Pete Sena

Nobody is going to care how many AI seats you bought.
 
Your board will ask one question:
 
What changed?
 
Not how many people logged in.
Not how many prompts ran.
Not how many tokens got burned.
Not whether the team “likes it.”
 
What changed?
 
Did a workflow get faster?
Did a decision improve?
Did a customer experience change?
Did revenue move?
Did margin improve?
 
That is the number.
Everything else is theater.
 
The most common AI mistake I see right now is companies confusing activity with strategy.
 
They buy the licenses.
Send the login.
Run the workshop.
Watch the invoice grow.
 
Then someone asks about ROI and the room starts telling stories about the future.
Before you sign another AI contract, ask three questions:
 
What changed?
What is it worth?
What do we want next year?
 
I wrote the full breakdown on Substack. (link in first comment)

Silicon Valley is token-maxing.
I’m outcome-maxing.
 
Here’s the difference:
I can tell you what AI cost yesterday.
And what it changed for tomorrow.
 
If your CFO asked right now,
“How much did we spend on AI last month?”
Could you answer?
 
Not just the obvious subscriptions.
The whole iceberg.
 
Embedded AI tiers you forgot you upgraded.
Copilots inside tools you already pay for.
Contractors doing work in personal AI accounts.
API calls.
Agent loops.
Shadow tools.
 
You don’t have a tools problem.
You have an awareness problem.
 
Spend that produces learning is tuition.
Spend that doesn’t is burn.
 
So I run a tiny ritual with receipts:
 
1. Run /cost (see the receipt)
2. Run /insights (see the work)
3. Answer 3 questions:
- What worked today
- What didn’t
- What I’d change tomorrow
4. Write one change for the next day
 
That’s it.
90 seconds.
Compounds fast.
 
And before any workflow, I run Token Triage:
Does this need a genius, a specialist, or a script?
 
Stop paying Ferrari prices for warehouse work.
 
AI is the gear. You’re the guide.
 
I wrote the full operator stack on Substack (frameworks, prompts, templates). Link in the first comment.

A picture is worth a thousand bucks.

Claude Fable 5 is brilliant. It's also expensive as hell.

A library called pxpipe found a weird crack in the billing model:

Turn dense context into an image.

Not a pretty image.
A packed PNG full of code, tool docs, system prompts, READMEs, and session history.

Here’s the trick:

Text gets billed by text tokens. Images get billed by visual tokens.

Visual tokens mostly care about pixel space.
They do not care how many tiny words you jam into that space.

So a giant wall of context gets packed into a small image and sent to the model for less.

The pxpipe repo claims roughly 25,000 text tokens became about 2,700 image tokens.

It's a quick hack but also a sign of what's next:

Most companies are about to waste insane amounts of money on AI.

Not from bad prompts or wasteful teams. It'll come from lazy architecture.

Your team is sending every file, every meeting note, every customer record, every policy doc, and every Slack thread into the most expensive model. That's like driving a Bugatti to the grocery store and being mad at your gas bill.

The best teams think like operators, not AI tourists.

They ask:
What context matters?
What gets cached?
What gets compressed?
Which model is enough for this job?
What never needs to be sent?

Token spend is the new media spend.

The companies that track it protect margin.
The companies that ignore it bleed money in silence.

AI isn't expensive. Careless AI is expensive.

Receipts:
pxpipe GitHub: https://lnkd.in/e-ZzMSA7
Anthropic vision token docs: https://lnkd.in/eAAHmScF
Anthropic Fable pricing: https://lnkd.in/egRTU4hi
Anthropic pricing docs: https://lnkd.in/edVa2k3M

Wanna act like an operator not an AI tourist? Follow Pete Sena for more AI education.

I just caught someone using AI.

So I did the brave thing and outed them. Reported them to AI police chief Karen. Shouted it to my whole network on blast.

They had a thought. So they talked it out. And instead of scrolling in silence they posted something. But they asked a tool to help them clean it up.

Sickening. How dare they!

Hate to break it to ya, AI Karen, but my best drafts start by talking out loud. I'm not a great writer... yet. But I'm a damn good thinker, and I work on it daily. AI is just another tool. And yes I use it in my process.

Real writers know the pure way to share ideas:

scratch them into stone
ship them by mule
reject spellcheck
never borrow cadence from anyone who came before you

Just raw, untouched genius. Preferably written in cursive.
But guess what... People have copied people forever.

We borrow structure. Steal rhythm. Repeat stuff our parents said. We quote books. All genius is stolen no matter what machines are used to make it so.

Ai is a parrot. A reflection of human works. All our genius. All our stupidity. Stolen genius parked into a latent space.

AI slop is real and it's damn cringe. But hate to break it to ya but every smart, successful creator today is using AI somewhere in their process. But the best of them either choose to hide it or made it invisible to you.

Paste the AI tells that give you the ick most below. I'm sure there's dozens of "It's not X it's Y" phraseology that boils your blood. Drop it below...

This post was written by a human. A very flawed very curious human. Follow Pete Sena for more stuff like this.

Everyone says keep a human in the loop.

That turns you into a spell-checker for a machine. I say F**k that.

Here's the better frame.

AI raises the floor.
Your job is to raise the ceiling.

The floor is competent.
And AI gets anyone to competent in minutes.
First drafts, passable strategy, decent design. Done.

So competent isn't the job anymore.
The job is the standard the model can't hit on its own.

The taste that makes someone feel something
The call that wasn't in the data
The bar you refuse to lower

You're not the approver checking its work.
You're the ceiling it keeps falling short of.

And your old role was never really your job.
It was sized to your limits. How much one person could push in a week.

Take the limit off and the question flips.
Not "which of my tasks survive."
But "what was I always capable of and never had room for?"

AI didn't shrink the work.
It handed back the part that was always yours.

What would you do at work if competent took zero time?