Taste & Token Cost Controls

Taste & Token Cost Controls

Taste

If you spend any time in the Valley, you've heard the discussion about how, in the age of AI, the only thing left is "taste". But no one has defined what that means.

Taste is a feeling, not a set of requirements. But it's also a Persistent Difference (see our prior posts on the definition of Life).

If you copy others, you have no taste. You're copying someone else's taste. Whether that's fashion, a user experience, a product, a tagline, a story, design, code, etc. You're infringing on someone else's copyright. Which means there is no Difference, by definition.

If you're different but so crazy nobody cares or wants it except you, it will eventually die out. Meaning you are a special snowflake, which I'm here for, but you're not Persistent.

Great products, therefore, maintain a Persistent Difference. And that's what defines taste. Case study: iPhone.


Token Cost Controls

We've gone from tokenmaxxing to tokenminimizing in only a few weeks. Reactions have been robust and swift. People were let go. People were cut off from access to frontier models. This is happening in the enterprise and federal governments.

Why? Because tokens cost: money and national security.

So you, as an investor or operator competing against the best in the world, need an edge. How do you get it?

You start using open source AI models like GLM-5.2, you buy your own high-memory local computing cluster (whether multiple M3 Mac Studios with 256GB or 512GB running MLX if you can even find them in stock, 2 NVIDIA DGX Spark boxes at 128GB each connected via a proprietary cable, or an M5 Max MacBook Pro with 128GB unified memory). And if you're really going for it, solar panels and generator, and Starlink for connectivity. Grab Ollama, Codex, and you're off to the races.

The flex is no longer that Range Rover or big ole monthly mortgage. Those assets produce far less value than running your own local AI cluster. The former is a tax with no or low growth. The latter lets you become the master of your own domain: build, sell, grow, reinvest, flywheel.

Now all you have to worry about is an EMP and Faraday cages.

The benefit being:

  • You can run agentic coding without being connected to the grid to build whatever you want whenever you want with zero variable cost. Push it until the computer literally breaks.
  • You can begin training your own AI models for your proprietary products. Defensibility, eat your heart out.
  • You've simultaneously removed your downside risk and removed the ceiling on your upside. Hedge funds, I see you.
  • And nobody can turn you off. Self-sovereignty, the Bitcoin way.
  • Break-even periods are within months to definitely less than a year compared to running frontier models in the cloud. CFOs are our best friends.

The future is here. It's just not evenly distributed.

--Sean

Previous
Previous

Drunk Driving Emerging Tech

Next
Next

We Build Fortune 200 AI-Native Platforms