A faster horse
Most of us got an upgrade this year. A few people got a different job without changing jobs.
Scroll LinkedIn for five minutes and you will find an AI playbook for every job there is.
Sales. Engineering. Product. The C-suite. There is a playbook for writing playbooks.
Everyone is an expert now.
I'm one of them. I've written some of those posts. I teach a course.
What most of them are missing is evidence. Not claims about what AI could do. People whose work actually changed.
I have some. Two people, to be exact.
What the room told me
This year I taught Beyond the Prompt to a few small cohorts. Each one starts the same way. I ask people how they actually use AI. Not how they think they should. How they do.
The honest answer, for most of the room, is search.
A better search box. One that answers instead of linking. Useful, a real improvement, and the same thing they were already doing.
By the end, most people were using it far better. Prompts with context. Projects with standing instructions. Work they'd been doing by hand for years now done in minutes.
It was like trading a Honda for a BMW.
Noticeable. Meaningful. You feel it every time you drive. And it's still doing the same job.
One person went exponential.
Not a better car. A different way of thinking about the work, and it changes what that person does every day. Not faster at the old job. A different relationship to the job.
The rest of the room upgraded.
The second person is me.
At the start of this year I used AI the way most of that room did. Today I run my work on a system of agents I built myself. It briefs me before I'm awake, drafts with me, argues with me, and remembers what I decided last month and why. I didn't get faster at my old job. The work itself changed.
That's the difference I keep coming back to.
The upgrade makes the work you already do better.
The transformation changes what the work is.
Most of us got the upgrade this year. That's not a failure. It's what an upgrade is for. But it's worth knowing which one you got, because the people selling playbooks are describing the second and delivering the first.
It wasn't skill
I expected the people who went furthest to be the most technical. That's not what I saw.
Skill is hard to pin down in a room like that. Some of the most capable people stayed in the BMW. Some of the least technical started asking questions that took them somewhere else.
The difference was curiosity. By a lot.
Not the curiosity of "what can this tool do." Everyone has that one for about a week.
The other kind. What would this work look like if nobody had ever done it before? What am I doing only because it's how it was done when I learned it? What would I try if I weren't afraid of looking like a beginner?
Skill gets you a better result from the same question. Curiosity changes the question.
The frontier
The companies building these models call themselves frontier labs. I think we should take the word seriously.
Frontiers have a known shape. A land rush. Very little law. Everyone selling maps. And the people who lay the railroads end up owning a lot of the land.
On a frontier, most people get a faster horse. A few start laying track.
That's this year, as best I can read it. AI is embedded now, the way the phone is. Nobody is asking whether to use it anymore. The open questions are how much we hand over, and what, and who decides. Nobody has enough evidence yet. So most of us are trying everything and seeing what sticks.
That's the subject of something the Studio is publishing in December: the first Annual Thesis. A State of the Union for people and the machine. What the year proved, what it only claimed, and a question for the people building the railroads.
This issue is my piece of it, from the seat of one person.
Praxis
This week, run the upgrade test on yourself. Fifteen minutes.
First, pick the thing you use AI for most. Write down how you did it before. Now write down how you do it now. If the second list is the first list, faster, you got the upgrade. That's fine. Now you know.
Second, take that same task to the machine and ask it a different question. Not "help me do this." Ask: "If this task didn't exist yet, and we were inventing it together from scratch for what I actually need, what would we build?" Then argue with whatever it gives you. Don't accept the first answer. Ask why three times.
Third, write down one thing you're doing only because it's how you learned it. Just one. You don't have to change it. You just have to see it.
That's what curiosity looks like when it has a job to do.
I need your help
I'm writing the Annual Thesis now. I don't want it to be one person's read of the year.
A state of the union written by one person is a speech. I want an inquiry.
Two cases is evidence. It isn't proof. That's the gap I want the Thesis to close, and I can't close it alone.
So I'm asking you directly. Three questions, five minutes:
Did AI give you the upgrade or the transformation this year?
Where do you think this is heading, for you and for the rest of us?
If you could ask the people building the frontier one question, what would it be?
If you ran the test above, you already have your first answer.
Answer the three questions here.
Or just hit reply. Either way it reaches me. I read every answer myself, and the best of them will shape what the Studio publishes in December. You decide whether I quote you by name, anonymously, or not at all.
This issue doesn't work without you.
The frontier doesn't stay open forever.
– Michael Founder, The Drop In & Author of 'Human Traits — a novel exploring humanity's relationship with AI'