Marc Andreessen just told Joe Rogan the only AI skill left is knowing what to ask. Here’s how he actually does it.
Marc Andreessen spent over three hours on Joe Rogan’s podcast in May 2026, and the headline everyone pulled from it was the boldest one: he believes AGI already arrived, about three months before the interview, with the release of GPT-5.5, Claude 4.6, Gemini 3, and Grok 4.3. Whether or not you accept that claim, the more useful part of the conversation was quieter, the specific way Andreessen says he actually talks to these tools. That’s the only AI skill left worth building, and it has nothing to do with which model you use. It also happens to be the one skill almost no one is deliberately practicing.
The Claim That Gets the Headlines
Andreessen told Rogan that for almost any topic, the top models now give him better answers than the world-class experts he could call by phone, and he has access to a lot of experts. He also said doctors are already using ChatGPT in the exam room, typing a patient’s symptoms in the moment they stop talking. His exact words: “at that point you’re asking the question of like, what do I need you for.”
Take both claims seriously and verify them yourself. Andreessen hasn’t published data behind the “better than experts” claim, and a separate Nature Medicine study on a comparable AI health tool, evaluated against 60 physician-written emergency scenarios, found it under-triaged 52% of cases that actually required immediate care. The gap between “impressive in conversation” and “safe to rely on for a real diagnosis” is exactly the gap responsible use of these tools has to account for.
The Only AI Skill Left
Here’s the part that matters regardless of where you land on the AGI question. Andreessen’s own framing. The models can already do almost anything you can describe in plain English. The bottleneck isn’t the model anymore, it’s you. The only AI skill left is knowing what to ask.
That’s a bigger claim than it sounds like. For most of the last few decades, the scarce resource was access, access to the world-class doctor, the world-class lawyer, the expert who’d seen your exact problem before. Andreessen’s argument is that access is no longer the constraint. The constraint moved entirely into your own head. Your ability to frame the problem well enough that a generalist expert, sitting on the complete corpus of human knowledge, can actually help you.
How He Actually Uses the Only AI Skill Left
The specific techniques are worth stealing regardless of your view on AGI. When a topic is genuinely hard, Andreessen escalates simplicity on purpose, he asks the AI to explain it, then asks again like he’s 10, then again like he’s 5, then again like he’s 2, until it actually clicks. When a question has two legitimate sides, he doesn’t ask the AI for “the answer.” He asks it to steelman one side, then steelman the other, and reads both before he decides anything himself. And for the biggest, messiest questions, he has the AI role-play a panel, a doctor, a lawyer, a historian, a psychologist, and argue it out with each other while he reads the debate.
None of those three moves require a better model. They require deciding, before you type anything, what kind of answer would actually change your mind, which is a thinking skill, not a technology skill. That’s the only AI skill left that’s actually worth practicing on purpose, and almost nobody teaches it explicitly, in a classroom or in a boardroom.
Why This Is an IDEAS Problem, Not a Tools Problem
This is the IDEAS ingredient of the Kryptonite Defense from a completely different angle than I usually argue it. IDEAS was never really about having access to smart people or smart tools, plenty of people have both and still ask worse questions than they should. It’s about the discipline of framing a problem well enough that whatever’s in front of you, human or model, can actually help. Andreessen’s line captures it better than I usually manage to, the moment you think “I don’t know how to figure this out” is exactly the moment most people give up, and exactly the moment you should be doing more thinking about the question, not less.
The people and organizations who treat “what should I even ask” as a skill worth practicing are going to get more out of every tool they touch, this generation of AI and whatever comes after it. The ones who wait for a smarter model to save them from asking a sharper question are going to stay exactly where they are, no matter how good the only AI skill left theoretically becomes for everyone else around them.
This also isn’t a skill reserved for billionaire investors with access to frontier labs. Anyone with a laptop and a free-tier chatbot account can practice escalating simplicity, steelmanning both sides, or staging a panel-of-experts debate on their own hardest problem tonight. The tool is no longer the differentiator. The question is.
What’s one hard question in your business right now that you haven’t actually framed well enough to get a good answer to, from anyone, human or AI?
This is exactly the kind of shift I walk through in Distinct or Extinct, where the real threats and opportunities of AI disruption actually sit, and the 90-Day Deployment Plan for responding to them before your competitors do. If you want a clearer read on where your organization stands, the Kryptonite Scorecard will show you: realmikeevans.com/scorecard