Insight / signal
Three conversations about AI. Only one creates leverage.
The failure mode is not cynicism. It is category confusion. People are having a prediction conversation when the useful conversation is an operations conversation.
Three very different conversations about AI landed in my feed this week.
The first was Mo Gawdat on The Diary of a CEO. Former Chief Business Officer of Google X, not a crank. His argument: AGI is already here, 30% of jobs will be gone by 2027, and the structural problem is that capitalism breaks when the workers are also the buyers. He gives us roughly three years before it starts hitting hard.
The second was John Lennox, Oxford mathematician and philosopher, on the same show. His argument: Silicon Valley’s AI promises map almost exactly onto traditional religious eschatology. Immortality through merging with machines. An omniscient superintelligence that solves all suffering. His counterclaim is that consciousness cannot be reduced to computation, and that whatever AI can replicate, it cannot replicate personhood.
The third was Tom Bilyeu on Peter McCormack’s podcast. No grand theory. Just this: he rebuilt a significant part of his business operations using AI in seven days. Work that would have taken a full team and hundreds of thousands before. Done in a week.
I find Gawdat’s timeline alarming in tone but serious in substance. He has seen things most people have not. His structural argument about capitalism and displaced workers is not science fiction. It is economics.
I find Lennox genuinely interesting. The frame of AI as competing religion is a useful bullshit detector. When someone talks about uploading consciousness, achieving immortality, or solving the human condition through computation, Lennox’s question is worth asking: are these technical predictions or are they old promises in new clothes?
Both conversations are worth having.
But neither one will change your business position this week.
The Bilyeu story is the one that does.
Not because rebuilding something in seven days is magic. It is not. The specifics matter: what was rebuilt, what tools were used, what the team looked like, what broke, what stayed human. A one-line anecdote from a podcast is not a blueprint.
But the direction it points is clear.
Small, capable operators using agents well are already working in a different operating reality than the businesses watching podcasts about AI doom. The gap is not philosophical. It is not about which AI predictions turn out to be right. It is about whether you have started rebuilding yet or whether you are waiting for the debate to settle.
The debate is not going to settle. That is not how these things work.
Gawdat’s timeline might be right. It might be wrong. Lennox’s argument about consciousness is worth taking seriously, but it does not tell a business leader what to do on Monday morning. The philosophical questions are real. They just do not unlock the next quarter’s revenue.
The failure mode I see in most business AI conversations is not cynicism. Cynicism is fine. Healthy, even.
The failure mode is category confusion.
People are having a prediction conversation or a consciousness conversation when the useful conversation is an operations conversation.
What can a small team with agent leverage actually do that a large team without it cannot? What did it cost before? What does it cost now? What is the bottleneck in your business that is not being touched because it feels too complex or too expensive? What would it take to rebuild that in seven days rather than seven months?
Those questions are not glamorous. They do not make for compelling podcast content. Nobody is getting half a million downloads asking “what are your weekly agent usage patterns.”
But those are the questions with answers that compound.
I have been building Foundry as a proof layer for this argument.
Not proof that AI is conscious. Not proof that AGI is coming in three years. Proof that a small team with agent infrastructure can do commercial marketing work, client delivery, content operations, and internal tooling at a velocity that an equivalently staffed agency without that infrastructure cannot match.
That is a falsifiable claim. Either the builds happen or they do not. Either the delivery is faster or it is not. Either the economics are better or they are not.
The post-agency model is not a prediction about 2027. It is a bet on what is already possible now, and what compounds fastest if you start this week rather than next year.
Gawdat’s three years might be the timeline for the big disruption. Fine. That gives you three years to be inside the disruption rather than downstream of it.
Lennox is right that consciousness is a harder problem than any current model solves. Also fine. You do not need conscious AI to rebuild an operations layer. You need reliable agents, good tooling, and someone willing to do the unglamorous work of making the system trustworthy.
Bilyeu’s seven days is not the point. The point is that he did not wait for the debate to end.
The practical move is not to dismiss Gawdat’s warning or Lennox’s philosophy.
It is to have the third conversation first.
What is one thing in your business that agents could rebuild, speed up, or replace in the next month? Not in theory. Actually. What would the first seven days look like? What would you need to put in place? What is the blocker?
That is the conversation worth having.
The other two will still be running when you get back.