There Is No Plan

A question showed up on Reddit recently that I haven't been able to stop thinking about:

"If AI takes all the jobs and only 10 trillionaires own everything... who buys the products?"

It's not a hypothetical anymore. It's a macroeconomic math problem. And neither Wall Street, nor the Fed, nor the government has a credible answer.

Here's the math nobody wants to say out loud. A major blockbuster employs over 1,000 VFX artists, the people making $80k to $180k a year to build the shots that make a $200 million Marvel movie worth watching. Vehicles through debris. Complex physics. Action sequences that used to require entire teams and weeks of render time.

ByteDance just dropped Seedance 2.0. Curious Refuge stress-tested it and their verdict: "A shot like this would typically be insanely expensive. Seedance 2.0 did a great job." The SAG-AFTRA strike in 2023 was literally a fight over this. Studios agreed to "consult" before using AI on likenesses. They consulted.

The $13 trillion white-collar mortgage market built on those salaries doesn't have a borrower anymore. Corporate earnings look fine. The movie still comes out. But the money never makes it back to the humans who made it. Ghost GDP: profits that exist on balance sheets and nowhere else.

The endgame of that loop isn't just displaced VFX artists. It's a commerce layer where machines recommend products, machines fulfill orders, and machines track engagement. The human is fully removed from the transaction. Not replaced by a cheaper human. Removed.

That's the scary part. The liberating part is what happens after you accept it.

---

The End of the Middle

Most people assume someone is working on the AI displacement problem. No one is.

In 1919, the Supreme Court ruled in Dodge v. Ford that a corporation's primary obligation is to maximize shareholder profit. That ruling still governs corporate America. If AI increases margins, every publicly traded company is legally required to deploy it. Klarna replaced 700 customer service agents in 2024. Valuation surged. Every other CEO had a board meeting the next morning.

The disruption isn't going to look like a clean, linear transition. It's going to look like a barbell. The middle (the 50-person creative agency, the mid-market SaaS company) is getting crushed between hyperscalers squeezing compute margins from above and solo builders deploying platforms for 10,000 users in a weekend from below.

This leaves an economy of extremes: capital-rich monopolies on one side, hyper-agile one-person unicorns on the other.

There is no plan for the people caught in the middle.

---

The Inversion Nobody Sees

Every AI job-loss story follows the same script: AI does task, human is redundant, multiply by millions, collapse.

It's clean and logical. And it's missing something enormous.

When a technology drives the cost of execution to zero, it doesn't destroy value. It moves it.

When every phone got a camera, the cost of "taking a picture" dropped to zero. Did photographers disappear? The ones selling "I can operate a camera" did. But the ones selling their eye, their ability to see something the rest of us miss? They charge more now than ever.

The task became free. The taste became priceless.

Here's the part nobody in these AI job-loss conversations actually says: the people who built the models don't know how they work. The model that learned to identify a cat from millions of YouTube frames took a mathematical path that is entirely opaque. Even the Google engineers who ran the experiment couldn't follow it.

Nobody wrote those instructions. The machine wrote them for itself.

The opacity is structural. The system can't explain its own reasoning. So someone with taste has to stand between the output and the world. That role will never be automated.

You can't replace the judge with the thing being judged.

The cost of digital execution has hit zero. But someone still has to decide what to build, absorb the liability, and know the difference between brilliant and garbage. That someone is you. That role cannot be automated, because the value was never in the output. It was always in the intent behind it.

---

The 2028 Event Horizon

Right now, the physical world is buying us time.

The grid can't keep up. Data centers are behind schedule. High-bandwidth memory chips are on 18-month waitlists. The compute crisis is real, and it's the only thing slowing the spiral down. But in 2028, when the next generation of TSMC fabs come fully online and the supply floods the market, the price of intelligence crashes.

That's the event horizon. When compute goes free, the friction disappears. The reckoning begins. You have until then to build the thing that makes you impossible to replace.

2023's ChatGPT was the Netscape browser. Not the breakthrough. The interface. The internet existed before Netscape. Netscape just made it touchable.

The real inflection point was years earlier, for the people actually in the labs. Around 2016, before the public had vocabulary for any of this.

AlphaFold folded 200 million proteins in a single day. A problem that would have taken conventional science centuries. Nobody outside the research community noticed.

And taste is exactly what's next on the AI roadmap. Linguistic intelligence came first: read, write, translate, summarize. Intuition, emotional intelligence, deep reasoning are the next benchmarks.

The same layer we're calling taste. The moat is real.

But it's a window, not a wall. Current AI can't feel pacing. Can't sense when a scene lands wrong. Can't know the difference between technically correct and actually true. Not yet. The race is to build something real before that gap closes.

2028 isn't when AI gets powerful. It's when power gets cheap. Same pattern, bigger wave.

---

The Thesis in Real Time

I came across a reaction video last week that stopped me cold.

An animator has been working on "Anime War 2," hand-drawn frame by frame, for three months. He has three minutes of footage. The AI generated a fully rendered, fully voiced, three-minute fight scene in a single day.

Three months of work. Matched by a few dollars of compute.

Watch his critiques. He's not attacking the art. He admits it's gorgeous. Every single thing he hates is something the AI can't do:

"The pacing is horrible."
"It doesn't understand storyboarding."
"It's not cohesive."

This guy is literally holding the moat and doesn't even know it. The AI has infinite execution and zero judgment.

If he stopped competing on drawing frames and started using the AI as his studio (writes the storyboard, directs the pacing, edits the output) he'd be unstoppable. One person with taste and a render farm that never sleeps. But he can't see it. He's still measuring his value in frames per hour. And by that metric? He already lost.

He also mentions Seedance 2.0 has already been sued. Copyrighted characters like Goku and Gojo were scrubbed from the public models. The technology can produce a perfect Goku. Human copyright law says it can't. Same mechanism as the compute bottleneck. Friction that buys humans time.

At the end he says: "If you can just click a button and do the whole thing... you don't have any skill at all besides writing the prompt."

He's wrong. Writing the prompt was never the skill. Knowing the output has garbage pacing and having the editorial vision to fix it. That's the skill. That's the moat. The people who refuse to see it will feel exactly what he's feeling right now. The ones who embrace it become One-Person Unicorns.

---

Be the Orchestrator

The system has no plan for you. Good. That means nobody's in your way.

For the first time in history, one person with clear thinking and the right tools can build what used to require $5 million in VC funding and a team of twenty engineers. I'm doing it right now: two production platforms, one engineer, no VC. The creative side isn't far behind. Give it two years and 5-10 people with taste will produce what currently takes a thousand-person studio and $200 million.

The word I keep coming back to is Orchestrator. Not "the person who prompts AI." The orchestrator has agency, taste, and compute independence. They don't optimize old workflows. They build entirely new ones, guided by a point of view that no model can replicate.

That's what turns one person into ten. And it's what makes someone impossible to replace, even when intelligence itself is free.