Everyone Is Using AI. Almost Nobody Knows How.
The tools are extraordinary. The way most of us are using them is… not.
A reasonable estimate suggests that somewhere north of 400 million people have used an AI assistant in the past year. A much smaller number are getting anything close to the actual value out of it. The gap between what these tools can do and what most people ask them to do is genuinely striking — and mostly not the technology's fault.
A reasonable estimate suggests that somewhere north of 400 million people have used an AI assistant in the past year. A much smaller number are getting anything close to the actual value out of it. The gap between what these tools can do and what most people ask them to do is genuinely striking — and mostly not the technology's fault.
The single most common mistake is treating AI like a search engine. People type a short, vague question and expect a useful answer, then feel underwhelmed when it sounds generic or misses the point. Which is fair enough, because that's exactly what happens. The tools built on large language models are not retrieval systems. They're reasoning systems. They respond to context, specificity, and nuance in ways that a Google search simply doesn't.
The gap between what these tools can do and what most people ask them to do is genuinely striking.
The gap between what these tools can do and what most people ask them to do is genuinely striking.
Ask one to 'write me an email' and you'll get a perfectly serviceable nothing. Ask it to 'write me an email to a client who I think is about to pull out of a deal — they went quiet after we sent the revised quote, I don't want to be pushy but I need to know if they're still in — keep it short, confident, no desperation' and something genuinely useful comes back. The difference isn't the tool. It's the instruction.
The other thing worth knowing: AI doesn't get tired of context. You can give it an enormous amount of information about your situation, your tone, your constraints, your audience — and it will use all of it. Most people give it almost none of this. They write the shortest possible prompt, as if being brief is a virtue, when the opposite is usually true.
There are real limitations too, worth being clear-eyed about. These tools can be confidently wrong in ways that are hard to detect without expertise. They have cutoff dates on their knowledge. They'll tell you wha
The other thing worth knowing: AI doesn't get tired of context. You can give it an enormous amount of information about your situation, your tone, your constraints, your audience — and it will use all of it. Most people give it almost none of this. They write the shortest possible prompt, as if being brief is a virtue, when the opposite is usually true.
There are real limitations too, worth being clear-eyed about. These tools can be confidently wrong in ways that are hard to detect without expertise. They have cutoff dates on their knowledge. They'll tell you wha
t you seem to want to hear if you push back on them, which is not the same as agreeing with you because you're right. None of this makes them not worth using. It just means they reward some thought about how to use them, rather than being things you can point at a problem and expect magic.
The people who are genuinely benefiting — writers, analysts, marketers, small business owners — tend to use them with a lot of specificity and a healthy scepticism. They check the output. They push back when something doesn't land right. They treat the tool as a capable collaborator rather than an oracle.
That sounds obvious, written down. It's also clearly not how most people are approaching it yet. But it's learnable. Quickly. And the gap between those who've figured it out and those who haven't is already wider than most people realise.
The people who are genuinely benefiting — writers, analysts, marketers, small business owners — tend to use them with a lot of specificity and a healthy scepticism. They check the output. They push back when something doesn't land right. They treat the tool as a capable collaborator rather than an oracle.
That sounds obvious, written down. It's also clearly not how most people are approaching it yet. But it's learnable. Quickly. And the gap between those who've figured it out and those who haven't is already wider than most people realise.
Features
