Contrast settings

Ways of Working with AI

The market for AI advice feels completely saturated, and by the time you’ve got a partial grip on what AI can do, it’s moved on again. Here are some ideas I think might be useful if you’re exploring what’s possible now, and whatever comes next.

  1. Card 1 of 7: Three ways to work with AI

    Talk, Do and Build drawn as three strands of one helix, each joined to it by its own line: two voices taking turns, a task track with two steps ticked, and a small tool with a chart.

    Talk. Explore a decision. Test an idea against strong objections.

    Do. Bring together research, compare documents and turn findings into something you can use.

    Build. Create personalised software tools, interactive dashboards and simulations.

    These tools don’t just answer questions any more.

  2. Card 2 of 7: Build a way to explore the answer.

    An illustrative chart of ticket price against attendance, with costs held constant. A break-even curve separates profit from loss. From the current plan, fewer people moves the event into loss; a cheaper ticket still sits above the break-even curve.

    Suppose you want to know whether an event will break even. Ask AI to build a small interactive model. Change attendance, ticket price and costs to see what makes the event viable. Keep the calculations visible.

    Check the calculations, then explore the possibilities with your team.

  3. Card 3 of 7: Let it work while you sleep.

    Explore, then develop, then test; the findings return to explore for the next round, three times, and the only way out leads to a handover in the morning.

    A single instruction can set off a productive loop.

    “Tonight: research what would strengthen this idea, develop it, test it against our agreed brief. Three rounds. Leave me the result, what changed, and what needs my decision.”

    Come back to something you can take further.

    Use a tool that supports scheduled tasks.

  4. Card 4 of 7: The part you edit out of your prompt may be the context it needs.

    A voiceprint of someone explaining aloud, with the passage they would leave out of a written request lit.

    “We tried that last year and nobody came.”

    You might say that to a colleague, but leave it out of a neatly written request.

    Try explaining it aloud, including what you’ve tried and why you’re hesitating. Ask AI to organise it afterwards. Check that the clearer version still says what you meant.

    Wispr Flow is one option for dictating into the text box you’re already using.

  5. Card 5 of 7: Ask your AI what it has started taking for granted.

    A line of ten messages drifting from the first to the tenth.

    You mention that time is tight.

    Ten messages later, every suggestion favours speed over quality.

    You never agreed to that trade-off. In a longer conversation, pause and ask:

    “What have you inferred about what I want during this conversation? Which of those assumptions have I actually confirmed?”

    If it has misunderstood your priorities, say so: “Time is tight, but I don’t want to sacrifice quality for speed.”

  6. Card 6 of 7: Sometimes you need something to disagree with.

    Three rough drafts drawn as loose contour sketches, one of them polished and certain.

    You’re not sure how to introduce an idea to colleagues. Ask for three rough approaches.

    One comes back as a persuasive pitch.

    Your reaction: “Hang on. We haven’t decided whether to do this. I want to open a discussion.”

    That’s useful. You’ve worked out something the brief was missing. Explain your objection and keep talking.

  7. Card 7 of 7: Make the work portable.

    Dozens of messages, with four useful moments gathered into a file called handover.md, under the headings Agreed, Why, Still open and Next.

    Bank the progress you’ve made: what you’ve agreed, why, and what’s still unresolved. Keep decisions separate from ideas you’re still exploring.

    Check it and save it as a Markdown file. You now have a record to build on, in this conversation or another, even on a different platform.

    Markdown is a plain-text file with simple headings and lists, usually ending in .md. Easy for you and AI tools to read, edit and share.

In brief

  1. 1 Three ways to work with AI

    Talking, doing and building are equally useful, and one piece of work can move between all three.

  2. 2 Build a way to explore the answer.

    Ask for a small model you can change yourself, with the calculations visible.

  3. 3 Let it work while you sleep.

    One clear instruction can run a few rounds overnight and leave you something to decide on.

  4. 4 The part you edit out of your prompt may be the context it needs.

    Explaining aloud keeps the experience and hesitation a tidy request can lose.

  5. 5 Ask your AI what it has started taking for granted.

    Check which priorities it has assumed, and correct the ones you never agreed to.

  6. 6 Sometimes you need something to disagree with.

    Your reaction to an imperfect draft can show you what you actually want.

  7. 7 Make the work portable.

    Keep a short record of decisions, reasons and open questions to carry into the next conversation.

Download PDFPDF, 125 KB

Back to Ideas