A plan before agents: decide what each role should do with AI first.

Picture one page that shows what every role in your company will be able to do with AI by the end of next quarter. The project manager prepares every client update in minutes. The bookkeeper has overdue follow-ups ready each morning. The business development lead turns a client brief into a first proposal the same day. Everyone can see what is coming to their work, who owns each piece and what gets built first. That page is your plan, and it is how you 10X the ROI of your existing AI spend: across every role, not only for the one enthusiast.

Without it, AI arrives one purchase at a time. An agent for the inbox, another for meeting notes, a third that a team lead found and liked. Each answers the question “what can this product do?” None answers “what should our people be able to do?”

Start from the role, not the product.

A role is a kind of job: a project manager, a bookkeeper, a site coordinator. It is where capability lands, because it is where the work is done and where someone answers for the result. Planning by role does three things a list of products cannot.

• It covers the whole company.Walk every role and you find the work nobody pitched a product for, which is often the work that matters most.
• It outlasts any one person.A role’s plan holds when the person in it changes, and it never ranks one colleague against another.
• It tells you what to buy, build or leave alone.Once you know what a role should be able to do, a product either serves that or it does not.

What the plan settles.

1. What each role should be able to do with AI.Written as outcomes in the role’s own work: “prepare the Friday client update for review”, not “use AI for reporting”.
2. The work between teams.The work that matters crosses teams: a won proposal becomes a project, and a project becomes an invoice. Record each piece of work once and list it under every team that takes part. An AI org chart should show the work between teams shows how.
3. One owner for each piece of work.One role answers for each workflow’s result and approves every change to it. Who owns an AI capability after the pilot? sets out what that owner does.
4. What to build first.Rank the work on what you can see today: how often it happens, how many people do it and what it costs when it goes wrong. An unknown stays unknown, never a guessed zero. Choose your first AI wins has the scorecard.

Agents have a place in this plan. They go on work the plan has already understood, with an owner who knows what good looks like.

A one-page plan for a 40-person firm.

Here is a plan on one page for an invented firm: five teams, eight of its roles, and the order in which the work gets built.

Illustrative example with invented data. An invented 40-person commercial interiors firm. Its plan for the next three quarters.

Capability templates already exist for several of these: sales proposal preparation, client-status update preparation, overdue-invoice follow-up preparation and the employee onboarding plan. Each is adapted to the firm’s rules and tested on its own work before the role uses it daily.

Five questions before an agent.

An agent acts on its own. That is its value and its risk. Before one goes on a piece of work, answer these five. Five yeses and the work may be ready. Any no, and the next step is a capability your people run, with the AI preparing and a person deciding. Running it is also how you get to five yeses.

1. Is the work in your plan, with one role that owns it?An agent on unowned work does work nobody checks.
2. Are the rules written down and tested against your best person’s results?If the standard lives in one person’s head, an agent learns the usual way, not the best one.
3. Do you know the exceptions, and could an agent recognize every one?Three months of a capability’s exception log answers this. Without a log, you are guessing.
4. Is a wrong action cheap and quick to reverse?A rebooked site visit is. A message to a client in dispute, a payment, a legal deadline or a hiring decision is not, and those stay with a person.
5. Will someone read what it did, every week?Name the role and the day. An agent nobody reviews drifts from the standard without anyone noticing.

Weighing an agent product against a team that runs its own capabilities? An AI employee or a more capable team? compares the two on one piece of work.

Where Fully Powered stands.

We start with your plan: what each role in your company should be able to do with AI, the work between teams, an owner for each piece and what to build first. Then we build the first capabilities with your people, test them against your best person’s results and roll them out to the roles that do the work. AI prepares, people decide. An agent has its place once your plan and the five questions say a piece of work is ready.

Fully Powered Planner is where the plan lives: your teams in a tree, your roles, each workflow listed under every team that takes part, and Now, Next and Later with an owner on each. It is priced by the number of roles in your plan, never by the number of people. It opens soon, and until then we build the first version of your plan with you. The first 90 days of an AI adoption plan shows what follows, from the first build to deciding whether to expand.

Everything we build runs in the AI your team already uses: ChatGPT, Claude, Copilot or Gemini. Your platforms, your pace, your people, and your organization keeps control of how the work gets done.

Bring a list of your teams and the roles in each. The call is the first working session in your plan.