My biggest lesson from two years of working intensively with AI has surprisingly little to do with AI. I have to relearn how to think.
Not how to prompt better. Not which model to use. Not how to build agents. How I think about what is possible in the first place.
I still catch myself having an idea or running into a problem and immediately thinking:
I don't know how to do that. That sounds complicated. That'll take forever. We'll need someone for this.
And sometimes that's where the thought ends.
Then, an hour later or sometimes days later, I catch myself: Wait. Why did I just assume that? Why didn't I ask?
Our intuition is running on old data
For most of my career, this reflex was useful. Experience teaches you to recognize complexity before you walk into it. You don't just see a feature. You see the architecture, dependencies, edge cases, people, time and money behind it.
Over decades, you develop a pretty good intuition for what things cost. That mechanism isn't stupid. It's one of the benefits of experience. The problem is that the environment it was calibrated for has changed.
AI changed what's possible faster than we changed our intuition about what's possible.
We're still estimating the difficulty of new problems using the cost of solving them in the old world. Experience still tells me what can go wrong. But I'm learning not to let it automatically tell me how difficult it will be to make it right.
I keep catching myself doing it
Recently, we were working on an internal platform for Founder.org. It connects information across the organization, orchestrates different agents, automates workflows and brings everything into one shared knowledge layer. Basically, a fully AI-powered operating system for the organization.
At some point, I wanted Claude to interact directly with the platform. We'd need an MCP. My immediate reaction was: Fuck. That sounds complicated. Who builds that? How much work is this? Do we really need it?
And then I caught myself. I had just decided that the problem was difficult without checking whether it was difficult now. So I asked. About an hour later, it worked.
The surprising part wasn't that AI could do it. It was how wrong my initial estimate of the problem had been.
And once I started noticing this reflex, I started seeing it everywhere.
Do I really need an agency to figure out how much of our marketing can be automated?
Do I need to fully understand something like loop engineering before I can apply it to a product I'm building?
Is connecting a payment terminal actually the integration project my experience tells me it is?
Why am I still doing that repetitive task manually every week?
Different problems. Same old reflex.
I'm deciding what they require before finding out what they require now.
"I don't know how" has changed meaning
For most of my life, I don't know how was useful information. It meant I probably needed to learn something, find someone who knew how, spend money, invest significant time, or reconsider the idea altogether.
Today, I'm trying to treat that same thought differently. Not as a conclusion. As a trigger.
I don't know how. Okay. Ask.
Asking doesn't mean committing. It costs almost nothing to explain what I want, explore a possible path and understand where the actual complexity lies.
That first exploration used to have a cost itself. Research it. Find someone. Explain the problem. Get an estimate. Now I can often get enough clarity to make a better decision in minutes.
And this doesn't only apply to new things. I still catch myself doing repetitive things manually simply because that's how I've always done them. There's no "I can't do this" reflex to interrupt. The task just feels normal. So I do it.
Old thinking doesn't only stop us from doing new things. It keeps us doing old things the old way.
The question becomes the same: Does this still need to be done this way?
Ask doesn't mean do
There's an obvious trap here. Once you realize how much you can suddenly attempt, it's easy to swing too far in the other direction.
I'll quickly do this with AI. Three hours later, I'm deep in a rabbit hole wondering why the supposedly simple thing still doesn't work.
AI has created a strange estimation problem. The old instinct says: This will take three weeks. The new AI-euphoric instinct says: Claude will do it in 20 minutes. Both can be wrong.
Relearning how to think doesn't mean assuming everything is easy now. It means becoming less confident in your assumptions about what something should cost. In either direction.
If testing the next step is cheap, test instead of speculate. Then ask the question that increasingly matters more to me than "Can I do it?"
Is the potential upside worth my attention?
Because increasingly, the answer to can I? is some version of: Maybe. Let's find out.
A larger space of possible
For a long time, the boundary of what I could realistically attempt was closely connected to what I knew how to do. Then it expanded to the people around me: Who knows how to do this? Then money: Who can I pay to do this?
AI introduces another question: What can I figure out? And that is a much larger space.
That doesn't mean AI makes everything possible. It doesn't mean expertise no longer matters. And it definitely doesn't mean every idea deserves to be pursued. It means something much simpler:
More ideas deserve to survive the first five minutes.
A lot of ideas don't die when they meet reality. They die before reality. We reject them because they sound expensive, difficult, unfamiliar or outside our capabilities, often based on assumptions that were completely reasonable a few years ago. But reasonable then doesn't mean accurate now.
So when my brain says I can't do this. I don't know how. That's too complicated. That'll take forever. I don't want to automatically believe it anymore.
I want to notice it. Ask. Find out what the problem actually looks like today. Try the cheapest next step. Then decide whether the potential upside is worth my attention.
Sometimes the answer will still be no. But at least it'll be today's no, not yesterday's.
(C) 2026 — ante.design