How a one-person studio directs an AI team.

The point is not to imitate a large team. It is to separate judgment from repeatable work, give each AI role a narrow contract and keep responsibility visible from brief to release.

01 Automate friction, not judgment

Start with the recurring friction around a decision. Sorting a long brief, comparing references, preparing variants, converting one source into several formats and recording accepted changes are defined tasks that can be delegated.

Keep judgment with the person responsible for the work. AI should protect attention for positioning, tone and selection; it should not decide those questions by default.

02 Give each role one contract

Separate research, synthesis, production and review. When one conversation invents an idea, approves it and writes the final version, confidence rises faster than quality.

Write a narrow contract for every pass. Research gathers labelled evidence; synthesis creates distinct options; production follows an approved direction; review searches for contradictions, missing states and unsupported claims. The studio owner accepts, rejects or redirects each result.

The advantage is not automatic creation. It is visible separation between generating, deciding and checking.

03 Keep one source of truth

Give the team one current brief, one decision record, one content inventory and one release checklist. AI can create plausible alternatives faster than a person can reconcile conflicting instructions.

Treat every other document as working material. Drafts may multiply, but accepted decisions must return to the source before the next pass begins; otherwise yesterday’s direction quietly reappears in tomorrow’s output.

04 Review in a separate pass

Use a distinct review pass to catch synthetic repetition. Repeated sentence shapes, generic certainty, identical compositions and unnecessary headings are production defects, not a house style.

Test writing for evidence and rhythm; test design at phone and desktop widths; inspect generated images at full size; test systems on the ordinary path, empty state and failure state. The reviewer should not inherit the producer’s assumptions.

05 Mark the human gates

Keep positioning, tone, omission, fairness and final approval deliberately slow. These are not clerical delays; they are the decisions that make the work accountable.

State only what the process can control. AI can shorten research and production, but it cannot guarantee audience response, business results or a perfect first draft. Nothing should advance only because the previous task finished.

06 Measure fewer returns

Track how much confusion reaches the next stage. Did the brief remain intact? Were the options meaningfully different? Can every decision be traced? Does the phone layout preserve the desktop priority? Is the release package complete?

Fewer corrections and cleaner handovers matter more than output volume. They show whether the system creates capacity or merely creates more material to review.

07 Keep responsibility singular

A one-person studio gains continuity, not headcount. The person who understands the question can stay close to every decision while AI carries repeatable transformations and exposes gaps before release.

The work can include research, production, criticism and quality control without pretending that many people made it. The useful model is one accountable practice organised well enough that care survives a larger workload.