The short answer.
Letting employees figure out AI themselves is right for learning and wrong as the whole approach. Exploring is how people learn what ChatGPT, Claude, Copilot or Gemini can do, so keep that freedom. But the company owes its people three things no individual can supply alone: which piece of their own work to hand over, what good output looks like for that work, and a person who checks it with them. Without those three, people stay at asking and drafting, and the company gets a fraction of what it pays for.
The fix is one sentence to every team: explore as much as you like, and here is the one workflow your role runs.
What people learn by exploring on their own.
Employees who explore an AI assistant on their own learn what it is good at, where it goes wrong and how to ask for what they want. Experimenting was the right way to start, and the people who did it are the ones a company most needs next.
- How to ask.People learn that context, an example and a clear instruction change the answer.
- Where it slips.People learn, by catching mistakes, that output needs checking before it goes anywhere.
- Where it helps them.People find the questions, drafts and summaries that save them effort every day.
Where figuring it out alone stops.
Employees left to figure out AI alone stop at asking questions, drafting and summarizing, because handing over a whole piece of work needs decisions an individual cannot make. A coordinator can try an assistant on the weekly client report. The coordinator cannot decide that the report should be prepared this way for everyone, cannot set the standard the report has to meet, and cannot assign someone to check it. Those are the company’s decisions.
So undirected AI use plateaus. A few enthusiasts build private habits, most people use the assistant now and then, and each good find stays on one desk. What happens when a company uses AI without a plan? sets out what that costs.
What do we owe our people if we expect them to use AI?
A company that expects its people to use AI owes them three things, and each one is a decision only the company can make.
- The piece of work to hand over.Name one piece of each role’s own repeat work, in the words the team already uses: “the renewal summary”, “the month-end note”. People cannot hand over work nobody has said they may hand over. How to plan AI by role shows how to choose it.
- What good looks like.Write down the standard for that work: what it starts from, what it must include, what it must never say, and an example of a finished one the team is proud of. That standard becomes the workflow: your best way of doing the work, written precisely enough that the assistant prepares it the same way every time.
- A person who checks it with them.Name one person, by name, who reviews the output alongside the person running it until both trust it. A workflow prepares; a person decides.
What work should you not give to AI? covers the limits people are also owed.
Should we encourage experimenting or direct it?
Encourage experimenting and direct the work, both at once. The two do different jobs. Experimenting is how each person learns the tool. Direction is how the company gets its repeat work done with AI, the same way, by everyone in the role.
- Keep the freedom.People go on exploring for their own questions and drafts. Nobody needs permission to try something.
- Add one workflow per role.Each role gets one named workflow, with a standard and a checker. That one is not optional, because it is how the work is now done.
- Bring good finds in.When someone’s experiment works on repeat work, it goes to the plan as a candidate for the next workflow, and the person who found it helps write it.
- Never scold.Nobody is told they used AI wrongly. People explored because the company left them to, and that was a sensible start.
Direction here means focus: one named piece of work per role, done on purpose. How do we get people to actually use AI at work? covers building each workflow with the team.
A page to share: what we ask of you, what we give you.
A leader can send this page to the team as it is. Fill in the blanks for each role.
What we ask of you.
- Keep exploring.Use our AI assistant for your own questions, drafts and summaries whenever it helps.
- Run your role’s workflow.For your role, that is: ______________________. Use it every time that work comes up.
- Check before it leaves.Read every output before it goes to a client, a colleague or a system. You decide; the workflow prepares.
- Log what you correct.Note each case where you changed the output, and tell ______________________.
- Share what works.If you find a good way to do other repeat work, tell us. It may become the next workflow, and you will help write it.
What we give you.
- The work to hand over.One named piece of your own work, chosen with your team.
- The standard.What good looks like for that work, written down, with an example.
- A person who checks it with you.______________________ reviews your outputs with you until you both trust them.
- Plain limits.What never goes into the assistant, and what always waits for a person to approve.
- No rating.Nobody is ranked or reviewed on how much they use AI. The workflow is judged by whether the work is useful.
One brokerage adds the direction.
Illustrative example.
Take a 45-person insurance brokerage whose account managers had used an assistant on their own for a year. One used it to compare policy wordings; most opened it now and then. The managing director did not ban anything or audit anyone. She told the team the exploring had been the right start, and that each role would now get one workflow. For the account managers it was the renewal summary sent to each client before a renewal meeting. The account manager whose summaries clients praised most walked the team through how she wrote them: what she pulled from the agency’s management system, what she always explained, and what she never promised in writing. That became the standard and the workflow. The service team lead was named as the checker for the first month. The account manager who had been comparing policy wordings brought that method to the next planning meeting, and it went on the plan as the second workflow.
What is different in a company of under 200 people.
Most advice on AI at work is written for large enterprises, with acceptable-use programs, training catalogs and champions networks. A company of under 200 people owes its people the same three things and can supply them faster.
- The leader can say it in person.One message at a team meeting from the owner or managing partner settles that exploring was welcome and that each role now gets a workflow.
- The standard already exists in one person.A small team usually has one person whose work everyone copies. Write that person’s method down and the standard is written.
- The checker is down the hall.The person who reviews the output is usually the team lead, who already reads the work.
That is how a small company gets 10X the ROI of your existing AI: the same seats, the same freedom, and one directed workflow in every role.
Where to go next.
Choosing the work for each role is part of a larger plan. How organizations plan their use of AI sets out the three steps: plan it, develop the workflows, deploy them. Why plan how your company uses AI? sets out what changes once the plan exists.
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. Colleagues add their own teams. Free workflow templates give each role a starting point, and we develop workflows with your team in a 3-Day Workflow Sprint or a 2-Week Workflow Sprint.