A colleague recently asked me what a “10X marketer” means. My answer was capability expansion. A strategist can explore a data question, build a working prototype, or test an idea that would previously have stayed in a deck until someone with a different skill set had time to help. The distance between having an idea and doing something with it is shrinking.
Then came the better question: how do you tell the difference between expanding your capabilities and overestimating your competence?
That is the question I keep coming back to. It connects how we work, how we learn, what we build, and even how we argue about AI safety. We are gaining access to intelligence we can direct. We still have to decide what deserves doing, recognize when the result is weak, and take responsibility for what happens next.
I.From an idea to something you can interrogate
There is a version of the AI productivity conversation that stops at making the same report faster. I understand the appeal. But the part that excites me is being able to pursue an idea I would otherwise have had to put aside.
Imagine a marketer who sees a useful pattern in customer behavior. With AI, that person might work through the data, explore competing explanations, and build a rough tool that makes the pattern visible. They can bring a data scientist something concrete to examine. A creative partner can respond to a working experience. The conversation gets more specific because there is something to challenge.
That does not make the marketer a data scientist or a creative director. It lets the marketer get further into the question before asking for their help. Those experts can then apply their judgment to a more developed idea, including telling the marketer that the apparent pattern is an artifact or the prototype solves the wrong problem.
The distinction matters. A polished result can make you feel as though you understand a field you have barely entered. I want the ability to explore across disciplines while retaining enough humility to know where someone else's expertise becomes essential.
II.Human judgment, machine precision
Consider a hypothetical automotive campaign. A strategist identifies a practical reason that someone shopping a competing vehicle might switch: more usable space, a better fit for a family's routine, a feature the competitor handles poorly. Turning that insight into many distinct messages, landing pages, tests, and budget decisions creates an enormous coordination problem.
This is the kind of work I want AI to help make possible. The ambition is to carry a good strategic idea through all the details that usually limit its execution. Each variation still needs an accurate claim, a meaningful audience, and a way to tell whether it worked. Producing seventy versions of a weak idea would only make the problem larger.
I think of the opportunity as human judgment combined with machine precision. But precision has to be earned through checking. An automated system can execute the wrong instruction very consistently.
The same distinction applies to software and services. If you can build more of the machinery, the questions become sharper: whose problem are you solving, what does a good outcome look like, and how will you know when the system has failed? The ability to produce an application does not answer those questions for you.
III.Protect the skills that make judgment possible
I came to these tools with roughly twenty years of professional experience. That matters. I have a store of mistakes, patterns, and hard conversations to compare an answer against. Someone beginning their career is still building that store.
So I worry about the instruction to “just review the AI's work.” Review requires knowledge. You need to recognize a missing assumption, a misleading comparison, or a recommendation that is technically plausible and practically useless. Those are harder to catch than a misspelled word or an incorrect sum.
We need to protect the skills that support that judgment. Learning to work through a problem, estimate an answer, explain a causal argument, and question evidence still has value even when a tool can produce the finished-looking output. I would redesign learning around that purpose: decide which work a person needs to do themselves to develop understanding, then use AI to help them go further.
That applies to experienced people too. My preferred use is to arrive with a point of view and ask AI to find the holes in it. I can push back, examine the criticism, and take the argument to another model for a different challenge. The work still requires me to think.
Of course, I can be wrong about my own expertise. Years in a field do not make anyone immune to a persuasive answer. The useful habit is being able to explain why I accepted a recommendation, beyond the fact that the model sounded confident.
IV.The safety question is also a steering question
This is where the workplace discussion connects to my frustration with the AI safety debate. Too often, I hear people answering different questions. One person is worried about concrete misuse. Another is arguing about whether artificial general intelligence exists. A third is concerned that regulation will protect incumbents. Those concerns can coexist.
I do not need to settle a forecast about AGI to care about a system making harmful actions easier. Nor do I need to dismiss AI's promise to ask who can direct it, what they can authorize, and how mistakes will be caught.
The same expanded capability that lets a thoughtful person pursue a useful idea can give a careless or malicious person more reach. That is a reason to take the design of permissions, verification, and accountability seriously. It is also a reason to be specific about the harm we want to prevent, so we can evaluate whether a proposed safeguard would actually help.
In a business workflow, that starts with ordinary questions. Can the system draft a message, or send it? Recommend a budget change, or spend the money? Who reviews an exception? Can an action be reversed? A more capable model does not remove the need to answer them.
V.Build the habit of steering
The discipline I want is a repeatable one. Begin with an objective I can explain. Give the system the context and constraints it needs. Explore possibilities. Check the result against evidence and against the people who understand the problem. Watch what happens after it leaves the prototype stage.
That discipline needs to exist across a team. If one person can move from an idea to a working model while the rest of the organization can only engage through the old sequence of handoffs, much of the advantage stalls. People need shared ways to examine the work, contribute their expertise, and decide when it is ready.
This is why I find the “10X” label less interesting than the behavior behind it. What can you now attempt that you could not attempt before? How much better can you make the result? What are you learning as you do it? And can you still recognize when you are out of your depth?
Those are the questions I want to keep asking as the tools improve. I want to use the new capability fully, and build the judgment to match it. Steering intelligence is becoming part of the work. Learning to do it well is work of its own.