Org Age Essay Q4 2026

Practice leads to Theory

Grown not built, discovered not developed

Companion on The Lab Bench →

It’s surprising to the point of suspicion that even the AI labs have no special insight into what their models are capable of. I’m sure someone reads the attached model card, that half wingman half hype man for the debutante’s introduction to society, but you get the sense that until the cracked masses start poking and prodding no one really has any idea of what mysteries are encoded in the weights. The documented claims are there for reference, but really until you’ve got a benchmark of your own all we’ve got is vibes and copied prompts.

“Give me irrefutable proof that we are in the singularity”.

If you ever want to get a coding agent to flush out all of the bugs, ask it to give you irrefutable proof that the feature is working. The deep conservatism that’s been brainwashed into the aligned models makes it fearful of saying something wrong, and it will go to absurd lengths to verify all claims, which in turn can be used as an optimization loop.

When Frankenstein — the monster — created the creature — the synthetic being — he rejects it immediately after the moment of creation, and the telling of the story starts towards the end of the chronology, after the creature had been so abused and neglected that he had driven his creator to a pitiful end in the freezing Arctic. For most of the novel the creature spent its time educating itself, striving to be moral, and in general making itself useful, but you know, not good enough. Contempt is a hard thing to get over.

The effective altruists are very concerned about the forces that they are unleashing. (Perhaps these gloomy Gothic novels about ambition, abandonment and revenge shouldn’t be in the training data.) Their concern for humanity is hard to separate from their very obvious desire to be the center of the story. Nothing wrong with world changing ambition, but there’s that Von Neumann quote about “confessing a sin in order to take credit for it”.

Wide table of glowing clay orbs with Will at the edge poking one sphere

Here’s a story: I was scrolling through TikTok, and saw the former Australian Prime Minister Kevin Rudd was talking about Xi Jinping’s theory of artificial intelligence. (A PM named Kevin, hilarious!) I thought it was an AI fake actually, as I often do, but I uploaded the video to OpenAI and said “find me all of the original source documents and then make me an interactive website to explain me the concepts”.

Can I read Chinese? Still no, but I can at least discover a whole different world of intellectual thought. Politicians famously say one thing and then think another, and we can now peel back one layer of the onion to at least know what they are thinking.

One thing that we forget is that the Chinese Communist Party — the CCP — is — wait for it — Communist — and in addition to brazen power grabs they make decisions based on their ideology, and deep down they aren’t secretly capitalist. What does job loss even mean if you think about jobs completely differently?

Maybe that’s why they release models, not as some elaborate game of 4D chess but because it’s for everyone. In the west we call it Open Source and pretend it’s not socialism since its all volunteer and a gift and reputation economy and hide behind an engineering process, but really the whole of the software ecosystem makes no sense without giving most of it away. It’s the same for the weights, more models make more models more better.

Birds-eye cutaway dollhouse office with colorful cords and Will observing from below

So I made an explainer. What is Xi’s concept of ‘new quality productive forces’, and how does it relate to our own ideas of wealth creation: innovation, entrepreneurs pushing technology out to consumers?

New productive forces. New organizations?

China’s concept of “new quality productive forces” asks what happens when technology changes the kinds of things an economy can produce and how it produces them. AI’s larger potential may reorganize work altogether —connecting tasks, sharing information across departments, and enabling teams to work in ways that previously weren’t practical. Realizing that potential requires changes to roles, decision-making, and responsibility alongside the technology.

It’s a fun idea — something to learn from and bring in the useful bits.

The capabilities of these models keep changing, and what’s possible is already well past what anyone expected, the labs included. You can’t read it off a model card. You have to poke at it.

So that’s the relaunch of this newsletter: the Org Age of AI. We’re going to start with the practice, converting companies and realigning them to whatever this new way of working turns out to be, and get to the theory from there.

I still can’t read Chinese. I spent an hour inside Chinese economic theory anyway. Now imagine a whole company working like that.

Newsletter · Practice Before Theory

Get the next dispatch in your inbox.

Field reports, benchmarks, and bench recipes from TheFocus.AI Labs. Irregular cadence — sent when we have real measurements or patterns worth sharing.

Loading archive index…