You're using AI for about five percent of what it could do for you.
Not because you're not smart enough, or technical enough, or because the tools aren't ready. They are. You're using AI for five percent because nobody has shown you what the other ninety-five looks like, and the gap between where you are and where you could be is hidden in plain sight.
Here's what I mean.
If you're a typical knowledge worker in 2026, you probably use ChatGPT or Claude a few times a week. You draft an email and ask it to make it sound less stiff. You paste in a long document and ask for a summary. You ask it to suggest a few subject lines, or rewrite a paragraph, or come up with a list of names for something. Useful. Time-saving. A small upgrade on the old workflow.
Meanwhile, the same tool — the exact same tool you're using to polish emails — could be producing your entire weekly review deck. Or writing the first three sections of your monthly report. Or running the analysis you usually spend Wednesday afternoon on. Not drafting parts of these things. Producing the whole deliverable.
The leap from drafting-help to entire-deliverables is the move most people haven't made yet. It's the difference between AI as a slightly better thesaurus and AI as a colleague who handles the work you don't want to do.
This piece is about how to make that leap.
Why most people haven't crossed yet
Three reasons, in roughly this order.
First, they don't know what's possible. This sounds patronising and it's not meant to be — it's just true. Most people's mental model of AI was set sometime in 2023 or 2024, and the tools have moved much faster than their mental models. They've heard the hype, but the hype is abstract. What they actually know how to do is what they've seen demonstrated to them. And what they've been shown is almost entirely surface-level — drafting, summarising, brainstorming. So they use it for that.
Second, they don't have a method. Even people who suspect the tools can do more get stuck at "okay, so how?" They try once with a big ambitious prompt, get back something that's 70% there but with weird mistakes, conclude the tool can't really do it, and go back to using it for drafts. The 70% problem is real, but it's a method problem, not a tool problem. There's a way to get from 70% to deliverable that takes ten minutes once you know it. Most people don't.
Third — and this one almost nobody says out loud — they're quietly nervous about using AI well. Worried it counts as cheating. Worried their boss will think they're cutting corners. Worried that if they get too good at this, they'll be the first to be automated away. So they hold back. They use AI just enough to be productive, not enough to look suspiciously productive.
I'll address the third one directly because it matters. Using AI well is not cheating. The people in your organisation who are already doing this — and there are some, in every organisation — are not getting in trouble. They're getting promoted. The risk is not that you'll get fired for using AI too well. The risk is that you'll get left behind by people who already are.
The first two — knowing what's possible, and having a method — are what the rest of this piece is for.
The leap, made concrete
Let me show you what crossing looks like in practice, with the most universal example I can think of: the weekly review deck.
Almost every office worker has some version of this. Maybe yours is called the WBR, or the weekly team update, or the Monday morning sync deck, or the leadership readout. The shape is the same. Every week, you pull data from a few sources, identify what changed, write some commentary, build a few slides, and present it.
If you do this seriously, it takes you somewhere between three and six hours. Pulling the data. Updating the charts. Writing the "what's going on" narrative. Formatting. Last-minute panic when something doesn't reconcile. The kind of work that fills an afternoon and leaves you slightly drained even though nothing about it was actually hard.
Here's what the same deck looks like after the leap.
You sit down on Monday morning. You open your AI tool. You give it the raw data — usually a CSV or a couple of spreadsheets. You tell it what last week's deck looked like (you can paste in the previous version, or describe the format). You tell it what's important this week and what's noise. You ask it to produce the deck.
Forty minutes later — including the time you spent reviewing and adjusting — you have a deck. The charts are right. The narrative is reasonable. The formatting matches your template. The three things you flagged as important are highlighted. The minor noise is filtered out.
You didn't draft slides and ask AI to polish them. You described what the deliverable is and what good looks like, gave it the inputs, and let it produce the whole thing. Then you edited. The edit pass is real — usually fifteen, twenty minutes. But it's an edit pass on a finished artifact, not a build from scratch.
The gap between six hours and forty minutes is not magic. It's three moves.
The three moves
This is the whole method. Once you internalise it, it transfers to almost any work product an office worker produces.
Move one: describe the deliverable, not the task.
Most people prompt AI by describing what they want done — "summarise this," "draft a paragraph about X," "list some ideas for Y." This anchors the AI to small tasks and prevents the leap.
The shift is to describe the finished deliverable. Not "summarise this report" but "produce a one-page executive briefing for a busy CEO, with three sections — what changed, what it means, what we should do — each section three sentences, no jargon, formatted as bullet points under each heading." The finished deliverable framing pulls the AI up to work at the level of artifacts, not the level of subtasks.
For the weekly review deck, this looks like: "Produce a ten-slide weekly business review deck. Slide one is a title slide. Slide two is a one-page summary of the week. Slides three through eight are one slide per metric with a chart and three lines of commentary each. Slide nine is risks and watch-items. Slide ten is the ask. Format matches our template."
That description is the unlock. Everything downstream gets easier when the AI knows what it's producing.
Move two: hand over the inputs the way you'd hand them to a colleague.
If you were asking a junior colleague to produce this deck, you wouldn't expect them to read your mind about what data matters. You'd give them the data files, point them at last week's deck so they understand the format, and tell them which metrics are the focus this week.
Do the same for the AI. Paste the data in. Paste last week's deck in. Tell it which two or three metrics are this week's story. Tell it what to ignore. If there's context that matters — "the dip in metric three is because of the holiday, not because of a problem" — tell it that too. This is not over-explaining. This is the work that gets you the right output on the first or second pass instead of the seventh.
Most people skip this step because it feels like a lot of typing. The typing takes four minutes. Skipping it costs you an hour of bad outputs and re-prompts.
Move three: tell it what good looks like.
This is the move almost nobody makes, and it's the difference between 70% outputs and 95% outputs.
After describing the deliverable and handing over the inputs, add one more thing: a short description of what a good version looks like. "The commentary should sound confident but not promotional. Numbers should be rounded to one decimal. The headline insight goes first; the supporting detail goes second. If something looks unusual in the data, flag it explicitly rather than smoothing it over. The tone should match the way I write — direct, not corporate."
You can adjust this once and reuse it forever. The first time you do it for your weekly deck, it takes ten minutes to articulate. From then on, you paste it in and the AI hits the standard.
That's the whole method. Describe the deliverable. Hand over the inputs. Tell it what good looks like. Three moves, applied in order, every time.
Where this transfers
The three moves are a knowledge worker's framework, but the underlying shift — from drafting-help to whole-deliverables — applies almost anywhere.
A homemaker organising five years of saved recipes can describe the deliverable ("a categorised, searchable collection grouped by cuisine and meal type, with the messy duplicates merged and the unclear ones flagged for review"), hand over the inputs (the saved PDFs and screenshots and bookmarks), and tell it what good looks like ("merge anything that's essentially the same recipe even if the wording differs; preserve original sources; flag anything that looks incomplete"). The deliverable is different. The shape of the request is identical.
A college student writing a research paper can describe the deliverable (the paper, with structure and length and citation style), hand over the inputs (the source readings and their own notes), and tell it what good looks like (their professor's standards, the way they actually write, what counts as a good argument in their field). The output is a draft they then edit hard — not a paper to submit, but a starting point hours ahead of where blank-page-writing would have left them.
A freelancer producing client work can describe the deliverable (the report, the analysis, the strategy document), hand over the inputs (the brief, the data, the past examples), and tell it what good looks like (the client's expectations, the freelancer's voice, the format that's worked before). Two hours of work becomes thirty minutes plus an edit pass.
Same three moves. Different work. The leap is the same leap.
What changes after you cross
You stop being someone who uses AI for drafts and become someone who uses AI for deliverables. The visible change is time — work that took an afternoon now takes forty minutes. The invisible change is more interesting.
You start aiming higher. The things you avoided because they were too time-consuming — the proper analysis instead of the quick one, the well-structured document instead of the email, the actual research instead of the gut call — become tractable. Your work product gets better, not just faster. The ceiling on what you can ship in a given week goes up.
You stop being the person who's drowning. The Monday mornings stop being grim. The evenings stop being a question of which deliverable you sacrifice your weekend to. The work fits inside the workday in a way it didn't before.
And — this is the part people don't talk about, but it's the most consequential — you become much harder to replace. Not because you're using AI as a shield. Because you're the person who's already absorbed what AI does, integrated it into how you work, and is now using the extra capacity to do harder things. That's the worker organisations want to keep. The one who's already on the other side of the leap.
What's next
I'm building a series of short videos and pieces on start.naklitechie.com that walk through specific versions of the leap, in detail. The weekly review deck. The monthly report. The research paper. The recipe collection. The client deliverable. Each one is a worked example you can copy the moves from.
Drop your email below and I'll send you the next piece when it's ready. No spam, no funnel of upsells, no AI hype newsletter — just the next walkthrough.
And if, somewhere down the road, you find yourself wanting to build the tool you wish existed — the small app that would make your work easier, the thing you'd buy if anyone made it — there's a separate place for that. You don't need it yet. Start here.