← All exercises

Chapter 2 · Demonstration · 5 minutes

Start with your problem, not with AI

From Actually Using AI

What this shows

Most people meet AI the same way: they open it up wanting to try it on something, and they ask for something to try. “What’s a good prompt for getting started?” “What can you help me with?” “Give me ideas for how to use you.”

You’re about to watch what happens when the same person, on the same evening, asks two different opening questions. The first is the curious one — what could AI do for me? The second is the specific one — here’s the thing I’ve actually been carrying around. This isn’t about setting AI up beforehand or pasting in any clever instructions; it’s about the question you bring on the day. Same default AI, same person, two starting points. The difference between what comes back tells you everything you need to know about how to start.

The demonstration

Maria is at her kitchen table on a Wednesday evening. Something has been sitting at the back of her mind for months: her closest friend Sara. The two of them met in their NCT group fifteen years ago and used to talk every week. Over the last year, the rhythm has just dropped. Fewer texts. No spontaneous catch-ups. The WhatsApp threads going quiet. Nothing happened that Maria can identify — no row, no incident — but the friendship has thinned, and Maria has noticed she’s the one initiating now, and increasingly not bothering.

She doesn’t open AI to think about Sara. That feels like the wrong tool for the job. She opens AI because she’s seen people at work using it and thought she’d have a look herself.

Take 1 — looking for something to try

Here’s what comes back:

AI: That’s great that you want to explore! AI is incredibly versatile. Here are some popular ways people are using AI:

1. Drafting and editing. Use AI to help write emails, polish your writing, or generate first drafts of documents.

2. Research and summarising. Paste in long articles and ask for a quick summary, or ask AI to explain unfamiliar concepts in simple terms.

3. Planning and organising. AI can help you plan trips, organise tasks, brainstorm ideas, or think through decisions.

4. Learning and skill-building. Ask AI to teach you about a topic you’re curious about — it’s like having a knowledgeable tutor available any time.

5. Work-specific tasks. Especially for someone in HR, AI can help with drafting policies, summarising employee feedback, preparing for difficult conversations, and explaining HR concepts.

What kind of tasks interest you most? I’d be happy to go deeper on any of these!

Maria reads it. None of it is wrong. None of it is useful either. It’s a tour of AI’s capabilities — a brochure for everything she could do, with no purchase on what she should do. She closes the conversation thinking I’ll have a proper look at this another time. She doesn’t.

The thing actually on her mind isn’t a use case for AI. It’s Sara. But she didn’t bring it because AI doesn’t feel like the right tool for that.

Take 2 — bringing the actual thing

Rewind. Same Wednesday evening, same kitchen table. This time Maria pauses before typing. She thinks about what’s actually been at the back of her mind. Not what AI is good at — what she’s been turning over. Sara surfaces immediately.

She hesitates. This feels strange to type. But she types it anyway.

Here’s what comes back:

AI: This is a useful kind of question to bring, actually — these situations are often clearer when you say them out loud than when you carry them around. Let me try to help you think through it rather than tell you what to do.

A few things stand out from how you’ve described it. You said nothing has happened, but you’ve also described a pattern that’s continued for about a year — fewer texts, you initiating, threads going quiet. That’s not nothing. It might not be a something in the dramatic sense, but it’s a sustained drift, and sustained drifts usually have a cause even when there’s no incident. The two most common ones in friendships at this life stage are circumstantial (different priorities, different work pressures, kids at different ages) and emotional (something one person hasn’t said). It’s worth being honest with yourself about which one feels more likely. You probably already have an instinct.

On whether to do something: I’d gently push back on the framing that the choices are “say I miss you” or “pretend it’s fine”. There’s a wider range. You could send a specific, light message that creates an opening without putting weight on it — a memory, a question about her kids, something only you would think to ask her. If she takes it up, you’ll know the friendship has more in it than the recent pattern suggests. If she doesn’t, you’ll have a clearer answer about where things actually are.

One thing worth checking honestly: how much of the energy you have for this is about repairing the friendship versus understanding why it changed? They’re not the same. If it’s the first, the move is to put effort in and see what comes back. If it’s the second, you might already have your answer — and the work might be about accepting the friendship has shifted rather than getting it back to where it was. Both are reasonable, but they ask for different things from you.

I should be honest that I don’t know Sara, and I don’t know what’s going on in her life that you might not be seeing. The pattern you’ve described is consistent with either of the two causes I mentioned, and there are versions of this where she’s quietly going through something and the right move is to be more present, not less. So whatever you do, leaving some kind of opening that’s easy for her to use seems worth doing.

Maria reads it twice. Then she sits with it.

She didn’t get told what to do. She didn’t get a five-step plan, or a generic “prioritise the friendships that matter” line. She got two things she didn’t have when she opened the tab: a reframe she hadn’t quite let herself see (am I trying to repair this, or trying to understand why it changed?), and one specific small move that doesn’t require her to commit to either direction. She also got an honest acknowledgement that AI doesn’t know Sara and can’t see the whole picture — which is true, and useful to have stated.

She closes the tab. On Saturday, she sends Sara a photo of something she knows Sara will recognise from years ago, with no caption. Just the photo. No pressure, no demand for a reply. Easy to use as an opening if Sara wants one.

The point

Both responses came from the same AI. The first was a brochure. The second was movement on something Maria had been carrying alone.

What changed wasn’t the AI, or even the length of the message. What changed was where Maria started. The first message was about AI. The second message was about Maria’s actual situation, with the part that felt awkward to type included rather than smoothed over.

AI can’t do anything useful with the first kind of message — it doesn’t know enough about you to be specific. The second kind of message gives it something to work with: a real situation, a real stuck point, a real ask. “How would I think about this” is a much more answerable question than “what should I do with you”.

Try it yourself

Don’t open AI looking for something to try it on. Open it because there’s something on your mind you’ve been carrying around anyway — at the kitchen table, in the shower, on the walk to the train. Something a thoughtful friend would actually have an opinion about if they knew the situation.

Bring that. Type it the way you’d say it out loud, including the part that feels awkward to say to a machine. Then ask for help thinking about it — not for an answer.

The conversation that follows is different from anything else AI does, and different from any AI conversation you’ve had if you’ve only ever asked it to help with work things.