Fifteen small habits, four families, and a ten-minute way to practise each one on work you already have.
Fifteen human-AI collaboration skills in four families, each with a definition, a daily-work example and a ten-minute drill. Score yourself on what you did in the last two weeks, pick three skills by role, and log one drill a week. Use, persistence and impact come from scoring again a month later.
People who work well with AI usually aren’t using a better tool. They’ve picked up a handful of small habits that nobody ever named, so it gets called “being good with AI” and can’t be taught.
Two analysts ask the same assistant for a summary of a competitor’s pricing changes. The first reads it, finds it sensible and sends it to the team. The second opens two of the pages the summary refers to, sees that one says something narrower than the summary claims, and fixes the sentence before it goes anywhere. The tool was identical. The difference was a habit: check one claim against its source before forwarding.
This guide names fifteen habits like that and sorts them into four families: setting the work up, judging what comes back, working with the people around it, and owning the outcome. Each skill has a one-line definition, what it looks like on an ordinary day, and a way to practise it in about ten minutes. My view is that these skills are learned by doing small, real reps, not by reading a list, so each drill uses work you already have.
This list comes from how we teach non-technical teams. It’s our judgment, not a validated competency model or a research finding, and other lists will slice it differently. For the narrower argument about which human skills gain value as AI gets cheaper, see our piece on human skills in the AI era. This one is the checklist.
Practise one skill at a time, for ten minutes, on work you already have.
Most of the effort that decides how a session goes happens before the first prompt. A marketer who types “write a launch email” gets a competent, generic email and concludes the tool is average. The sentence didn’t say who the email was for, what the launch was or what the reader should do next.
Briefing is the skill people point to when they say “prompting”, and it’s one of fifteen. Here is a filled-in brief for a recruiter, with the parts that usually get left out.
| I’m a recruiter writing to candidates who reached the final interview and weren’t selected. Audience: people who gave us several hours of their time, so the tone is warm and direct, not corporate. Purpose: tell them the decision, thank them specifically, and invite them to apply for future roles. Length: under 130 words. Constraints: no reasons for the decision, no promises about future openings, no phrases like “unfortunately we regret”. Context: the role was a customer success lead and the team chose someone with more experience in our sector. Before you draft, ask me up to three questions if anything is missing. |
A Future Factors example. Swap in your own role, audience and constraints. The last sentence asks the assistant to ask you, which is the quickest way to find what you left out.
Have a look at the last line of that brief. Asking the assistant to question you first is a small move that tends to surface the context you forgot to give.
AI output arrives in the same polished voice whether it’s right or wrong, so the checking comes from you. A draft that reads well is not evidence that it’s correct.
Skill 6 has the clearest research behind it. In a preregistered experiment with 758 knowledge workers, those using AI on tasks inside its capability finished 12.2% more tasks and 25.1% faster, and on one complex task outside it they were 19% less likely to produce a correct solution.[1] It’s one study on a specific set of tasks, so don’t read it as a number for your own work. The useful part is the shape: the boundary between “good at this” and “confidently wrong” isn’t visible in the output, so you learn it task by task.
| Here is the recommendation I was about to accept: [paste it]. Give me the three strongest objections to it. For each one, say what would have to be true for the objection to be right. Then list any claim in the recommendation that you can’t support from the material I gave you, and mark it “unsupported”. |
A Future Factors prompt. Use it after the answer looks finished, not before. The “unsupported” label gives you a short list of things to check by hand.
Our guide to critical thinking for AI has a longer set of questions.
Much of AI skill is invisible because it happens in a private chat window. These three skills make the work visible enough for colleagues to trust it and learn from it.
A manager who receives a report sees a clean document and can’t tell whether it was typed, drafted or pasted. That’s not a reason for suspicion. It’s a reason for a shared habit of saying so, so the manager’s attention goes to the parts that need it. An L&D designer who finds a brief that reliably produces a decent first draft of a quiz has made something the whole team could use, and it stays in her chat history unless she shares it.
The last family holds up as the tools improve. Someone still answers for the work, decides what information is safe to share, and decides whether AI belongs in the task at all.
Name the person who answers for the output before the assistant starts drafting.
Here is what that looks like across four roles. The pattern is the same in each row: the assistant prepares, the person decides, and there’s a specific check with a specific person doing it.
| Role | What AI does | What the human still owns | How it gets checked |
|---|---|---|---|
| Analyst | Drafts a summary of a data export and suggests possible explanations for a trend | Which explanation is credible, and what goes into the recommendation | The analyst traces headline figures to the export. A senior colleague asks how it was produced |
| Recruiter | Drafts outreach and rejection messages from a brief | The decision about each candidate and the tone of the message | The recruiter reads every message before it sends. The assistant never judges suitability |
| Marketer | Produces variants of a campaign email from an approved message | Which claims are true, and the final choice of message | The marketer checks each claim against the product page. Legal reviews any number |
| L&D designer | Turns source material into a draft outline and practice questions | What learners need, and whether the questions test the right thing | A subject expert reviews the questions, after the designer tries them on a few learners |
A Future Factors working split. Adjust it to your organisation’s policy and to what your tools can reach. For the full approach to review, see our guide to writing human-in-the-loop guidelines and the manager’s guide to managing the responsible use of AI.
Skill 14 is the one people resist, because it can feel like admitting the tool isn’t useful. Deciding a task isn’t for AI is a good outcome.
The check below takes ten minutes and asks what you did in the last two weeks, not how confident you feel. That’s deliberate. In a survey of 319 knowledge workers, higher confidence in generative AI was associated with less critical thinking, while higher confidence in their own ability to do the task was associated with more.[2] The data was self-reported, so treat it as a pointer. It suggests “I feel comfortable with AI” is a weak test and “I did this thing” a better one.
| # | Skill | In the last two weeks… |
|---|---|---|
| 1 | Choosing the task | I decided on purpose which tasks to hand over and which to keep. |
| 2 | Briefing | My last request named the audience, purpose, length and one thing to avoid. |
| 3 | Giving context | I pasted in the background the assistant couldn’t have known. |
| 4 | Working in steps | I gave specific feedback on a draft instead of starting again. |
| 5 | Checking against a source | I opened the source for at least one claim that mattered. |
| 6 | Calibrating trust | I trusted the output less on a task where I knew it was weak. |
| 7 | Challenging the answer | I asked for the sharpest objection before accepting a recommendation. |
| 8 | Holding my standard | I changed a draft until I’d put my name on it, and could say what I changed. |
| 9 | Saying how AI was used | I told a reader what the assistant did and what I checked. |
| 10 | Reviewing others’ work | I asked a colleague how an AI-assisted deliverable was made. |
| 11 | Sharing what works | I passed on a brief or a check to someone else. |
| 12 | Naming the owner | I could name who answers for each AI-assisted piece of work. |
| 13 | Judging what goes in | I removed sensitive details or used the approved tool. |
| 14 | Knowing when to stop | I kept at least one task away from AI, and knew why. |
| 15 | Adjusting how I work | I reviewed one task and changed something about how I do it. |
Score one point per yes. 0 to 5: getting started, so pick three skills and run one drill each. 6 to 10: you have the basics, and the gaps are often in skills 9 to 15. 11 to 15: working well, so share it, because skill 11 is how a team catches up. A Future Factors working check, not a validated assessment.
Today’s score is use. Score again in a month for persistence, whether you’re still doing these things unprompted. Whether the work is better or faster for it is impact.
Nobody needs all fifteen at once. A recruiter and an analyst share the list but not the starting point, because a missed habit does different damage in each job. Pick the three that would hurt most to skip this month.
| Role | Start with | Why these | First ten minutes |
|---|---|---|---|
| Analyst | 5 Checking, 7 Challenging, 12 Naming the owner | Numbers travel further than the caveats attached to them | Trace the three most important figures in your latest summary to the source |
| Manager | 10 Reviewing others’ work, 9 Saying how AI was used, 14 Knowing when to stop | Your team copies what you ask about and what you accept | Write the two review questions you’ll ask of the next AI-assisted deliverable |
| Recruiter | 13 Judging what goes in, 14 Knowing when to stop, 8 Holding your standard | Candidate data is sensitive, and decisions about people stay with people | Rewrite one recent prompt with identifying details removed |
| Marketer | 2 Briefing, 8 Holding your standard, 5 Checking | Volume is cheap, and a wrong claim in a campaign is costly | Rewrite last week’s brief with audience, purpose, length and one thing to avoid |
| L&D designer | 3 Giving context, 11 Sharing what works, 7 Challenging | A good brief for one course helps the whole team, if it’s shared | Add two paragraphs of real learner context to a brief and compare the draft |
A Future Factors suggestion by role, based on where a missed habit costs most. If your role isn’t listed, ask which of your outputs travel furthest without anyone checking them, and start with the skills that sit closest to those.
For the team-level view, see what AI skills employees should be trained in.
A drill done once is a nice afternoon. Done weekly, it becomes a habit. Here is a log a recruiter might keep over four weeks.
| Week | Skill | Task I used it on | What I did | Keep, change or drop |
|---|---|---|---|---|
| 1 | 13 Judging what goes in | Outreach message to a passive candidate | Removed the candidate’s current employer and salary from the prompt | Keep. The draft was just as good |
| 2 | 14 Knowing when to stop | Feedback after a final interview | Wrote the feedback myself and used AI only to tighten it | Keep, and add it to my “not for AI” list |
| 3 | 8 Holding your standard | Job advert for a customer success lead | Edited the draft down and noted I often cut the opening line | Change. Add “no opening slogan” to my brief |
| 4 | 13 Judging what goes in | A second outreach message | Did it without being reminded | Keep. This one has become routine |
A Future Factors example. The last column turns a drill into a decision, and the week 4 entry is the persistence signal.
| Here are my notes from this week: [paste what you used AI for, what you changed, what went wrong]. Group them into things to keep, change and stop. For each “change”, suggest one specific adjustment to my brief or my checking step. Don’t invent anything I didn’t write down. |
A Future Factors prompt. Paste your own notes, since the assistant can’t remember your week for you.
You can build all of this on your own, and many people do. When a group of people are building the same habits at once, working with someone who has seen where these stall can shorten the learning curve, build a shared standard across a team rather than in one enthusiastic person, and help people apply the habits to their real work instead of generic examples. That is the work our corporate workshops are built around.
This week, run the self-check, pick your three skills and put three ten-minute slots in your diary.
Start with the first slot today.
No. Most people already do a few without having named them. The self-check shows which, and the role table suggests three to start with. Treat the rest as a menu.
Skills 2 to 4 cover what people usually mean by prompting, and they’re a fifth of the list. The rest is about what happens before and after the prompt: choosing the task, checking the result, and deciding who is responsible.
Use work that is already finished or low-stakes, such as last month’s report, a draft that went out last week or an internal note. Drill 13 is the exception: do it first, because it tells you what’s safe to paste into the others.
In our workshops, skills 9 and 14 tend to feel the most awkward: saying how you used AI, and deciding a task isn’t for it. Both can feel like admitting weakness, and both make colleagues more willing to trust AI-assisted work.
Use it as a conversation, not a scorecard. It asks about behavior in the last two weeks, which is fair to discuss in a one-to-one. Turned into a rating, it invites people to tick yes. For a team-level view, see our AI skills matrix guide.
I read two research sources on October 26, 2026: the Harvard Business School working paper on the jagged technological frontier (Dell’Acqua and colleagues, 2023) and the Microsoft Research and CHI 2025 survey on generative AI and critical thinking (Lee and colleagues). The first is a field experiment on specific tasks, and the second is self-reported survey data, so neither says how much any of these habits will change your own results. The fifteen skills, the four families, the self-check, the role table and the log are Future Factors working methods, not research findings or a validated competency model. I did not test any tool hands-on for this article, and the scenes are illustrations, not client stories.