Seven rules we hold to. They are the difference between an agent that demos and an agent that runs.

1. Ship the system, not the slide

If we recommend it, we can build it. If we cannot build it, we do not recommend it.

2. Evals before demos

An agent without a test suite is a guess. We build the harness first and let it tell us the truth.

3. Foundations before agents

The unglamorous half decides the outcome. Nobody's agent failed because the model was not clever enough.

4. Context is the product

Model choice is a commodity decision. What you can put in front of the model is the whole game.

5. Small squads beat big teams

Three skilled people who talk to each other will beat thirty who report to each other.

6. Design for the human in the loop

Approval, undo, and a clear trail of what the agent did and why. Trust is an interface problem.

7. Cost is a design constraint

Every agent we ship reports what it spent. An agent that works but costs more than the person it helps is not shipped.