If intelligence is nearly free, where does the value go?

TLDR: The investment thesis of WunderOS is now available; read it here. I wrote it to be endorsed, which is to say I wrote it to be repeated. Read on if you want to know why.

I have been accused of being stubborn and rightly accused. For the longest time I thought the easiest way to get something done was to do it myself. Then I learned that other people doing the things you want is a force multiplier, and I was stubborn about that too, about the best way to get them to do it.

What I learned, oh so slowly, is the only rule on the subject worth keeping: the way to get a person to do X is to make X the easiest thing for them to do.

Want someone to buy your software? Make it easier to buy and use than to refuse. Want someone to endorse your thesis? Make the thesis easier to endorse than to ignore.

I have been applying that rule lately to capital formation. When I think about how WunderOS gets funded, and who its partners should be, the tempting mistake is to make the ask…“believe me.” Which asks the investor to do the work, and it misreads the work. An investor’s real work is not done in the room with me. It is done in the room I am not in: the partners or investment committee meeting where someone who liked the pitch has to stand up and defend the check to people who never heard the pitch from me.

The WunderOS investment thesis

The thing to make easy is not understanding the thesis. It’s repeating the thesis. Because I’m stubborn, it took me too long to learn the most important lesson of all: a thesis simple enough to be repeated by someone who is not me is a thesis that survives the second room.

So this week I wrote the thesis down, in public, and held it to the rule my blog posts follow: a claim has to be sayable in one sentence at the right altitude or it isn’t finished. If it needs a paragraph of warm-up before you can utter it, either the claim is wrong or I haven’t found it yet. Here is the sentence I found.

The cost of intelligence is collapsing toward zero, the cost of authorized action is not, and the value concentrates in the thin layer that turns the first into the second. WunderOS is built to be that layer for the regulated enterprise.

That is the whole of it, and the rest of this post records why it’s true.

The AI economy has three parts

Picture the AI economy in three regions.

1. On one side are the inputs to intelligence: compute, models, data, energy, capital. These are abundant and getting cheaper, and each new frontier model turns last year’s into a commodity.
2. On the opposite side are the outputs, that is, what a business actually pays for: a trial initiated, a component manufactured, a claim adjudicated, a mission completed. These do not commoditize, because each is bound to the particular rules and liabilities of its industry.
3. Between the two sits a narrow region inputs get turned into outputs, that is, where abundant intelligence is converted into outcomes someone will stand behind and pay for.

The venture people have a name for the shape, the bow tie, and the narrow middle is where the money is, because the inputs are too abundant to command a price and the outcomes are captured industry by industry, by the firms that already own the trial, the factory, the claim.

Scarcity and value, as always

What is both scarce and general is the conversion in the middle.

The scarce thing in the middle is not intelligence. It is permission. No regulated enterprise will let a cheap, abundant, non-deterministic process act on the systems that matter, on weapons, factories, clinical trials, money, without a layer that makes each action one it can authorize beforehand and prove afterward.

Frontier models produce plausible actions cheaply, and will produce them more cheaply next year. Producing an action a bank or a hospital can defend under audit is the hard part, and it is the part no abundant input performs on its own and no incumbent supplies to its rivals.

That is the middle, and building it is what WunderOS is.

None of this makes the thesis a small claim.

Fighting over the middle ground

The page argues the hard version:

1. the most valuable node is the one many industries have to cross, so the layer is a bridge and not a bet on any single vertical;
2. the timing is not incidental, since the inputs are commoditizing fastest right now, which is exactly what moves the value off the inputs and into the conversion; and
3. the layer compounds, because a deterministic system improves with use while staying provable, which is the one thing an abundant input cannot do and an incumbent will not build.

Simplicity here is compression, not omission. The one sentence carries the claim; the page in full carries the proof: pentad.ai/about/why. I wrote it to be endorsed. I wrote it to be repeated.

If you finish it and can say it back in a sentence, that was the whole intention, I did my job acceptably well, and you have my permission to use it as your own. Spread the love.