🤚 The Open-Palm Illumination

Ladies and gentlemen of the velvet-roped intelligence community, we have a problem. Not an artificial one — a definitional one. Salim Ismail, founder of OpenExO and the man who wrote the book on exponential organizations (literally), has decided to say what the rest of us have been politely coughing around at cocktail parties: “I call bullshit on AGI.”

Not because AI isn’t powerful. Not because the models aren’t impressive. But because the entire AGI debate is built on a foundation made of vibes. The argument is elegantly simple: how can you declare that artificial intelligence has achieved “general” intelligence when nobody — not the neuroscientists, not the philosophers, not the computer scientists, and certainly not the CEOs reading from prepared remarks — can agree on what intelligence actually is?

It’s the equivalent of declaring you’ve climbed the tallest mountain in a country that hasn’t finished its topographical survey. You’re standing somewhere high, breathing heavily, planting a flag, and hoping no one checks the map.

👐 The Two-Handed Reality Check

Let us now survey the magnificent wreckage of everyone’s competing definitions.

OpenAI originally defined AGI as “highly autonomous systems that outperform humans at most economically valuable work.” Elegant, practical, and conveniently measurable in revenue. Microsoft’s secret contract with OpenAI allegedly defined AGI as technology generating $100 billion in profits — which is less a definition of intelligence and more a definition of a very good quarter.

Jensen Huang, Nvidia’s CEO, declared on the Lex Fridman podcast in March 2026 that “we’ve achieved AGI” when asked if AI could start a billion-dollar company. He then immediately hedged, noting the company need not remain valuable. Later in the same episode, he said AI agents could “never replicate Nvidia” — thereby undermining his own claim with the confidence of a man who makes the chips everyone else is fighting over.

Meanwhile, Google DeepMind tried to bring actual science to the circus by publishing a cognitive taxonomy identifying ten key cognitive faculties — perception, reasoning, memory, learning, attention, social cognition, and more — against which AI systems could be measured. Their finding? Current AI has what researchers call a “jagged profile”: superhuman at mathematics, undergraduate-level at common sense, and functionally absent at social understanding.

In other words, the most advanced AI on Earth has the cognitive profile of a physics prodigy who cannot read a room. We have all worked with this person. We did not call them generally intelligent. We called them “difficult to seat at dinner.”

🌿 The Gentle Awakening

Ismail’s critique cuts deeper than mere semantics. Intelligence, as a concept, has been contested since the invention of the word. Is it the ability to learn? To adapt? To solve novel problems? To understand that your colleague’s “I’m fine” means they are categorically not fine? The field of psychology spent the better part of a century arguing about this and settled on approximately nothing.

The AI industry, in its characteristic modesty, decided to skip that entire conversation and jump straight to declaring that the artificial version of the thing they can’t define has been achieved. This is like skipping medical school and going directly to performing surgery because you’ve watched every season of Grey’s Anatomy and your hands are steady.

What makes this particularly consequential is that trillions of dollars in investment, government regulation, and corporate strategy are now anchored to a term that has no fixed meaning. When OpenAI hits $100 billion in profits, does AGI arrive because Microsoft’s contract says so? When DeepMind’s cognitive taxonomy shows perfect scores across all ten faculties, does AGI arrive because academia says so? When Jensen Huang says it on a podcast, does AGI arrive because the man who sells the infrastructure says so?

The answer, apparently, is yes to all of the above and none of the above simultaneously.

👑 The Crown Verdict

Here is the uncomfortable truth that Salim Ismail has articulated with admirable directness: the AGI debate is not a technical debate. It’s a marketing debate wearing a lab coat.

Every major AI company has a vested interest in defining AGI in a way that either makes their product the first to achieve it or makes the achievement perpetually five years away (depending on which narrative better serves their current funding round). The goalpost doesn’t move because of scientific progress. It moves because someone’s Series G depends on it.

DeepMind’s cognitive taxonomy is the closest thing we have to an honest framework, and even it reveals more about what AI isn’t than what it is. The jagged profile problem isn’t a bug to be fixed in the next model release. It’s a mirror showing us that we’ve been building savants and calling them generalists.

Perhaps the most intelligent thing we could do — artificially or otherwise — is agree on what we mean before we declare victory. But that would require the kind of nuanced, multi-stakeholder consensus-building that AI was supposed to replace.

Inspired by The AGI Debate Has a Huge Problem | MOONSHOTS by Peter Diamandis.

Your definition is showing. Define wisely.