When Harry Met AI

You wouldn’t let a teenager run your business unsupervised, would you? Of course not. Not even if they watched every YouTube tutorial on spreadsheets and negotiation tactics. Teenagers are brilliant, frustrating, full of potential, and in constant need of guidance.

Oddly enough, the same is true of AI.

So rather than bang on about models, tokens, algorithms and hallucinations, let’s talk about Harry. He’s 15. Thinks he’s ready to take on the world. Spend a day with Harry, and you’ll have an idea of what you need to know about managing AI in the workplace.

7:10am - The Wake-Up Call

Mum yells upstairs: “Harry! You’ve got football! You’re going to be late!”

“I know!” comes the reply. He didn’t. He’d forgotten.

AI, like Harry, will confidently say it’s got something under control. It’ll give you an answer with absolute certainty, even if that answer is completely wrong. You can't take either of them at face value. That’s why setting clear expectations is essential. Define what “ready” looks like. Spell out what success means. If you’re vague, you’ll get vague results.

7:45am - Breakfast

Harry’s in the kitchen half-dressed, one sock on, brushing crumbs off his homework. Mum’s already packed his kit bag and laid out everything he needs.

Teenagers function best with structure and boundaries. Left to their own devices, they’ll improvise. AI’s no different. If you just plug it in and let it run wild, don’t act surprised when it invents facts or rewrites your pricing model. Like Harry, it needs a defined routine and limitations to avoid mess.

9:00am - Football Practice

Harry tries to show off, attempting nutmegs and scissor kicks. Coach pulls him aside: “Stop messing about. Keep it simple.”

Harry grumbles but nods. Coach sticks with him, correcting his positioning and giving him another chance to try.

AI needs a responsible adult too. Someone to review what it’s doing, step in when it’s going off-piste, and help it improve. Just as Harry gets guidance on the pitch, AI needs human supervision. You can’t just let it run blind and hope for the best.

Harry’s stumbling, flushed, and short-tempered. Misses an easy goal, argues with a teammate, nearly throws up on the pitch. The coach looks concerned. Mum steps in. “You’re not right, are you?”

Harry shouts something inappropriate at Mum

Turns out Harry’s running a temperature. No wonder he’s not making sense.

Sometimes it’s not just youthful chaos. Sometimes something’s actually wrong. AI systems can go off track. They pick up odd patterns, start behaving unpredictably, or make decisions that just feel wrong. But unless you’re watching closely, you won’t spot it. This is why familiarity matters. You need to know what healthy output looks like. And when it starts sounding weird, like Harry shouting something at Mum about kangaroos, it’s time to take it offline and call in the expert.

9:30am - Trip to the GP

Harry’s been pulled from school. He’s in no state to be around others. Fever, a foggy head, making poor decisions. Possibly contagious.

The doctor confirms it.

“It’s something viral. Give him rest, fluids, and most importantly, keep him away from other people.

If your AI tool starts to drift, or behave erratically, or produce wildly inconsistent output, you pull it. Straight away. You don’t let it keep working in a live system where it might “infect” decisions, influence other tools, or confuse users. You isolate it, investigate, and only put it back once you’re sure it’s healthy. The problem is, many businesses don’t recognise the signs until it’s already caused damage. Just like you wouldn’t send Harry back to class coughing, sneezing and sweating, you don’t leave faulty AI in the loop. You pause it. Diagnose the issue. Take remedial action.

12:30pm - Homework Time

“Have you finished your maths homework? It needs submitting today.”

“Yes, I’ve done it,” says Harry from the sofa, pale and sniffing.

“You’ve done all three questions?”

He pauses. “...I looked at all of them.”

He's done one question, and it’s nonsense.

Teenagers will tell you what you want to hear. So will AI. It’ll act like it’s done the job, even if it’s only done a quarter of it, or completely misunderstood the task. Honesty and transparency are crucial, but with both Harry and AI, you often have to dig a bit. Review. Double-check. Ask follow-ups. And crucially, explainability matters. When Harry hands in wrong answers, Mum asks, “Show me how you worked that out.” With AI, the business should be asking the same thing. If the answer can’t be explained, or it’s clearly not right, you don’t let it continue unchecked.

2:00pm - Study Hour

Mum usually sits down with Harry and helps him break his work into chunks. 25 minutes on, 5-minute break. They check in every half hour. Mistakes are spotted. Progress is made. Frustration bubbles up, but the structure helps him stick with it.

Both need feedback loops. You don’t just assign a task and disappear. With AI, you run test outputs, assess quality, tweak the instructions, maybe retrain it entirely. Without regular feedback, neither teen nor machine improves they just carry on in the same loop, mistakes and all.

Today though Mum breaks work into tiny chunks. Harry’s got tissues, squash, a duvet. Progress is slow. She checks in constantly.

Not the day for big demands, just stability.

When AI systems are under strain or have recently been “unwell,” you ease them back in. Monitor output. Reintroduce tasks gradually. You keep feedback loops tight. You don’t just switch them on and assume all is fine again. Without oversight, things can spiral again. Illness and algorithmic drift both take time to bounce back from.

4:00pm - The Big Idea

Harry, despite being ill, bursts into the kitchen: “What if I made a YouTube channel for dogs doing football tricks?”

“You’ve still got a fever,” says Mum. “Brilliant idea, but maybe wait until you can see straight.”

Mum doesn’t dismiss it. It’s a decent idea that could go somewhere.

AI can suggest things when it's not in a great state. It may still sound clever. But it doesn’t mean you act on it. You can listen, note it down, and revisit it later when the system’s stable. Same rule for Harry. But when well, teenagers, and AI, can surprise you. Once you’ve got guardrails in place, once they’ve learned how to behave, they can generate ideas and solve problems in ways you didn’t expect. Recognise their strengths. Harness their creativity. Just don’t leave them unsupervised with a corporate Twitter account.

5:45pm - The Text Message

Mum gets a text from school: Harry’s maths homework wasn’t submitted on time.

“Harry?”

“Oh. I did it. Just forgot to upload it.”

Mum takes a deep breath. “Look, in the end, I’m the one who gets called in if this stuff keeps happening.”

Harry looks sheepish. He hadn’t really thought about that.

This is accountability. Mum doesn’t get to say, “Well, Harry did it.” She’s the one answering the phone if something goes wrong. Likewise, a business can’t hide behind its AI. If it causes harm, makes a dodgy decision, or spits out offensive rubbish, you are accountable. The AI won’t apologise. You will. That’s why businesses need not just controls, but ownership. Know who’s responsible. Don’t assume someone else is checking.

6:00pm - School Email

Notification arrives. “Harry was seen behaving oddly in science last lesson.”

Mum replies, “He’s home sick. He must have been ill yesterday as well. I’m really sorry. Maybe check that no other pupils have been infected”

Accountability again. Even when AI is misbehaving due to a fault, someone has to step in, explain what happened, and make sure others aren’t affected. This is where notification and escalation procedures come in. If something goes off, someone needs to be told. You don’t let a sick system go around the office coughing out dodgy predictions without telling a soul. Someone has to own it.

8:00pm – Quiet Debrief

Mum sits with Harry. “Feeling any better?”

“A bit,” he says. “Sorry I shouted at you earlier.”

They chat through what happened and what to do next time he feels off.

They agree on some adjustments for tomorrow.

Reflection helps. With teenagers, with AI. If something went wrong, sit down with the logs, the output, the users. Learn from it. How did we miss the signs? What checks failed? How can we improve the oversight next time? AI will fail at times. You must accept imperfection and be patient. Just like Harry, it needs room to grow, space to learn from missteps, and someone willing to have a quiet word when it veers off course. Blame rarely helps. Calm course correction does.

9:00pm - Mum’s Backup Plan

While Harry’s upstairs in bed watching football clips, Mum checks his school app. Homework due tomorrow? Done. Uniform clean? Yes. Chargers plugged in? Of course not. She plugs his phone in, tops up Harry’s water, leaves some paracetamol, and sets an alarm in case he spikes a temperature overnight.

Always plan for the unexpected. Whether it’s Harry forgetting his charger or your AI system crashing during a client demo, make sure you’ve got backups, audits, and rollback procedures. That extra step will save you from a panic later on. And if you’ve had to take your AI offline once, build in alerts, backups and fallback processes so the next time you can act faster. Don’t wait for it to go haywire again in front of clients.

Managing AI in your business? Just think of Harry.

Most days he’s fine. Occasionally brilliant. But sometimes, like today, he’s not himself. He makes poor decisions. Becomes unpredictable. A risk to others. But because his mum knows him, and knows what’s normal for him, she spots the change. She pulls him out of school, gets him seen by a professional, and doesn’t send him back until he’s ready.

You need that same level of attention with AI.

Know what healthy looks like. Watch for the signs. Be prepared to take it offline. Don’t let it run in live systems when it’s clearly not right. Someone must be accountable. Someone must be watching. Someone must have the authority to intervene.

Treating AI like a teenager might seem flippant, but it’s surprisingly effective. Teenagers can be smart, creative and genuinely helpful, but only when supported with the right combination of trust, boundaries and grown-up involvement.

And just like a parent is ultimately responsible for Harry, your business is responsible for your AI. Its decisions, its mistakes, its actions, all roll back to you. You can’t shrug and blame the system.

Because if Harry messes up, it’s Mum who gets the call.