I was traveling with a group of friends recently, and volunteered to guide one leg of the trip through a city I had never visited and knew essentially nothing about. I had less than 12 hours to prepare. So, I handed the job to my favorite LLM: here's the group, here's what they want to see, here's what they don't, a few of them can't manage too many steps, we have this much time.
After some back and forth, refining a route, swapping out a stop, and adjusting the pace, it produced a very good tour: a route, a sequence of stops, practical notes on timing and terrain. It also gave me talking points for every stop, the history, the context, and the things worth noticing. I was the one who stood in front of the group at each stop and said them. I guided the group through a city I'd learned essentially from scratch the night before. Not one stop felt like a guess.
A few days later, a friend invited me to see what his company had built: an AI-enabled tool that compresses a task customers used to spend weeks on down to a few hours. The task still demands real domain expertise; the AI isn't generating that knowledge from scratch; it's automating the work of applying it.
What's actually being equalized here isn't the expertise; it's the software: building and running a program to do a job that used to require a human to do it. I was genuinely impressed. Then he asked what I thought, and I asked him a question back: what stops another team from building something similar? The models are available to everyone. Anyone with decent judgment and a good prompt can get most of the way there. So where, exactly, is the edge?
That question is really the same question twice, once about a group trip and once about a venture-backed company, and the answer is the same both times. AI is the great equalizer. It is handing out, freely and instantly, the kind of domain knowledge that used to take years, or a specialist's salary, to acquire. That's progress. It's also a serious problem for anyone whose value used to come from holding that knowledge alone.
What “equalizer” actually means
For most of modern work, expertise functioned as a moat. A tour guide knew things about a city that a tourist didn't. A specialist consultant knew a methodology a generalist couldn't replicate without years of training. A technical founder could build something a non-technical one simply couldn't, because the knowledge required to build it was scarce, and scarcity is what gave it value.
AI dissolves that particular kind of scarcity, fast, and it does it indiscriminately. It doesn't care whether the knowledge is trivial or hard-won, whether it took you a weekend to pick up or a decade of specialization. If it's pattern-based and has been written down somewhere, a capable model can reconstruct a usable version of it, on demand, for anyone who asks the right way.
The tour I built in an evening, the product my friend's team engineered, both draw on the same underlying capability.
The gap between a novice and an expert, for a large and growing category of tasks, is compressing toward zero.
This is democratizing. People with real ideas, but without the traditional credentials, networks, or capital that used to gate entry, can now get much further on their own.
A nurse with a good idea for a clinical workflow tool doesn't need to partner with a software team just to get a working prototype in front of someone who could fund it. A small business owner doesn't need a six-figure consulting engagement to get a credible first pass at a strategy. That's a real, substantial advantage. I don't think the industry talks about it enough; too much of the conversation about AI dwells on what it threatens rather than who it lifts up.
So, what's actually scarce now
My friend's question deserves a real answer, not just a diagnosis. If the knowledge itself is no longer the moat, what is?
Judgment is the first answer. It's the hardest one to fake. A model will happily generate ten plausible options. It won't reliably tell you which one is right for your specific situation, your specific constraints, your specific risk tolerance, unless you already know enough to ask it the right follow-up questions and recognize a wrong answer when you see one.
The tour I built worked because I knew which group member actually couldn't handle the stairs and which one was being polite about it. This is a judgment call no model could make for me. It's a sensing and weighting problem that still needs a human touch.
Execution and distribution are the second answer. A good prompt gets you a good draft. It does not get you the relationships, the trust, the regulatory approvals, the customer base, or the operational muscle to actually deliver the thing at scale, repeatedly, under real-world conditions.
My friend's company isn't defensible because of what the AI does. It's defensible, if it's defensible at all, because of what happens around the AI, the data it's trained on that competitors don't have, the workflow it's embedded in, the customers who already trust the brand enough to adopt it.
Intuition and insight are the third, and the most underrated. When everyone has access to the same raw capability, the differentiator becomes what you choose to do with it, what you leave out, what you refuse to ship, what you notice is subtly wrong when it technically passes every test. That's a human faculty that AI amplifies but cannot replace.
How to actually work in this world
So, let’s be practical. Stop competing on access to information. That race is already over. Compete on judgment instead: the ability to decide what to do with the information you have.
Spend less time building the kind of expertise a model can now hand you in 30 seconds, and more time building the intuition and insight it can’t: pattern recognition earned from being wrong in the real world, a feel for your specific customer that no training set captures, the instinct for which plausible-sounding idea is actually a trap.
Build the parts of your advantage that live outside the model: relationships, proprietary data, operational capability, a reputation people trust. Get genuinely good at directing AI rather than just using it; the person who knows precisely what to ask for, what constraints matter, and how to spot a confidently wrong answer will consistently outperform someone with access to the identical tool but weaker judgment and intuition about how to wield it.
I expect my friend's company will be fine because everything they've built around their AI – the data, the workflow, the customers who trust them – isn't something a competitor can prompt their way into overnight. That’s a moat. And I will keep volunteering to guide trips to cities I've never seen because I've gotten better at knowing which of the model's suggestions to trust, and which ones need a second look, or edit, before I walk 20 people there.
Everyone now has the same model. Not everyone has the same judgment and intuition. That's the whole game.