The wrong question everyone's asking I get asked this roughly three times a week now. "Should we use AI?" People say it like they're asking whether they should buy insurance or update their website. Binary. Urgent. Like there's a right answer that applies to everyone. There isn't. What I've learned working with people building things across London and beyond is that the AI question is almost always the wrong question. It's a symptom of something else. Either you're feeling pressure to keep up. Or you've seen someone else use it and thought it looked easy. Or you're genuinely stuck on something and you're wondering if AI could help. Those are three completely different situations. The real question is simpler: what's actually broken right now? And could AI fix it without breaking something else? Why caution isn't paranoia Look at what's happening in the financial sector this week. Finance ministers and bankers are raising genuine concerns about new AI models potentially identifying cybersecurity weaknesses in ways no one anticipated. These aren't Luddites. These are people whose job is managing risk at scale. They're not saying "don't use AI." They're saying "understand what you're deploying before you deploy it." I think they're right. I've watched someone at a small business get excited about an AI tool, implement it in 48 hours, and then spend three weeks fixing the data quality problems it created. The tool worked exactly as designed. But it was designed for a different problem than the one they actually had. They'd grabbed the solution before they'd properly diagnosed the issue. This isn't unique to AI. It's just more visible with AI because the hype is louder. Start with what you're losing time on Here's where I'd begin: what takes up hours of your week that feels like it shouldn't? Not what's hard. What's repetitive. What's tedious. What pulls you away from the stuff only you can do. Maybe it's writing the same emails over and over. Maybe it's data entry. Maybe it's summarising customer feedback or drafting reports. Maybe it's scheduling. These are the places AI actually works well right now. Not because AI is magic. Because these tasks have clear inputs, predictable outputs, and low tolerance for creativity. If you're spending 6 hours a week on something like that, and an AI tool could cut it to 1 hour, that's 260 hours a year. That's real. That matters. But if you're hoping AI will fix something that's actually a process problem or a people problem, it won't. I've seen people try. It doesn't work. An AI tool won't fix unclear handoffs between teams. It won't solve a product that nobody wants. It won't replace good management. The framework that actually works When someone brings me a problem and asks about AI, I work through this: First, is the problem actually time-based or is it something else? If you're not doing something because you don't have time, AI might help. If you're not doing something because you don't know how, or you don't have the skills, or it's genuinely unclear whether it matters, AI probably won't fix that. Second, what could go wrong? I don't mean existential risk. I mean: if this tool makes a mistake, what happens? If it hallucinated a fact, or gave you bad advice, or got the tone wrong, what's the cost? High cost equals high scrutiny. Low cost equals maybe you can move faster. Third, do you need to trust it without checking? This is the bit people skip. Most AI tools work best when you treat them as drafts, not finished products. They work best when someone still reads the output. If you need something that works unsupervised, that's a different category of tool and a different category of risk. Fourth, is the data you're feeding it any good? Garbage in, garbage out. This one catches people. You'll be amazed how many businesses realise their data is a mess only when they try to use it with an AI tool. What you probably shouldn't do Don't adopt AI because everyone else is. Don't adopt it because it feels like you should. Don't adopt it in a panic because you read something that made you nervous about falling behind. Don't implement something without a clear measure of success. "We'll use AI" isn't a plan. "We'll use AI to cut data entry time from 8 hours to 2 hours per week" is a plan. You can measure it. You can tell if it's working. Don't deploy it in customer-facing contexts without understanding what you're risking. I've read about people using AI to file complaints against businesses, with false statements generated by the model itself. That's what happens when tools are used without thought to the downstream effects. Don't assume the tool you've heard about is the right tool for your specific situation. There are hundreds of them now. The one your mate recommended might be brilliant for his work and useless for yours. What to do this week Write down three tasks that took up at least an hour each last week. The ones you found tedious. Be specific. Not "admin" but "copying customer feedback into our spreadsheet" or "writing follow-up emails to leads who didn't respond in 72 hours." For each one, ask yourself: could an AI tool do the first draft, leaving me to review and adjust? If the answer's yes, spend 30 minutes researching two or three tools that claim to do that thing. Read the reviews. See what people actually struggled with. Don't buy anything. Just understand what's available and what the tradeoffs are. Then pick one. Just one. Run a two-week pilot. Track whether it actually saved you time. If it did, great. If it didn't, stop using it. You've lost two weeks and learned something real. That's better than spending six months on something that doesn't work.