Training teams to use AI at work has given me a front row seat to a new kind of occupational inequality.
Some people leave everything to machines and stop thinking. Others don’t touch it at all.
But there is a third group. They learn to be critical of AI and treat it like a bright, enthusiastic intern who needs management and support to do their best work.
difference? It’s rarely technical ability. It’s curiosity. A willingness to experiment, make mistakes, and figure out what AI is actually good at.
Here’s what I’ve learned so far:
Most people fail at AI because they don’t understand what it actually is
People I’ve worked with tend to vacillate between the extremes of treating AI as an all-knowing oracle and ignoring it completely after a single mistake.
Today’s AI has as much in common with the human brain as it does with the brains of birds and A380s. Both can fly, but that’s where the similarities end. Large-scale language models simply predict words based on patterns in the training data. That’s why they can produce fluent prose on well-covered topics, yet confidently construct stories in unfamiliar areas.
Once users understand this, their approach changes to provide clear goals and appropriate context. When you say that everything you get from an AI is crap, most of the time you find that they give generic answers to generic prompts.
People who get the best results treat AI as a skill, not a shortcut.
The biggest predictor of success is not technical ability. It’s about whether someone treats AI as a skill to be learned, rather than a magic box that may or may not work. The people who can use this best are those who experiment every day and reflect on how to get better results next time. The goal is to make the machine work for us, not think for us. This means using the machine proactively, critically, and proactively.
Just like humans, AI needs direction, feedback, and correction.
The skills needed to use AI are those that many people already have: communication and delegation. You’re not going to hand them a project and disappear like that intern did. Break it down, check it regularly, and course correct as needed. The same goes for AI.
And, just like with interns, you, as their manager, are ultimately responsible for what they produce. This is the true meaning of “human involvement.” It’s your job to keep the AI on track and make sure its output always reaches zero.
Don’t delegate decisions or give sensitive data to AI
A few months ago, a manager at a small retail chain was proudly showing me an HR dashboard he had coded using AI. Unfortunately, he was also importing sensitive information without thinking about what would happen if it were to be compromised and what policies he would need to follow. I sent him directly to the IT department.
But the risks go beyond safety. AI systems are trained on data created by humans and reflecting our collective biases. Avoid asking AI to make highly subjective judgments that can be prone to bias, such as “Should I pass this candidate through for an interview?” Instead, focus on evaluating facts, such as “Does this candidate have the right number of years of experience?”
Ignoring AI will not stop its impact
The environmental, ethical, and social impacts of AI are significant and growing. At a recent session for an environmental charity, one director was torn between our ability to do more as an organization and the moral costs of doing so, such as the carbon impact of running AI systems. But AI isn’t going away. It is far better to have a population that is knowledgeable about AI and can demand that it be built in a responsible and democratic way. AI is not a train waiting for us to board. I’m already on my journey. The only question is who will take the helm.
The speed at which AI is evolving leaves no room for slow decision-making.
Today’s version of the AI is the worst ever, but it’s improving faster than most people think. Tasks that were impossible a year ago are now routine. Where once you spent long nights hunched over your keyboard trying to figure out why your code wasn’t performing as expected, you can now write entire applications in just a few hours of prompts. When Anthropic’s CEO said last year that 90% of code would soon be written by AI, many developers laughed. Today, many admit that he was not that far off.
Unlike past technological revolutions, this one is moving faster than our ability to adapt. It took a century for the steam engine to become a locomotive, and 50 years for Faraday to become Edison’s power plant. Currently, there is a gap of several months between breakthroughs and global adoption. We cannot afford to have a 10-year debate. We must build social and democratic responses as quickly as we build technology. Otherwise, you risk being dominated by tools you don’t yet understand.
The people shaping how AI changes the world don’t have to be the engineers building these systems. They are likely to be people who take both ability and risk seriously and are willing to experiment. We all have a responsibility to not only understand AI ourselves, but also to encourage employers, communities, and governments to use it to ensure no one is left behind.
Tom Hewitson is the founder Chief AI Executive at General Purpose, a London-based AI training company
