The short answer.

You share what works with AI across a company by turning each prompt that works into a named workflow, attaching that workflow to the roles that do the work, and giving it one owner who keeps it current. Collect what people use, keep one version for each piece of work, build it properly, and put it in the plan under the role. People then find a workflow by asking “what does my role run?”, not by searching a folder.

Sharing is how the second step, develop the workflows, feeds the third, deploy them. Our AI use is all over the place covers the first week of taking stock, and Using AI versus deploying AI draws the line between a person’s habit and a team’s method. This piece is about the handover in between: how one person’s good prompt becomes something every person in the role runs.

Why a shared folder of prompts decays.

A shared folder of prompts decays because a prompt in a folder has no owner, no home in anyone’s week and no record of what it needs. The folder starts with a handful of useful files. It fills with near copies, and nobody can tell which one the finance team actually runs.

  • Nobody owns any single prompt.When the person who wrote a prompt changes role, the prompt stays and goes stale. Nobody is asked to update it when a price list, a policy or a system changes.
  • A prompt does not say what it starts from.The writer knew to paste in last week’s export. The next person pastes in something else and gets a different result.
  • A prompt does not say who checks.The output goes out unread, or everyone reads it, and neither is a method.
  • A folder is sorted by file, not by job.A new hire opens it and has no way to tell which files belong to their role.

A workflow fixes all four, because a workflow is written down with its inputs, steps, checks and the decision a person keeps, and the plan says which role runs it.

Collect what people actually use.

Collect prompts from what people ran in the last month, not from what they remember. Ask each person to open their history in ChatGPT, Claude, Copilot or Gemini and bring the three conversations they would be sorry to lose. Memory produces the clever prompt they used once; history produces the one they run every Monday.

Collect through the team leads, team by team, with the same short form for every entry (below). The form matters more than the prompt text: the prompt alone is the part that travels worst. What the prompt starts from, who reads the output and what the person still decides are the parts a workflow needs.

Choose one version for each piece of work.

Choose one version for each piece of work, and let the others go. Group the collected entries by the piece of work they do, not by who wrote them. Where three people each have a prompt for the same weekly report, the team lead picks one, borrows the best check from the other two, and names it.

Then sort each piece of work into one of three piles.

  • Build now.Work that repeats every week or month, starts from records someone can export, and has a person who already reviews it.
  • Build later.Work that repeats but whose inputs are scattered or whose owner has no room this quarter. Put it on the plan with a date.
  • Leave personal.One-off help: a first draft of a letter, a summary of a long document. That stays each person’s own use, and that is fine.

Build it as a workflow and assign it to a role.

Build each chosen piece of work as a workflow, then attach the workflow to the role in your plan. Building means writing down the five things How to hand a piece of work to AI lists, and testing the workflow on last month’s real cases before anyone relies on it. The workflow runs in your own ChatGPT, Claude, Copilot or Gemini, on your account, run by your people. A person decides; the workflow prepares.

Assigning is the step a folder skips and the one that makes sharing last. In FullyPowered Planner you set out your teams, their roles and the work between them, and each workflow sits under the roles that run it, with its owner named. Anyone in the role sees what the role runs, and the next person hired into the role starts with that list. How new hires learn the AI workflows their role runs covers that first week.

Free workflow templates in the Workflows Directory give a starting point for common work. When nobody inside has time to build and test, we build workflows with you in a 3-Day Workflow Sprint, starting from the prompts your people already trust.

A collection form to copy.

Copy these headings, one form per piece of work. Write “Unknown” where nobody knows, with the name of who will find out.

  1. The piece of work.In plain words, as the team says it: “the weekly order confirmation”, not “email prompt”.
  2. Team and role.The team, and the role that does this work today.
  3. Who uses a prompt for it now.Every name. Several names means several versions to compare.
  4. How often.Weekly, monthly, per order, per customer.
  5. What it starts from.The export, file, form or message pasted in.
  6. Who reads the output before it goes anywhere.One name, or “nobody yet”.
  7. What the person still decides.The judgement that stays with a person.
  8. The prompt text.Pasted in full, from the history, not retyped.
  9. Pile.Build now, build later, or leave personal.
  10. Owner.The one person who keeps the workflow current once it is built.

A brewery shares what works.

Take a 95-person craft brewery with a brewhouse, a taproom and a wholesale side that sells to distributors and local accounts. The operations director asked each team lead to collect with the form.

Illustrative example.

  • Wholesale.Two sales reps each had a prompt for confirming distributor orders, started from different exports. The wholesale manager kept the one that starts from the order system export, added the other rep’s check for keg returns, and named it the distributor order confirmation. Pile: build now. Owner: the wholesale manager. Role: wholesale sales rep.
  • Taproom.The taproom manager wrote the weekly tap list and staff briefing from the brewhouse’s release schedule. Nobody else knew the prompt existed. Pile: build now. Owner: the taproom manager. Role: taproom shift lead, so every shift lead can produce it.
  • Brewhouse.The head brewer summarised batch notes with a prompt, but the notes sit in handwritten logs. Pile: build later, once the logs move to a shared sheet.
  • Finance.The bookkeeper used an assistant to draft one-off letters to suppliers. Pile: leave personal.

The operations director decided to archive the brewery’s old shared prompt folder once both workflows sat in the plan under their roles, so that wholesale reps and taproom shift leads look in one place for what their role runs.

What is different in a company of under 200 people.

In a company of under 200 people, sharing what works with AI is a few conversations with team leads, not a programme. A large company sets up a community of practice and a prompt library with moderators. You know who wrote the good prompts and can ask them this week.

  • The best prompt often belongs to one person.That person is usually the natural owner of the workflow. Ask them first.
  • One role can be one person.Attach the workflow to the role anyway, so it survives a holiday or a resignation.
  • Duplicates are few and visible.Two versions of one prompt are easy to compare side by side with both writers in the room.

Collect this month, choose one version for each piece of work, and put each workflow under its role. FullyPowered Planner is where the plan lives, and your colleagues add their own teams. How organizations plan their use of AI covers the whole method.