Why You Still Need a Human Coach in the Age of AI
AI can summarize your market, draft your strategy doc, and roleplay an investor. It still can't do the one thing that actually changes how you lead.
Last week, I read Gangesh Pathak’s thought piece called, “Why AI’s Next Frontier Belongs To Human Judgment.” That put something into words I’ve been chewing on for a while. His argument, in short: the AI labs have quietly stopped chasing more data and more compute as the path to better models. What they’re chasing now is better human judgment. The expertise that goes into shaping what the model learns, what counts as a good output, what’s almost-right versus actually-right.
His piece is worth reading in full. The data is striking. Salaries for specialized AI trainers have climbed past $180K at the top end. Demand for these roles has grown more than 150% in two years. As the machines got smarter, the value of human expertise went up rather than down.
The line from his piece that stuck with me most: “You can scale the production of answers. You cannot scale taste.”
I want to take Gangesh’s argument and apply it somewhere specific. If human judgment is the scarce input for training the most advanced AI systems on the planet, what does that say about the messier, higher-stakes problem of running a company?
The encyclopedia problem
Here’s how I think about using AI as your business advisor.
In 1985, having an encyclopedia at home was a real advantage. You could look things up. You could answer questions your neighbors couldn’t. It was a genuine information edge.
By 1995, every middle-class household had one. By 2005, physical encyclopedias had mostly migrated to software. By 2015, Wikipedia made the entire genre obsolete. The encyclopedia didn’t get worse. The moat just evaporated, because everyone had it.
That’s where AI sits in 2026 as a business advisor. ChatGPT and Claude can give you a credible answer to almost any general business question. They can frame your strategy. They can pressure-test your reasoning. They can role-play an investor objection or draft a board update. All of that is genuinely useful, and you should be using these tools.
But it isn’t a moat. Every founder you’re competing with has the same tools. The same access to the same models. The same ability to ask the same questions and get the same well-reasoned, slightly-too-polished answers.
If your competitive edge is “I regularly use an LLM,” you don’t have an edge.
What AI cannot do for you
A model can tell you what other founders in your situation have done. It can synthesize the conventional wisdom. It can list the seven frameworks for thinking about your hiring problem. What it cannot do is look at you, in your specific moment, with your specific blind spots, and tell you which framework is actually wrong for you. It also can’t tell if you’re not really being honest with yourself or with it.
That requires three things AI doesn’t have:
It requires reading the room. Not the room in a strategy doc, but the actual room. The way you said something, the thing you didn’t say, the place you got defensive. A good coach catches the half-sentence you backed away from.
It requires having watched a hundred other founders navigate the same problem and remembering which ones got it right and which ones flamed out, and why. That’s pattern recognition built from experience, not from training data.
And it requires the willingness to disagree with you in a way that actually lands. AI is trained to be helpful, which often means trained to be agreeable. A coach worth paying for is willing to tell you that you’re wrong, and to keep telling you until you actually hear it.
The deeper thing
There’s something even harder to name that gets lost in this conversation, which is that humans only get really honest with other humans.
I sit across from founders every day. The things they actually need to work on, the real questions underneath their stated questions, almost never come out in the first five or even 20 minutes. They come out when there’s enough trust in the room that the founder is willing to stop performing competence.
The moment someone stops performing and starts actually thinking (with both the mind and the heart) is the moment coaching becomes valuable. And it doesn’t happen with a chatbot. It happens with another human, present, paying attention, who has earned the right to ask the next question.
This is simply how humans are built. We’re a social species. Our deepest cognition happens in dialogue with other humans we trust. You can simulate a form of that with AI. You cannot replicate the impact.
What this means for you
Gangesh’s bet is that the next decade goes to companies with the best people wielding the tools, not the ones with the most tools. I think that scales down to the individual founder too.
The AI tools are a commodity. You have them. So does your competition. The differentiator now is the judgment you bring to using them, which is itself sharpened by working with people who have judgment sharper than yours.
Use the tools. Use them aggressively. Let them handle the analytical, repetitive, synthesize-able work so you have more bandwidth for the work that actually moves your company forward.
But don’t confuse access to information processing power with access to wisdom. They’re not the same thing. They were never going to be the same thing. The data is now confirming what the best founders already knew.
The machines are getting smarter. Your need for a real human across the table is getting more important, not less.
 | A guest post by
|