Being good at prompting isn't enough.
For a couple of years, "learn to prompt" was the advice everyone gave about AI. It was a reasonable first lesson. It was never supposed to be the whole course, and in a lot of organizations it has turned out to be exactly that.
The first rung of a longer ladder.
Most teams now have a handful of people who can get a good answer out of a chatbot. They give it some context, ask for a particular format and push back when the first draft is weak. That skill is worth having, and nobody should stop using it.
The limit is that a good prompt improves one conversation. A grant coordinator gets a cleaner summary of a funder's guidelines, an office manager gets a politer reminder email, and the next morning the work runs the way it always did. One conversation at a time rarely changes how a department operates.
The people we see getting steady value from AI think less about the wording of a request and more about the work surrounding it. They ask what goes in, what should come out, who checks it and what happens to it afterward. Those are questions about the job, and the prompt is only one line of the answer.
Why teams stall at the prompt.
Prompting is easy to teach in an afternoon, which is why it fills most training sessions. People leave with a page of tips, open their laptops the next day and return to the same workflows they had before. The tool gets used for odd jobs. Staff say it saves them a bit of time, and a year later the intake process, the monthly report and the proposal template all look the same.
The teams that stall usually have capable people. What they lack is any training in the skills on either side of the prompt, so the organization got better at asking questions and stayed where it was on everything else.
A great prompt improves one conversation, and one conversation rarely changes how the work gets done.
The skills that come after.
Four skills sit around the prompt, and each one is closer to management than to anything technical.
Briefing is handing over a whole task the way you would to a capable new hire: the goal, the reader, the constraints and what a finished version looks like. Judgment is knowing when an answer is wrong, thin or confidently invented, and having enough knowledge of the subject to catch it. A clinic's billing lead will spot a bad coding suggestion that a general user would wave through.
Workflow design is breaking a recurring job into steps and deciding which ones AI should do, which ones a person keeps and where the handoffs fall. Sharing is turning a chat that worked into something other people can use, such as a saved template, a configured assistant or a short written procedure that sits where the team already looks.
The people best placed to learn these are the ones who already run the work. A school district's transportation coordinator knows which parts of route planning need a human call. A firm's senior associate knows what a weak first draft of a proposal looks like. Neither needs to become technical. They need practice applying what they already know to a new kind of helper.
Four rungs, in order.
It helps to think of AI skill on a team as a ladder with four rungs. The rungs build on each other, so a team that skips one tends to slide back to it.
On the first rung, people ask: one-off questions, quick drafts, a rewrite of an awkward paragraph. On the second, they brief: they hand over whole tasks with full context and a clear standard for done. On the third, they build: someone redesigns a recurring workflow around AI and puts a check on the output. On the fourth, they share: what works stops being one person's private habit and becomes something the team uses.
Most organizations we talk with have a few people on the second rung and nearly everyone else on the first. That is a normal place to be. The mistake is answering it with another round of prompt tips, which only makes the first rung more comfortable.
Where does your team sit?
Use the checklist below with a manager or two. Go rung by rung and stop at the first one where most answers are no. That rung is the one to train for next, using tasks from the team's own week.
A four-rung team check
- Ask. Does most of the team use AI at least weekly for something in their own job? Can people name one task where it reliably helps? If not, start with short, low-stakes practice on everyday tasks like email replies and meeting summaries.
- Brief. When someone hands AI a task, do they include the audience, the constraints and an example of good work? Do they treat the first answer as a draft to revise? If not, train briefing: have people rewrite a thin request into a full one and compare the results side by side.
- Build. Has anyone redesigned a recurring task, such as invoice coding, intake triage or a weekly status report, so that AI handles a defined step? Is a named person checking that step's output on a schedule? If not, pick one workflow, walk through it with the people who do it, and decide together which step to hand over first.
- Share. When a method works, does it end up somewhere the rest of the team can find and reuse it? Could a new hire pick it up in their first month without asking the person who built it? If not, set a regular slot for people to show what they made and agree on where finished templates and procedures live.
A team that answers yes through the third rung is in good shape. The fourth is where the gains stop belonging to individuals and start belonging to the organization, so it is worth checking again every few months as people come and go.
