Is your team ready to let agents build?

Twelve questions, three minutes. Answer honestly for the team as it works today, not as you intend it to. "Yes" means it happens reliably, not that someone once did it.

Intent

Intent is the product

Is the purpose of your product written down in one place that agents and people both read before building?

Could you rebuild a slice of your product from your written intent alone and have it pass its checks?

When a decision is made, does something get written back (a new constraint, eval or log entry) every time?

Evidence

Build to decide

When the team disagrees, do you build two or three working versions rather than argue or sketch?

Do real users, not just simulated ones, pick the version that ships?

Do you have automated evals that a change must pass before it is considered done?

Judgment

Budget judgment, not effort

Is human attention planned each week (who frames, who judges, who reviews) rather than assumed?

Do you limit how many variants you build to how many you can properly judge?

Does someone own watching the product after release, with time set aside for it?

Agents

Two kinds of user

Does your product have a surface designed for agents (tools, APIs, protocols), not only screens for people?

Is there a written contract for what your own agents may and may not do without a person?

Has an independent agent completed real tasks through your product before you shipped a decision?

Where should the results go?

Your score, the reading, and a link to the book. One email; no sequence.