The AI jobapocalypse is cancelled. Token economics may put human brains back in front

The AI jobapocalypse is cancelled. Our human brains might outcompete frontier LLMs starting this fall. It's simple economics.

Here is a reflection I'd like to share with you.

If you've followed AI labs & access providers: token prices are going up. Or you get fewer tokens for the same money. Anthropic has been doing the second one gradually for months.

Here is the trend in numbers.

GitHub Copilot Pro+: €40/month. Under the usage-based billing that starts June 1st, the same heavy agentic usage at retail token prices runs closer to €960/month. Roughly 24×.

Claude Code Max 20×: $200/month. Equivalent retail API value for a heavy user: about $5,000/month. Roughly 25×.

But what does that actually mean?

Today's smartest LLMs get better by thinking longer on each question. DeepSeek's R1 uses the same base model as V3, just with reasoning added. More thinking means more tokens. When the subsidy ends, that cost shows up on the bill.

Jensen Huang (NVIDIA's CEO) on the All-In Podcast: "If that $500,000 engineer did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed." Huang meant this as the bull case at subsidised prices: half of base pay in tokens to amplify the engineer 10×. At 25× retail, the same usage runs $6.25M against a $500K engineer. The math inverts.

Let's look at two categories of work and what will happen with token costs of proprietary models.

A. Routine work.(i.e. customer service tier-1, document classification and extraction at volume, routine contract review.)

I think, here, AI will easily win against humans on price. "If you work like a robot, a robot will take your job." Gerd Leonhard.

B. Complex work.(i.e. senior software engineering on unfamiliar codebases, complex contract drafting and negotiation, multi-source strategic synthesis.)

At retail token prices, the human re-enters the competition. It's possible that humans might claim back some of the high cognition tasks.

At the same time, how the tech evolves from here is open.

Open-source may overtake proprietary or stay at least competitive.

Two paths for a 2026 H2 budget.

1- Rent AI from 1-2 AI Labs

What will happen? The AI oligopoly Anthropic, Google, OpenAI will capture the majority of the value AI generates by increasing prices, but just as much so that using their AI will still be slightly economical for you compared to not using AI.

They can because once you're locked in: prompts, tools, agent scaffolding, evals. They will have the power. The longer you stay the more stronger the lock-in.

2- Set-up your own Sovereign AI infrastructure

How does this look? You plug in open and closed models per task. Frontier where it earns the price, open-source for the rest.

The harness is yours. You hold the power. Costs stay lower as you scale, you switch models any time, you stay in charge.

Token economics killed the jobapocalypse.

But your bank account and your independence are now in danger.