There are two kinds of AI advice doing the rounds, and both are useless. One says AI will transform everything and you're finished if you don't adopt it tomorrow. The other says it's a toy that makes things up. The truth is duller and more useful: AI is genuinely good at a specific shape of work and genuinely bad at another, and the whole skill is telling which is which. Here's how I'd work it out for a real business.
Where it genuinely earns its keep
AI is strongest on the work that's high-volume, low-stakes, and easy to check. Think first drafts, summaries, turning rough notes into something readable, rewording the same message for three audiences, sorting a pile of feedback into themes. The common thread is that a human can glance at the result and instantly see if it's right, and being wrong occasionally costs you almost nothing.
It's also good at the boring middle of tasks you already know how to do. If you can describe a job clearly and you'd recognise a bad answer, AI can take the first pass and save you the blank-page time. I use it like that constantly. It doesn't do my thinking, it does the typing while I do the thinking, and that's a real saving when it's a job you'd otherwise put off.
Where it quietly costs you
AI is weakest exactly where the hype is loudest. Anything where being confidently wrong is expensive, AI will hurt you, because it is confidently wrong on a regular basis and does not know when it's guessing. A legal or financial statement, a medical claim, a number a customer will act on, a promise about what your product does, those need a person who carries the consequences, because the AI doesn't.
It's also a poor fit for work that depends on things only you know: your actual customers, the context behind a decision, the judgement you've built over years. AI will give you a plausible generic answer, and plausible generic is often worse than nothing because it looks finished. The danger isn't that it refuses, it's that it cheerfully hands you something that's 80 percent right in a way that's hard to spot.
The test I'd apply to any task
Before you point AI at something, ask two questions. Can I instantly tell if the answer is wrong? And what does it cost me if a wrong answer slips through? If you can spot mistakes easily and a mistake is cheap, let AI take the first pass, it'll save you real time. If mistakes are hard to spot or expensive when they happen, keep a person firmly in charge and use AI, if at all, only as a sounding board.
That single filter cuts through almost all of the noise. It's why "use AI to draft your newsletter" is sensible and "use AI to decide who to make redundant" is not, even though both are technically possible.
The honest bit
Most businesses don't need an AI strategy in the grand sense. They need to find the three or four genuinely repetitive jobs where AI would save real hours, set those up properly, and stay well away from the places it would quietly do damage. That's far less exciting than the hype, and far more useful.
Working out which of your tasks fall on which side of that line is exactly what I help with, as someone who builds with these tools every day and has no incentive to oversell them. If you want a straight, practical read on where AI fits in your business, take a look at how I work or book a free half-hour surgery and we'll go through it together. If the honest answer is "barely anywhere yet," I'll tell you that too.