Goal Engineering
The internet just discovered loops. Four were already running on my laptop.
The quiet shift from prompting AI to giving it goals, and why it matters to you before it gets cheap.
The internet just discovered loops. Four were already running on my laptop.
The quiet shift from prompting AI to giving it goals, and why it matters to you before it gets cheap.
While the internet spent last weekend discovering loops, four of them were already running on my laptop.
I did not plan it that way. I found out the way you find out you have been speaking prose your whole life. Two of the most-watched people in AI, one from Anthropic and one from OpenAI, posted within hours of each other that they have stopped telling their AI what to do. One tweet hit five million views in a day. I read it, looked at my own machine, and realised the thing they were calling the frontier was quietly humming away while I made coffee.
By the end of this you will know what changed, why it matters for you in particular, and how to tell whether you are still doing the old thing or the new one. No jargon you have to swallow to keep up.
We have gone from telling AI what words to use, to telling it what to know, to telling it what done looks like and letting it find its own way there.
Let me slow that down, because each step is a different way of being in charge.
The first way: you stand over its shoulder
You write a good instruction, the AI answers, you write another, it answers again. This is prompting, and most of the world still lives here. There is nothing wrong with it, it is just slow, and it ties your hands to the keyboard.
The second way: you furnish the room before the guest arrives
This one is quieter, and most people never notice they have started doing it. Instead of perfecting every instruction, you start curating what the AI knows before you ask it anything at all. The files it can see. The notes about how you like things done. The memory of who you are.
When I set up my own AI, I wrote it a document about my voice, my values, the way I work, and after that I stopped explaining myself every time. The clever name for this is context engineering. You might just call it finally not having to repeat yourself.
The third way: you describe done, and walk away
This is the one that went viral, and it has a plainer name than it deserves.
You stop describing the steps. You describe the finished state instead. Then you give the AI a way to check whether it has got there, and a moment to begin, and then you walk away, and it runs itself until it is done.
That is a loop, and it needs only two things, something to start it and a goal it can actually check.
The checking is the whole trick, so hold onto it, because a goal you cannot verify is really just a wish. "Make this good" is a wish. "Make every test pass" can be checked by a machine. The loop knows when it has arrived, the way you know a cake is done because the skewer comes out clean, and if there is nothing that can be the skewer, then there is no loop, only a machine quietly guessing.
And only three things can start one. Something happens, like a form being filled in. A schedule fires, like every Sunday at nine. Or a human says go. That really is the whole list.
My four loops, the ones I did not notice
So what were they, the ones running while I thought I was just a person with a busy laptop?
One scans for dangerous software every night at half past two and stays silent unless it finds something, in which case it pings my phone. One runs my task board's AI cards by itself, every hour, no clicking. One checks my own writing for the tells of a machine the moment I finish a piece. One quietly logs my working hours from the day's transcripts so I never have to time myself.
Each one has a trigger, a goal, and a way to check the goal, which is what makes each of them a small loop. I built them one at a time, for my own convenience, and only when the rest of the world started shouting the word did I see what I had.
Why I am telling you now, even though it is expensive
Here is the part that actually matters, because the people demonstrating this on stage work at companies that hand their staff effectively bottomless budgets to play with. One of them showed a monthly bill of 1.3 million dollars. They are the top sliver of the top sliver, and they are allowed to burn money the rest of us are being told to save, which makes the timing feel almost insulting. Everyone is cutting their AI spend, and the frontier is saying spend like water.
Here is why I still want you to understand it now, today, even if you never run an expensive loop in your life.
Because what is costly today is always cheap tomorrow, and every single time, with every technology, the price falls. The model that shipped this same week, the one built precisely for this kind of long, self-running work, already costs less than half what its predecessor did, and it remembers what it learned last time, so it gets better each round instead of starting blind.
The women who understand the shape of it before it is cheap are the ones who will use it the moment it is, instead of spending another year catching up to a word.
The one honest catch
There is one catch, and it is the part the courses tend to skip right over. The easy loops are genuinely easy. "Check my work and fix the obvious problems" is a loop you could set up this month. But "build my whole business while I sleep" is not, and anyone promising it is selling you the 1.3 million dollar dream without the 1.3 million dollars.
The hard part is defining done for a goal that has no clean finish line, because get that wrong and the thing runs forever, never sure it has arrived, quietly spending. Respect that, and you are already ahead of most people shouting about loops this week.
The question worth carrying out of this
Right now, a human still decides the direction. I choose the goal. The AI finds the path, but I point at the horizon.
The open question, the one nobody can answer yet, is whether that stays true. The day a system can choose its own goal, with taste, and design its own loops to reach it, is the day the last human job in the room changes shape entirely. We are not there. But for the first time you can see the road that leads there, and you get to decide now whether you want to understand it or be surprised by it.
If you want to walk it on purpose instead of being dragged along by the headlines, I built the hands-on version of this, where you design your first real loop with your own two hands and watch it run. It is free and open, no wall: the Goal Engineering masterclass. Same plain words. Same respect for your time. You leave with something running, not just something understood.
That is the difference I want for you, and it is a real one. You get to stand inside the future, with something of your own quietly working away while you make the coffee.
Originally published at https://herfrontier.com.