Prompt engineering is dead.

For two years, using AI meant prompting. You type a request, read the answer, and type the next one. You're in the driver's seat for every step, and the whole skill was writing better prompts.

But that era is ending.

It started with one tweet from Peter Steinberger, "stop prompting your agents, start designing loops that prompt them," and it cleared 8 million views.

The proof it's real is Boris Cherny, who built Claude Code. He deleted his code editor back in November and shipped a full month of work without typing a line himself.

So what is a loop?

A loop is a small program that does the prompting for you. You write it once: prompt the model, check the result, decide if it's done, run it again if it isn't. Then you walk away and it works on its own. You stop being the worker and become the person who built the worker.

But there's a twist. When agents run themselves in a loop, your costs don't always drop, they climb, because the loop runs hundreds of times while you're not watching. Without the right guardrails, you're just draining your account.

So the important skill now isn't writing prompts. It's controlling the loop.

But Eric, how do I control these loops?

Which brings me to my next point.

Tomorrow I'm sitting down live with Matt Van Horn, to finally explains all of this.

If you use AI at all, even if you never write a line of code, this is the shape of every tool you'll use over the next year.

We're diving deeper into what a loop actually is, why the people building the frontier have stopped thinking in prompts, and the layer most people are still completely missing.

Join us LIVE tomorrow at 9:30 AM PT / 11:30 AM CT / 12:30 PM ET

Register HERE

See you there,

Eric Siu