Anthropic Leaked Its Plan to Ruin You

As Warren Buffett used to say: “Never ask a barber if you need a haircut.”

Well, this week, AI’s barbers decided we all needed one.

A few days ago, two of the most influential people in AI decided to invent a new job title. Again.

Peter Steinberger, the creator of OpenClaw, who just got hired by OpenAI, posted a tweet telling everyone to stop writing prompts. Prompts, according to him, are already out of style.

Now the cool thing is loops.

For anyone who’s not technical, let me explain quickly and in plain terms what a loop is.

It’s telling an agent: “do this, check it yourself, and if it comes out wrong, redo it until it’s perfect. Don’t bother me until you’re done.”

The agent works alone: it tries, fails, fixes itself, tries again.

And under the hood it calls other subagents to run tasks in parallel, but basically that’s a loop with agents.

Sounds beautiful. Except there’s one small detail.

Every attempt costs money, and these days it costs a lot.

They bill you in the currency of the 21st century: tokens.

Up to this point, Peter’s take could’ve passed for just another opinion on Twitter.

The problem is he wasn’t the only one. He wasn’t even the first.

Weeks earlier, Boris Cherny, the creator of Claude Code and an Anthropic employee, had already been saying the exact same thing in interviews.

What you just heard is a guy telling you, completely calm, that he leaves his agents working alone in a loop. Like leaving the stove on.

I’ll say it again, it sounds incredible, it sounds like dystopian science fiction.

But there’s one detail neither Peter nor Boris mentions in any of their tweets or interviews. Neither of them has had to pay the token bill in months, maybe even years, I’d guess.

They work at the two biggest token dispensers in the world. What did you expect them to say? That you should save tokens? No.

Two people who clearly have an open bar of tokens.

For them, burning 50,000 dollars a month on tokens isn’t an expense. It’s just a Tuesday at the office.

Do you really think they lose sleep waiting for their weekly quota to reset?

No way. They’ve got unlimited balance, burning tokens left and right.

And when someone with that much influence tweets, the reactions come fast. Twitter exploded.

And it exploded in two directions. The first: the memes, obviously.

Within hours, Twitter was flooded with irony.

Where are all the Loop Engineers now?

New job openings for Loop Engineers.

Comparing vibe coding to slot machines.

And the memes, as always, showed up before any actual analysis.

Because behind the joke there was a pattern everyone could smell.

In this industry, a new job title is born on LinkedIn every 3 months.

Six months ago we were “Prompt Engineers,” three months ago we became “Context Engineers,” and this week we turned into Loop Engineers.

I’ve got more titles than a soap opera character at this point.

We’re starting to look like seasons of a TV show.

But the second group is what actually worries me: the people who took this literally.

In under 24 hours, LinkedIn bios got updated, articles got rewritten about how to become a Loop Engineer, and endless threads popped up about the “7 secrets of Loop Engineering nobody tells you.”

And guess how those articles got written? Exactly: with loops, ha.

It takes exactly 72 hours from tweet to course. There are already dozens of YouTube courses teaching you how to become a good “Loop-er.”

And here’s the thing about Twitter, it’s both its best and worst quality at the same time: it forces every opinion to become extreme.

Either you’re a token-maxxer running 30 agents on a loop while you sleep, or you’re some prompt dinosaur still typing prompts by hand like a caveman.

There’s no middle ground. Why? Because the middle ground doesn’t get views or likes.

But let’s be honest, if all of this goes viral this fast, it’s not because of the gurus.

It’s because it taps into a wet dream we all have.

The dream of sipping your coffee in the morning, in your bathrobe, everything just hanging out, totally relaxed, while your farm of 30 agents builds the next Airbnb for you.

Making money while you sleep, what a treat.

They used to call it dropshipping.

Then it was called crypto.

Today it’s called an agent farm.

The dream is exactly the same. The only thing that changed is who’s selling you the shovel.

And spoiler: the only thing your agent farm is actually generating while you sleep is the bill you’re going to owe Anthropic at the end of the month, the one that’s going to bankrupt you.

And as if that wasn’t enough, it didn’t stop at two tweets.

More voices started piling on, and they all came from the same place.

Tibo, from the Codex team, also an OpenAI employee, quoted Peter’s tweet, joking about whether anyone was writing nested loops yet.

Employees of the two biggest token dispensers in the world, cheering on the snowball.

And in the middle of all that circus, some guy named Thiago, I’m guessing just a random developer, asked the only honest question in the whole conversation:

How do I do this without infinite tokens?

And you know what Peter answered?

“Don’t expect me to solve every problem for you.”

That answer, those eight words, sum up the state of the entire industry right now.

If this keeps going, we’re going to end up exactly where Sam Altman already predicted:

soon every single human is going to have to pay a new monthly bill, just like we do for electricity, gas, and water.

We’re going to have to pay for our monthly tokens as a basic utility, the same way we do now with internet service.

And right when this whole snowball was at its peak, something happened that made me stop believing in coincidences.

On June 9th, Anthropic launches Claude Fable 5, the smartest and most expensive model to date.

The token costs double what Opus 4.8 costs. Double.

And guess what happened a few hours after Fable 5 launched?

Yep. Another Anthropic employee, Lance Martin, published a new article pushing, once again, working in loops.

But now using Fable.

Using the model that costs double. Recommending the technique that burns double. Coming from someone who pays zero dollars for every token spent. Great.

Now before you call me a conspiracy theorist, let me be fair:

I’m not saying this new loop based way of working is bad.

Quite the opposite, it might actually be the future. There’s a real case for working this way.

The problem is pushing other people to try it when you’re experimenting with free tokens that daddy Anthropic, or daddy OpenAI, hands you.

It’s like when your dad lets you borrow his truck. You crash it and nothing happens, because dad’s got four more trucks in the garage.

The problem is you and I don’t drive dad’s trucks.

We drive our own car. In a lot of cases, financed over sixty monthly payments.

So here’s my advice: set up your loops.

Read every article you find interesting on the topic.

Adapt them to your own workflow. If it makes you more powerful, great.

Let them run overnight if you want to.

But the looping agent should execute the decisions. It shouldn’t be the one making them.

Because judgment isn’t something you delegate.

Boris and Peter can afford to total their dad’s trucks as many times as they want. They’re not the ones paying for it.

You and I can’t. We’re the ones who have to pay the bill at the end of the month for every token we burn.

So drive carefully. Thanks for watching.