Adoption & strategy
The Three Ways of Working
You use a product, you direct a tool, you run a service - whose hands are on the work, and are you in the room?
Use → Product · Direct → Tool · Run → Service
Nothing about the model changes. Your relationship to it does.
Your hands on the work, AI advising.
Its hands on the work, your eyes on it.
Its hands on the work, nobody watching live.
Same engine, three arrangements: Claude in the chat app is a product; Claude Code under your direction is a tool; Claude Code on a schedule is a service.
A tool left running unattended is not yet a service - it carries a service's risks without a service's checks. A service is built for your absence from the start: checking designed in, permissions bounded, escalation defined.
Most work belongs at product and tool, some of it permanently; services suit narrow, high-volume, well-bounded work. Judge each piece of work from where you sit.
Not the Permission Ladder: the Ladder sets how much autonomy you permit a task to have; the Ways say who does the work and who watches. They cross - deep research is Run-level permission inside a product. Not the Implementation Tiers: the Tiers say how you source AI; the Ways say how you work with it once it's there. Not Where Does It Run: the Ways describe your relationship to the system; venue describes where its two parts sit.
When to use When placing any piece of AI work - and weighing what you take on if you move it: whether it acts or only produces (Produce vs Act), and where it runs.
Read the full post →