What did you watch this weekend?

What did you watch this weekend?
I watched an AI model burn through $200 in a few minutes.

$200... gone.
On a test run.
For a silly research project.

- Yes, I had auto-billing approved.
- Yes, I set a monthly cap (phew).
- No, I didn’t expect it to drain everything in seconds.

It reminded me of the famous line from Charlie Munger,
"Show me the incentive, and I'll show you the outcome"

In this new world of AI, incentives still matter.

Token-based billing is the new hourly rate.
Except it’s unconstrained by time or speed.

Token-based systems are not incentivized to:
- be concise
- filter out noise
- stop at “good enough”

Just like charge-by-hour services, efficiency gets punished.

We're in the honeymoon phase of this technology.
All fun and good vibes.
Little scrutiny on spend.

But if digital workers are going to sit inside real businesses, we need new standards for evaluating value.

Cost per token isn't it.
Value-based and performance-based economics.

- What is the maximum rational spend for a useful result?
- What does one more unit of real progress actually cost?
- What is “good enough” mean?
- Does success here make the next run cheaper?

Speed without standards isn’t leverage.
It’s just burn, delivered faster.