Almost everyone at work uses AI now. Very few people's work has actually changed shape because of it. That second thing is the one worth aiming at.
Using AI is now the entry ticket, not the achievement. What separates people is whether a specific, recurring piece of their work is genuinely done differently. Our framework for that is Tool x Workflows x Behavior, and the multiplication is the point: great access with nowhere to apply it produces nothing, and a perfect workflow you run twice and abandon produces nothing either. This article defines each variable, explains why most people stall on the first one, gives you a scored self-check to find out where you actually are, and ends with a 30-day plan built around one recurring task rather than a tool.
Two people sit on the same finance team. Same job title, same manager, same Copilot licence, issued on the same Tuesday in March. One of them opens it a couple of times a week, asks it to soften an email to a supplier, likes the result, closes the tab. The other one stopped thinking of it as a thing she opens about four months ago, because it sits inside the variance commentary she writes every month end, and if it disappeared she’d have to rebuild three hours of her week from scratch.
Same tool. Same access. Same training session, probably in the same room. Completely different working lives.
The useful distinction isn’t frequency, and it isn’t enthusiasm. It’s whether a specific, recurring piece of your work is now genuinely done a different way. That’s the whole difference, and it’s why “do you use AI?” has stopped being a question worth asking anyone.
Start with a task you already repeat, not a tool you were just given.
Something that surprised me when I went looking for numbers on this: Microsoft’s 2026 Work Trend Index screened out anyone who doesn’t use generative AI at work before it even started. Everybody in it, all 20,000 of them, was already using AI. And within that group, only 19% landed in what Microsoft calls the Frontier zone, where individual capability and the organisation’s readiness to support it are both high.[1]
So the population that “uses AI” and the population whose work has changed are not remotely the same population. Using it is the entry ticket. It stopped being a differentiator somewhere around the middle of last year.
Here’s what’s true of the second person on that finance team, and not the first:
None of those four things is about the tool. That’s the part that takes a while to sink in.
We’ve been teaching this in workshops for a while now, and it started as a way of answering a question we kept getting from people who felt like they were doing everything right and getting nothing back. They’d done the training. They had the licence. They’d read the prompt guides. And their week looked exactly the same as it did before.
The model we landed on is deliberately small:
Future Factors’ own framework, developed from our workshop practice. It is not derived from published research, and the multiplication is a way of describing what we see rather than a measured relationship.
The reason it’s written as a multiplication and not a list is that the three don’t add up, they gate each other. A list would let you feel good about scoring two out of three. Multiplication won’t.
Think about what a near-zero in each position actually looks like in practice.
Near-zero Tool: you’re trying to get a consumer chatbot on a free tier, with no file access, to reconcile figures across four spreadsheets it can’t open. Your workflow thinking might be excellent and your discipline might be perfect. The output is still wrong, so nothing changes.
Near-zero Workflows: you have a genuinely capable model, you’re confident with it, and you use it for whatever happens to be in front of you. Different task each time, no repetition, no accumulating skill. In our workshops this is the one we run into most often, and it feels productive, which is what makes it hard to spot.
Near-zero Behavior: you designed a genuinely good workflow. It worked. Then Q3 planning hit, you fell back to the old way for a fortnight, and you never came back. The design was sound. It just didn’t survive contact with a busy month.
I want to be careful here, because “if one variable is zero the whole thing is zero” is a tidier sentence than reality supports. In practice you rarely see a true zero. What you see is a 3 out of 10 somewhere, quietly capping everything else, and the person keeps working on the variable they’re already strong at because that’s the one that feels productive. The multiplication is a way of noticing that. It isn’t a formula you should try to calculate.
Abstract definitions are where frameworks go to die, so here’s the same three variables filled in for one real-shaped job: a demand generation manager at a mid-size B2B software company, roughly 40 people in the business.
| Variable | What it means | What it looks like for her, specifically |
|---|---|---|
| Tool | Access to something capable enough, plus a working sense of where it’s weak | Copilot through work, because the source material lives in SharePoint. She knows it’s good at reading and comparing what’s already there and unreliable at inventing a number, so she never asks it for a figure it hasn’t been given. |
| Workflows | A recurring task with a named input and a named output | Input: the six competitor pricing pages she saved last Friday, plus last week’s version of the note. Output: a Monday morning note listing only what changed, with links. Nothing else. |
| Behavior | Repetition that survives a bad week | Friday 4pm, saves the pages. Monday 9am, runs it, reads it properly, deletes the two lines that are noise, sends it. Eleven weeks running, including the week of the conference. |
A composite example of the shape this takes, not a named client. The point is the level of specificity: every cell names a thing, a time, or a limit.
Look at the Workflows row again. The input and output are both boring and both extremely specific. That’s not a coincidence. The workflows that stick are almost always the ones you could hand to a competent new starter with a two-sentence brief.
A workflow AI touches once is a demo. A workflow it touches every week is a workflow.
The advice does change by role, and it’s worth saying which way. For a content marketer, the Workflows variable is usually the easy one, because the work is already text-shaped and repetitive, and the hard variable is Tool, specifically knowing when the output is fluent and wrong. For a financial analyst it’s the reverse: the tool question is nearly settled the moment the data is in a spreadsheet, and the hard part is finding a piece of the month that repeats in the same shape. For an HR business partner the binding constraint is usually neither, it’s confidentiality, which quietly rules out most of the interesting workflows until someone has answered what can go into the tool at all.
There’s a Microsoft telemetry finding that reframed this for me. In a privacy-preserving analysis of about 105,000 Copilot conversations from one week in February 2026, 49% of them mapped to cognitive work: analysing, evaluating, problem-solving, thinking through. Only 17% mapped to producing output.[1] Which is a long way from how most people describe what AI is for.
If that pattern holds for you, the workflows worth designing aren’t the ones where you want text produced. They’re the ones where you’re currently doing tedious reading, comparing, or checking before you can make a call.
The stall is so consistent that you can almost predict the week it happens. Someone gets a licence, has a genuinely impressive first fortnight, tries a dozen different things, tells a colleague it’s changed everything, and then three months later uses it about as much as they use the advanced features of Excel.
Nothing went wrong. That’s the confusing part. The tool worked, the person was willing, the outputs were fine. What never happened was the second variable.
Tool is the only one of the three that someone else can hand you. Workflows and Behavior both require you to look at your own week and change something about it, and that’s a different kind of effort entirely. It’s also unglamorous. Nobody feels clever writing down that the thing they do every Thursday afternoon takes ninety minutes and mostly involves reading.
The organisation around you matters more than the self-improvement framing usually admits. Microsoft’s analysis tested 29 factors against how much real impact people reported getting from AI, and found that organisational factors like culture, manager support and talent practices accounted for more than twice as much as individual mindset and behaviour, 67% against 32%.[1] The report is careful to say these are statistical associations rather than causes, and I’d hold them the same way. Everything in it is self-reported by the same person at the same moment.
Still, one figure is worth sitting with. Around 10% of people land in what Microsoft calls blocked agency: they’ve genuinely built the skills and don’t have the systems around them to apply them.[1] If that’s you, the constraint isn’t your capability, and more courses won’t touch it. There’s just nowhere for the capability to land.
Which is a real limit on how far an individual can get alone, and worth being honest about rather than pretending grit fixes it. Three things you can do inside your own control anyway:
Self-assessment on this is unusually unreliable, because confidence with a tool feels a lot like capability with the work, and the two come apart quickly. So rather than asking you to rate yourself out of ten on something vague, here are nine specific questions. Three per variable. Answer them about one task, not about your working life in general, or the whole thing turns to mush.
| Variable | Ask yourself | Score 0 or 1 each |
|---|---|---|
| Tool | Do I have access to something that can actually reach the material this task needs? | ___ |
| Can I name one thing this tool is unreliable at, from my own experience rather than an article? | ___ | |
| Have I checked an output against the source at least once in the last month? | ___ | |
| Workflows | Can I state the input and the output of this task in one sentence each? | ___ |
| Does this task recur on a schedule I could name out loud? | ___ | |
| Do I know what a good version looks like well enough to reject a bad one? | ___ | |
| Behavior | Have I done it this way at least four times? | ___ |
| Did I do it this way during my last genuinely busy week? | ___ | |
| Would I notice within a week if I stopped? | ___ |
Score each variable out of 3, then multiply the three subscores. Interpretation bands below. A scoring aid, not a validated instrument.
Multiply your three subscores together. The maximum is 27, and the distribution is deliberately brutal.
You’re AI-powered when a piece of your work would visibly change if you stopped.
Question nine is the one that does most of the work in that list, and it’s borrowed from a habit the most advanced users in Microsoft’s research seem to have independently. Compared with everyone else in the survey, they were much more likely to say they deliberately pause before starting a piece of work to decide what should be done by AI and what shouldn’t, 53% against 33%.[1] They were also more likely to say they intentionally do some work without AI to keep their own skills sharp. Read as caution, that looks like hesitancy. Read properly, it’s the same instinct as knowing which of your colleagues to give which job to.
One caveat on scoring, because people get this wrong in workshops and then get discouraged. A 0 on your first attempt is the normal result, not a bad one. Most people’s honest first score is a 0 because they’ve never named a single task, which means the fix is thirty minutes of writing rather than months of study.
Almost nobody arrives here by accident. The people who get there have usually done something quite small and quite deliberate, and the thing they did was almost always to pick one boring task and refuse to be distracted by a better one for about a month.
Before the plan, the part that gets skipped: deciding what stays yours. Not as a disclaimer, as an actual design decision made in advance, because the moment a workflow gets useful is exactly the moment the boundary starts drifting.
| What AI does | What you still own | How it gets checked |
|---|---|---|
| Reads the six competitor pages and last week’s note, returns only the differences with links | Deciding which changes matter enough to tell anyone about, and what they mean for next quarter’s positioning | You open two of the six source links every week before sending. Not the ones that look wrong. Two at random. |
| Drafts the month-end variance commentary from the exported P&L | The explanation of why the variance happened, which lives in conversations the model wasn’t in | Every figure in the draft traced back to the export before it goes anywhere. Every month, no exceptions. |
| Groups 200 open-text customer comments into themes with example quotes | Which themes are signal and which are three loud people, and what you’re going to do about it | You read 20 raw comments yourself first, then compare against the themes it produced. |
The three-column split we recommend for any workflow where AI touches real work. Fill in your own row before the first run, not after something goes wrong.
Now the plan. This is one person’s, filled in, so you can see the level of detail that makes it work rather than a blank template you’d have to invent content for.
Write down three tasks you do at least twice a month. Pick the most boring one. She picked the Monday competitor note over the campaign copy, which felt like the wrong choice at the time.
Write the input and the output in one sentence each, plus the three-column split above. Twenty minutes, and it’s the step that decides whether the rest works.
Run it twice. Both outputs will be mediocre. Keep the prompt that produced the less bad one and change exactly one thing about it.
Run it in the slot, on the day, even though it’s still slower than doing it by hand. This is the week most people quit, and it’s the week the behaviour is actually being built.
Show one other person, and answer the persistence question honestly: would you still be doing this in a month with nobody asking?
An illustration of one person’s thirty days, not a methodology. The dates are the useful part; the specific task should be yours.
On measuring whether it worked, resist the pull toward counting. How often you opened the tool tells you almost nothing. We look at three things in this order, and only the third one matters in the end:
Plenty of people can and do get through this on their own, particularly if they’re stubborn and already have a task in mind. What structured training buys you is mostly speed and spread: someone who has watched a hundred people pick the wrong first task will stop you doing it, and a team that goes through it together ends up with shared standards rather than one enthusiastic person and eleven onlookers. That second part is the one that’s genuinely hard to do alone. If that’s the version you want, our corporate workshops are built around exactly this: one real recurring workflow per person, designed in the room.
Either way, the next move is the same and it’s smaller than you’d like. Pick the task you find most boring. Write the input and the output. Run it on Friday.
Someone whose actual work has changed shape because of AI, rather than someone who uses AI frequently. The practical test is whether a specific recurring task in their week is genuinely done a different way now, and whether they’d notice within about a week if the tool stopped working. Frequency of use is a poor indicator, because it’s entirely possible to open an AI tool daily for one-off scattered questions and have your job look identical after a year. The distinguishing features are that they can name the task, state its input and output, describe what a good output looks like well enough to reject a bad one, and keep doing it during a busy week.
It’s our model for what has to be present at once for someone to become AI-powered: a capable tool, a real recurring workflow it fits into, and behaviour repeated often enough to stick. We write it as a multiplication rather than a list because the three variables gate each other rather than adding up, so being strong on two of the three leaves you roughly where you started. Worth being clear about what it is: this is our own framework developed from workshop practice, not a research finding, and the multiplication describes a pattern we see rather than a measured relationship. Treat it as a way of noticing which variable is capping you, not as a formula to calculate.
On its own, no, and this is the most common version of the stall. Daily use with no repeating workflow behind it means every session starts from zero: new task, new prompt, no accumulating standard for what good looks like. It feels productive, which is exactly what makes it hard to spot. The tell is whether you can name the task. If your real answer to “what do you use it for?” is “writing” or “research” rather than something like “the first draft of the month-end variance commentary from the exported P&L,” the Workflows variable is near zero however impressive the daily volume looks.
Both, but they’re different problems and the organisational one is bigger than most self-improvement framing admits. Microsoft’s 2026 Work Trend Index tested 29 factors against reported AI impact and found organisational factors like culture, manager support and talent practices accounted for more than twice what individual mindset and behaviour did. Around 10% of the AI users surveyed fell into what the report calls blocked agency: strong personal capability, no system around them to apply it. If that describes your situation, more training won’t help, because the constraint isn’t capability. The team version of this framework runs at the level of shared standards and documented handoffs rather than personal habit.
For one workflow, roughly a month of deliberate effort, and the limiting factor is almost never learning the tool. Weeks one and two are cheap: naming the task and getting a usable first version takes a few hours in total. Week three is where it actually gets decided, because that’s when the new way is still slower than the old way and there’s no external pressure to continue. Most people who stall, stall there. Beyond the first workflow it gets faster, because the hard-won part is the habit of looking at your own week and seeing tasks with inputs and outputs, and that transfers.
The framework in this article is Future Factors’ own, developed from our workshop practice with non-technical teams, and it’s labelled that way in the text rather than dressed up as a research finding. Every external figure quoted comes from the Microsoft Work Trend Index 2026 annual report itself, read on 26 August 2026, rather than from press coverage of it, because several of the widely-circulated summaries conflate two different measures: the 19% who sit in the Frontier readiness zone and the 16% classified as Frontier Professionals are different groups measured different ways. Two figures from the same report that would have fitted neatly here were deliberately left out. Both are real, and both only confirm something the reader already believes, which isn’t a good enough reason to spend a paragraph on a number. Where the report says its own analysis shows statistical association rather than causation, that caveat is repeated here rather than quietly dropped.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.