The short answer.

People and AI work together on one team by splitting every piece of work the same way: AI prepares, people decide. AI gathers, sorts, compares, checks and drafts. A person starts the work, supplies what is needed, reviews what comes back and makes the decision.

Divide the work by writing that split down for each workflow, in five lines: who starts it, what AI prepares, who reviews, who decides, and what is never sent, paid or changed without a person. The split is the same on every team, which is what makes it easy to teach.

What AI does in a piece of work.

In a workflow, AI does the preparation: five kinds of work that come before a decision and take most of the time.

  • Gathers.AI pulls the records, emails, exports and notes the work starts from into one place.
  • Sorts.AI puts the items in the order the team uses: by client, by due date, by what needs attention first.
  • Compares.AI sets this month against last month, the invoice against the order, the request against the rule.
  • Checks.AI runs the team’s own checklist and marks what is missing, what looks wrong and what it could not assess.
  • Drafts.AI writes the summary, the reply, the brief or the cover note, with its sources in view.

The AI that does this preparation is the assistant your company already pays for: ChatGPT, Claude, Copilot or Gemini, on your own account, following a written workflow so it prepares the work the same way every time. AI prepares. People decide. shows how to shape what AI prepares so the review is fast.

What people do in a piece of work.

In a workflow, people do four things: start the work, supply what is needed, review what comes back and decide.

  1. Start it.A named role runs the workflow at the agreed moment: when a request arrives, on the third business day, before each client call.
  2. Supply what is needed.The person gives the workflow its inputs: the export, the file, the context only they know. AI works from what it is given.
  3. Review what comes back.A named role reads the prepared work against its sources, corrects it and notes every correction so the workflow improves.
  4. Decide.One person, by name or by role, makes the call: approve, send, pay, change the record, or stop. The decision stays with that person every time.

The person who starts the work and the person who decides are often different. Write both down. Which work goes into a workflow at all is a separate question; Repeatable work and ad hoc work sorts it.

Write the split down for each workflow.

Every workflow gets a written split, in five lines, before anyone runs it. The five lines answer the questions a new colleague asks on day one.

  • Who starts it.The role, and the moment the role runs the workflow.
  • What AI prepares.The output, in one sentence, and the inputs it starts from.
  • Who reviews.The role that reads the prepared work and corrects it.
  • Who decides.One person per decision. Two names on one decision means nobody owns it.
  • What is never sent, paid or changed without a person.The actions that wait for a person’s approval: a message to a client, a payment, a change to a record in your business systems.

A workflow can be designed to pull from the systems your company already runs and push data back. The fifth line is where those writes are named, and a person approves each one before it happens.

A “who does what” sheet to copy.

Copy these headings into a document for one workflow. Fill the sheet in with the person who does the work today and the person who decides. Keep the sheet next to the workflow, and update it whenever a role changes.

  1. Workflow name.The work, named for what it produces.
  2. Team.The team that owns the work.
  3. Workflow owner.The one person who keeps the workflow current.
  4. Who starts it, and when.Role, and the trigger or the date.
  5. What the person supplies.Each input, and where it comes from.
  6. What AI gathers.
  7. What AI sorts and compares.
  8. What AI checks.The checklist it runs, and how it marks a gap.
  9. What AI drafts.The output, and where it is saved.
  10. Who reviews.Role, and what the reviewer checks first.
  11. Who decides.One person per decision.
  12. Never without a person.Each send, payment and record change, and who approves it.
  13. Where corrections go.The log the reviewer writes in, and who reads it.

One workflow, split line by line.

Illustrative example.

Take a 45-person property management company that sends each owner a monthly statement. The accounting team owns the work, and the company wrote the split down before the workflow’s first run.

  • Who starts it.The accounting lead, on the third business day of the month.
  • What the person supplies.The month’s ledger export, the maintenance log and any notes from property managers.
  • What AI prepares.For each property, AI gathers income and expenses, compares them with last month, checks every repair charge for a signed approval, and drafts the statement and a short cover note to the owner.
  • Who reviews.Each property manager reviews the flagged charges and the cover notes for their own portfolio.
  • Who decides.The accounting lead approves each statement for release.
  • Never without a person.No statement is sent to an owner, no distribution is paid and no ledger entry is changed until the accounting lead approves it.

The same company wrote a second sheet for maintenance requests. The front desk starts it when a tenant’s request arrives. AI sorts the request by urgency, checks it against the lease and the vendor list, and drafts the work order. The property manager reviews and decides which vendor goes out. No vendor is dispatched without the property manager. Two teams, two kinds of work, one split.

The split is the same on every team.

The split between people and AI is the same on every team: AI prepares, people decide. Finance, client service, operations and sales each fill in different roles and different work, and the five lines do not change. A colleague who learns the split on one team already knows how every other team’s workflows run.

A team that works this way is a more capable team, with every decision still made by a person. An AI employee or a more capable team? sets out why that model holds up better than handing a whole job to software. When each team leader teaches the same five lines, rollout is a matter of repetition; How to deploy AI workflows to a whole team, and keep them working covers the steps.

What is different in a company of under 200 people.

In a company of under 200 people, the “who does what” sheet names real people you know, and one person often holds two lines. Most published advice assumes separate departments for review, approval and audit. A smaller company has none of those layers, so the sheet does their job.

  • The person who starts the work often decides too.That is fine when the sheet names a reviewer. When one person does the work and sends it, add one line naming who reads it first.
  • The workflow owner is the team lead.Nobody else will keep the sheet current. Who should own AI in a company your size? sets out who owns what.
  • The sheet replaces a policy binder.The “never without a person” line on each sheet tells your people exactly what waits for approval, in the work they actually do.

Start with one workflow per team. We have a plan. How do we build the first AI workflow? takes the first one from the sheet to tested work.

The “who does what” sheet belongs to all three steps in How organizations plan their use of AI: plan it, develop the workflows, deploy them. FullyPowered Planner is where the plan lives: you set out your teams, their roles and the work between them, and choose what to build first. We develop each workflow with you in a 3-Day Workflow Sprint or a 2-Week Workflow Sprint, with the split written down before the first run.