Most firms are adopting AI faster than they are recording it. Use spreads tool by tool and person by person, so nobody can answer the questions that matter later: where AI touched a project, who reviewed the output, and what the firm decided it would never hand over. In professional work, that record is the difference between adoption and exposure.
A short intake that turns a described AI use into a practical governance decision. You log the task, and the register returns a risk level, a review gate, the evidence the use needs to carry, and an entry the project team can keep and export.
Describe the task, pick the project stage and output type, and state whether sources are available and whether the output will leave the firm. The register then places the use at one of four levels: low for internal drafting with no project decision attached, medium for project support that a person reviews before use, high for external-facing or decision-support output, and blocked for anything that touches legal, engineering, consent, or compliance sign-off.
Nothing high-risk is ever marked as automatically approved. The register recommends the gate; a named person still owns the decision. The tool exists to make that ownership visible, with each entry recording the boundary, the required evidence, and the reviewer.
Whether AI governance can live next to the work as a five-minute habit instead of a policy document nobody opens. If logging a use is quick and the recommendation is practical, the register gets used. If governance arrives as a PDF, it gets skipped, and the record never exists.