"It would seem that Our Lord finds our desires not too strong, but too weak. We are half-hearted creatures, fooling about with drink and sex and ambition when infinite joy is offered us, like an ignorant child who wants to go on making mud pies in a slum because he cannot imagine what is meant by the offer of a holiday at the sea. We are far too easily pleased." >— C.S. Lewis, The Weight of Glory
I have used this quote from Lewis so often I am probably wearing it out, but it's great. A child offered a holiday at the sea, who instead wants to keep making mud pies in a slum — not because he weighed it up and preferred the mud, but because he genuinely cannot imagine what's being offered. He isn't greedy. He's small. We are, Lewis says, far too easily pleased.
That's where we are. That's the whole essay, really, but stay with me because the specifics are worse than the metaphor.
The miracle, and what we did with it
"AGI cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile — it is much more akin to the discovery of electricity or fire. If you stop to think about it, we've essentially found a way to make sand think. It's miraculous." >— Demis Hassabis
We found a way to make sand think.
Sit with that sentence for a second, because familiarity has already dulled it. We took silicon out of the ground, arranged it very cleverly, and it started reasoning. In the whole four-billion-year run of this planet that has happened exactly once, and it happened in the last handful of years, and you and I both got to be alive for it. It's like a miracle. AI feels like a powerful magic. Stop and think about it for just a moment how really, truly incredible this technology is. What we could do with it. What incredible wonders of the universe it could be leveraged to unlock, what — if pointed and harnessed well — human suffering could be relieved.
And then realise that almost the best we have come up with as a species is spending it on email summaries.
Meeting recaps. Slide decks. Slightly better autocomplete for people who write marketing copy. An unfathomable share of the most valuable compute ever built, and the best minds of a generation, allocated to helping enterprises produce marginally more content and employ marginally fewer people.
"If I had asked people what they wanted, they would have said faster horses." >— Henry Ford
Faster horses. Ford's line, apocryphal or not, has never been more literally true. Maybe Steve Jobs was right and the customer is an idiot. We asked the market what it wanted and it said the thing I already have, but quicker, and we said yes sir, right away sir, and pointed a magical machine that could plausibly cure cancer at somebody's inbox.
Two signals from inside the house
Two things happened this month that I don't think are unrelated.
Hassabis stepped down as CEO of Google DeepMind. He's now chair, and Alphabet's chief scientist — and crucially, he's still running Isomorphic Labs, the drug discovery arm. Read the shape of that move: away from the Gemini product roadmap, toward AGI strategy and curing disease.
I want to be careful here, because there's a version of this essay that overclaims. The reporting is mixed — Fortune has sources describing low morale, a talent exodus and model delays, which is a less romantic story than "visionary walks away from the product treadmill." Both can be true. But when a Nobel laureate hands over the flagship consumer AI product and keeps the medicine, I don't read that as a vote of no confidence in Google. I read it as a vote of no confidence in what we've collectively decided to point this thing at. As my friend said this week: "you don't win Nobel Prizes for allocating GPUs."
The second signal: Axios reported that Dario Amodei has privately worried that new hires are coming to Anthropic for the money rather than the mission. No direct quote, an unnamed source, and the internet — fairly — noted that Anthropic pays about the best in the business, so being surprised that money attracts people is a bit rich.
But strip the pile-on away and the underlying worry is real, and it's the same worry. When the mission is don't get this catastrophically wrong — or we are going to rewrite the future of our species and tell a better story — and the compensation is nine figures, the compensation wins the room. Not because people are bad. Because that's what money does to a room. Both men are, I think, circling the same anxiety from different sides: the gravity of the money is bending the field away from the point. I also believe that both of them are true believers. Both of them want to build the most powerful tool we have yet imagined to compound the greatest good it could possibly achieve.
The barbell, and the fat part in the middle — a race to Meh
Here's how I think the outcome space actually looks.
At one end: extreme upside. Cures for the diseases that have beaten us for a century. Fusion providing unlimited free energy driving the cost of everything to near zero. Quantum. Climate models that actually work and materials science that actually delivers. The genuine Star Trek future — post-scarcity, disease-managed, the species finally out of its own way.
At the other end: extreme downside. Mass structural unemployment, a shredded social contract, concentrated power, and the tail risks the safety people lose sleep over and get called hysterics for mentioning.
Both have a non-zero chance of being the real outcome, but in one we are going where no man has gone before, exploring the edges of the universe, and in the other we're huddled in a bunker hoping the robots don't use our bodies for batteries and brains for compute.
And in the middle, a big fat boring meh. A nothing burger. Modest productivity gains, some jobs quietly deleted, quarterly earnings marginally improved, the world basically unchanged except slightly more efficient and slightly more hollow with AI slop everywhere and a deep anxiety that a swing to either extreme is just on the horizon while CEOs complain about ROI.
We are playing squarely to the meh middle. We're not seriously mitigating the downside — there's no real appetite for it, it's expensive and unsexy. And we are absolutely not aggressively optimising for the upside. We've taken the most consequential technology in history and aimed it at the meh, which is the one outcome where nobody wins anything worth having.
There's a race, sure. Competition seeds and funds real progress and I'm not romantic enough to wish it away. But a race to what is the question nobody's answering. A race to mediocrity isn't a race worth running. A race to meh gets you tools that maximise an organisation's ability to return capital to shareholders while minimising headcount — which is, of course, the childhood dream of every visionary. That's the plan. That's the actual, stated, quarterly-reported plan.
Why it's shaped like this
It isn't stupidity. It's structure.
Decoupling AI from enterprise is untenable, because building frontier models costs more money than exists in most countries, and enterprise spend is the only pool deep enough to fund it. Enterprise money is slow, but the appetite is enormous and the cheques clear. So the labs need enterprise, and enterprise has enterprise-shaped problems, and enterprise-shaped problems are email and reports and headcount reduction.
The money doesn't just fund the industry. It shapes it. It makes it knucklehead-junk-shaped, relative to the scale of problems this thing could actually eat. Every incentive in the system points society at the mud puddle and invites us to enjoy the mud pies.
And I'm part of the problem
I should be honest about my own position here, because it would be very easy to write this as though I'm above it.
I'm not. I feel a genuine moral obligation to my kids and to grandkids I haven't met yet to take as much money off the table as I can, while I can — because history treats laggards badly, and every major technological transition has produced a group of people who were sincere and diligent and got flattened anyway. I don't want mine to be that group. Remember: bunkers, batteries and brains. There is a non-zero chance we get the bad outcome.
So I build the enterprise-shaped things. I'm good at it. I make organisations faster at the thing they already do. I'm one of the people carefully arranging the mud into a slightly better pie, a willing part of the problem because doing it pays better than not doing it, which is exactly what makes it a hard problem rather than a stupid one. Everyone in this system is being locally rational. That's the trap. Short-term incentives mis-aligned to long-term outcomes.
The naive part
Let me be wishful for a paragraph, and I'll own that it's wishful. I am a Walt Disney level optimist after all. I also think it's basically the reason Demis walked away from DeepMind.
What if we called it enough? What if we looked at the AI that already exists — which is, genuinely, more capability than enterprise has any idea what to do with, because adoption is the bottleneck and adoption is knucklehead-shaped — and drew a line. Said: this is sufficient for email. The enterprise problem is solved, we're just waiting for the org charts to catch up.
And then what if Google, Anthropic, OpenAI, the US and China held hands and treated the next five years as the largest unified research project in human history? A dialled-up-to-11 Manhattan Project, but completely jacked and aimed — not at each other, but at futility itself. Laid down arms in the arms race. Turned the opposing forces into collaborators and pointed the combined might of the best minds alive and near-unlimited compute at the things that are actually trying to kill us. Cancer. Dementia. Climate. Energy. Every threat, real and existential. The whole intractable list that has beaten every previous generation because they simply didn't have a tool big enough. Inbox zero may seem like a big deal, but it won't matter to you as much when the world is burning, I promise.
We finally have a tool big enough. That's the part I can't get past. When the problems are this huge, the time is this short, the threats so immediately real, and we finally — finally — have something that can generally solve hard things — why are we working on anything else? Why are we spending one more second trying to automate the job of Jan from accounting?
I know. Naive. Geopolitics, capitalism, game theory, the whole apparatus of why-we-can't. I'm aware that "everyone should simply cooperate" is not a strategy. But I notice we're very quick to call this naive and very slow to call the alternative what it is: narrow thinking, short-sighted bullshit.
The holiday at the sea
We don't get the Star Trek future by accident. It doesn't emerge from a sufficiently large number of productivity gains. It requires a hard, deliberate turn toward scale problems, with scale thinking and scale leadership. Everything else — every roadmap, every enterprise pilot, every one of my own projects, if I'm honest — is a distraction dressed up as progress, wrapped in a blanket of ego and a pillow of vanity.
The technology is here. It's already sufficient to attack problems that have been called insurmountable or intractable for a century. What we're missing isn't capability. It's vision, and leadership, and will, and the collective ability to not be stupid — to not play wealth accumulation against our own survival and call it strategy.
The holiday at the sea is right there. It's within reach, in our lifetime, for the first time ever.
And we're sat in the slum, making mud pies, because we cannot imagine what's being offered.