The Magic Combination Is Time and Curiosity

I fix cars for a living. I read one paragraph in December that explained why some of my friends are running agents and some are still asking a chatbot how to cook a turkey. The paragraph was written by a man who helped start Anthropic.

Jack Clark co-founded Anthropic, the company that makes Claude. He also writes a newsletter called Import AI. On December 22, 2025, at the bottom of issue 438, he put a short personal essay called "Silent Sirens, Flashing For Us All." He posted it on X the same day.

He opened by admitting something. He had a newborn and a toddler. He was doing midnight feeds and keeping a two-year-old from jumping off the furniture. For the first time in years, he had stopped thinking about AI every hour of every day.

That is the setup. A man at the center of the whole thing, and even he had run out of time.

The paragraph

Here is the part that stuck with me. I am quoting it the way it has been passed around since, and I would tell you to go read the whole thing.

"Most of AI progress has this flavor: if you have a bit of intellectual curiosity and some time, you can very quickly shock yourself with how amazingly capable modern AI systems are. But you need to have that magic combination of time and curiosity, and otherwise you're going to consume AI like most people do - as a passive viewer of some unremarkable synthetic slop content, or at best just asking your LLM of choice 'how to roast a turkey and keep it moist' ... all the amazing advancements going on are mostly hidden from you."

Read that again. He is not saying the tools are hard. He is saying the tools are hidden. Hidden behind a door that only opens for people who have two things most adults are short on.

The funnel

Clark describes what it takes to see the real capability as "the end of quite a deep funnel, and one which narrows a lot." He lists what narrows it. I will put it in my own words.

1. How curious you are. Do you poke at things, or do you wait to be shown?
2. How easily you can get to a good model. Can you afford it, and is it in front of you?
3. How well you turn curiosity into a task. Can you say what you want in a way the machine can act on?
4. How much time you have to experiment. Not to use it. To play with it.

Look at that list. One of the four is about the technology. Three are about you. Your habits, your words, your calendar.

That is the whole reason I am writing this. The people I know who are getting somewhere with AI are not smarter than the people who are not. They are not more technical. Half of them cannot program. They put in the hours, and they were curious enough to ask the next question instead of stopping at the first answer.

The prediction

Clark made a bet in that essay. He wrote:

"By the summer I expect that many people who work with frontier AI systems will feel as though they live in a parallel world to people who don't."

He compared it to crypto. The crypto economy moved fast compared to everything around it, but it never touched most people's real lives. He said the AI economy would move just as fast, with one difference. It already touches ordinary work.

Summer is over. I can tell you how it landed from a repair shop in Portland.

I have a friend who runs a shop like mine. Good operator. Sharp. He tried a chatbot in 2024, got a wrong answer about a torque spec, and decided the whole thing was hype. He has not touched it since. When I tell him I pointed an agent at my Shop-Ware export, asked it to find every ticket over two thousand dollars from the first quarter and draft a follow-up email, and it did the whole thing, he looks at me like I am describing a dream I had.

We are in the same trade in the same city. We are in parallel worlds. Clark called it eight months early.

Why the gap is a habit

Here is the thing nobody says plainly. The gap is not a talent gap. It is a compounding gap, and it compounds on time.

The person who tried one strange task in January is not one task ahead in September. They are a hundred tasks ahead, because every task taught them the next question to ask. Their curiosity got fed, so it grew. Their "translation skill," Clark's third item, got better every week because they used it every week.

Meanwhile the person who asked about the turkey got a good turkey and learned nothing about what else was behind the door.

I saw this on my own bench. In August my Tesla's battery died on Father's Day, and I did what thirty years of wrenching taught me: I photographed everything before I touched anything. Two months later I pointed Claude at that folder of photos and asked for a training deck for technicians. First shareable version took about ten minutes. It was also wrong. It told the story like a mystery, symptoms building to a big reveal at the end. Only I knew that we had the full diagnosis before the car left the driveway, and that ordering was the entire lesson. I told it the real timeline. It rebuilt the deck in a couple of minutes. One more correction and it was done inside the hour. I wrote the whole thing up here.

That was not my first job for an agent, but it was the one that taught me the rule. The machine does the assembly. I do the two or three corrections only a person who was there can make. Once I understood that, I stopped trying to write the perfect request and started handing over rough jobs and fixing what came back. That is the skill. It did not take intelligence. It took an hour I could have spent on YouTube.

The bottleneck moves to the humans

There is a side of this Clark did not write about, and I only see it because I am on the fast side of the gap.

When you move at the speed these tools allow, the hardest part of the day stops being the work. It becomes the explaining. I can hand an agent a job at breakfast and have it done by the time I get to the shop. Then I have to tell a person about it. That takes longer than the job did.

It is not that people are slow. It is that there is nothing to attach the explanation to. When I say "the orchestrator checked the agents' work," there is no shelf in their head for that sentence to sit on. So I back up. I explain what an agent is. Then what an orchestrator is. Then why the check matters. By the time I am done, I have spent twenty minutes on a thing that took me four to do, and they are polite and a little glazed, and I can tell most of it did not land.

Multiply that by everyone I work with. Employees. Vendors. Family. The two friends I have been trying to bring along. I am not stuck on the technology. I am stuck on the humans, and I am one of them. I do not have the time to bring everyone up to speed, and they do not have the time to get there on their own. That is Clark's funnel again, seen from the narrow end.

It is the loneliest part of moving fast, and nobody warns you about it. So consider this the warning. If you cross over, you will spend a surprising share of your new free time translating for people who have not. Be patient with them. A year ago you would not have understood you either.

Two of the four you can fix this week

Go back to the funnel. Access is mostly solved. Twenty dollars a month buys you a frontier model, and the free tiers are not bad. Curiosity you either have or you do not, and I think most people have more than they use.

That leaves the two you can change on purpose.

Time. Block it. Not "when I get a chance." An hour, on the calendar, with a real problem from your business. If you do not put it on the calendar, the turkey question is all you will ever ask, because that is all there is time for.

Translation. This is the one people skip. Turning "I wish somebody would deal with the no-shows" into a task a machine can run is a skill. It is the same skill as writing a good work order. You get better at it by doing it badly a few times. The model will help you. Tell it what you want and ask it what it needs to know. It will ask you good questions.

Silent sirens

The title of Clark's essay comes from a little story he tells. He is walking a ridge above a city at dusk. Traffic hums. A siren goes by, so faint it blends into the birdsong. Then the lights of the city go out. The cars stop. When they start again they move differently, in order, like a conveyor belt, and a voice comes out of every radio and says do not be alarmed.

That is a story about a world changing all at once while people hear the sirens as background noise.

He also walked around his own town and saw no drones, no robots, no self-driving cars. The change is real and you cannot see it from the sidewalk. You can only see it from inside, and getting inside costs an hour and a question.

I do not think you should be alarmed. I think you should be curious. The sirens are flashing. Most of my friends have not looked up yet. You do not need to be a programmer to look up. You need a Tuesday.

If you want to start this week

The essay is "Silent Sirens, Flashing For Us All," the personal note at the end of Import AI 438 by Jack Clark, December 22, 2025. He posted it on X the same day. The Hacker News discussion is worth a skim too.

I'm not an engineer. I fix complicated machines for a living and explain them to people who just need theirs to run. If you want help finding your Tuesday, that's what Wildash does.