One person does every step, and the method lives in their head. There is no template and no shared prompt. Just individuals doing the work the way they always have.

The work is only as available as the person who knows it. When they are busy, out, or gone, it waits.

Write down how the work is actually done, and give the agency one shared, repeatable version of it.

Everyone uses AI, privately and differently. One person has a prompt that works; nobody else knows it exists.

It feels like progress because individuals move faster. The agency has gained no shared method, memory or standard. Only more variance.

Pull the scattered know-how into one place the whole team can rely on, with an owner keeping it current.

The agency's own history, rules and approved material are current, findable and shared. Everyone works from the same source of truth.

Retrieval is solved, but the work still moves by hand. People stop searching and rebuilding; they still push each step forward themselves.

Let the routine work start itself, gather its own context and check itself, with a person on the exceptions.

Repeated work kicks off, assembles what it needs and checks itself against your rules. Anything uncertain or consequential is routed to a named person.

This is the right ceiling for most of the agency. A person still approves anything a client sees, anything with brand or legal exposure, and anything that affects someone's job. A ceiling is a commitment, and a commitment needs a check. The Lookout is what verifies the approval actually happened.

For a few low-risk, high-repetition areas, consider Learning. For most areas, stop here on purpose.

Outcomes feed back into a narrow, internal system so it gets better at one specific job over time.

Useful only in a few contained, low-consequence areas, and only after long, steady time at Running. It never removes a person from a consequential call.

There is no rung six. The goal was never maximum autonomy. It was the right autonomy for each area.