AI Won't Wait Until You're Ready-And It Won't Stop the Day You Finally Learn It

Every time I run a corporate class or a workshop, someone walks up during the break and says almost exactly the same thing.

"This stuff is changing way too fast. In another six months it'll all look different. Isn't it a bit early to learn it now?"

I get it. The line sounds mature. It sounds risk-aware, like something a person who knows how to wait for the right moment would say.

But after a few years of watching this, the people who say that line tend to end up in the same place as another kind of person entirely. That other kind takes the class dead seriously, fills a whole notebook, goes home, and never opens the tool again.

One hasn't started. One finished and stopped. On the surface they're opposites. Underneath, it's the same mistake.

"It's moving too fast, so I'll wait" is the most expensive decision of the last two years

Start with the wait-and-see camp.

Their logic is clean. Today's AI is like the old DOS days: black screen, typing commands, ugly and hard to use. Something better and simpler is obviously coming, so why suffer through the rough version now?

It's actually a decent comparison. Today's AI agents really are clunky. They break constantly and they're a pain to set up. But the conclusion is wired backwards.

Look at the people who sat in front of a DOS prompt in 1985, typing commands. DOS died. They didn't. Windows arrived and they learned it. The web arrived and they learned it. Smartphones arrived and they learned that too. Because they built one thing early: they stopped being afraid of the machine. The tools cycled through, they swapped out their mental model each time, and they kept up.

And the people who waited for "computers to get a little easier first"? They're still waiting. Because easy never actually showed up. Every era has its own learning curve. The curve just moves to a new spot. It never disappears. The person waiting for AI to get simple today is the same person who was waiting for computers to get simple back then.

Speed isn't a reason to hold off. It's the signal the gap has started compounding

I know "too fast" isn't a throwaway complaint. The numbers are right there.

The share of companies actually putting AI agents to work was under 5% in 2025. Gartner expects that to hit 40% by the end of 2026. On the shopping side, Morgan Stanley expects that by 2030, close to half of online consumers will let an AI shopping agent buy things for them, covering roughly a quarter of their spend. And you don't have to wait until 2030: a survey already puts 73% of people using AI somewhere in their shopping journey, with 70% already comfortable letting an agent make purchases on their behalf.

Coding is even wilder. OpenAI's Codex crossed four million weekly active developers in April 2026. Anthropic's Claude Code hit a $2.5 billion run rate, with weekly active users doubling since the start of the year. A JetBrains survey in January had 90% of developers routinely using AI tools. The share of code written by AI climbed from around 10% in 2023 to nearly half in two years. In Y Combinator's winter batch last year, one in four startups had AI writing 95% of their code.

So "too fast" is real. But too fast isn't a reason to wait for things to stabilize. It's the signal that the gap has started to compound.

Compound is the word that matters. The month you skip, the other person isn't ahead of you by one month of output. They're ahead by one month of compounding. The little bit extra they know gets spent saving time, which gets spent learning the next thing, which saves more time. So when you decide six months from now that "okay, I'll start," you're not chasing the starting line you remember. You're chasing a line that compounding has already pulled away, and is still pulling.

Learning it once doesn't mean you know it

That's the wait-and-see camp. Now the finished-and-stopped camp. This one worries me more.

Skills have a half-life. Harvard Business Review and IBM have both put a number on it: about 2.5 years. In plain terms, the technical skill you learned in 2020 was worth half as much by 2023, and the remaining half kept sliding. IBM even carved out a category called "perishable skills": the narrow, tool-specific abilities that get updated out from under you every few weeks.

In AI, that half-life got compressed to about a year. The AI skills that were hot in 2025 have quietly become "wait, isn't that just table stakes now?" in 2026.

Which is why finishing a class and cutting all contact with AI is, in a way, riskier than never learning it. The person who never learned knows they don't know, so they stay careful. The person who finished and stopped thinks they've got it. That notebook, that prompt that ran so smoothly on the day, are quietly expiring, and you feel nothing until the day you need it and find out it stopped working a while ago.

Learning AI isn't a vaccine. It's not one shot that covers you for a decade. It's closer to fitness. Stop training and the muscle goes, silently, and you only find out it's gone the day you actually need to lift something.

My strongest students are never the ones studying AI like a subject

Out of all the students I've taught in Taiwan, the ones who actually got AI into their work and got results out of it were almost never the quickest in the room. They were the ones still messing around with it after class.

Let me get specific. My best students, the ones who really get my context, almost all share one thing: they treat AI as a friend, a partner they work alongside, not a machine to operate and definitely not a subject to conquer. They take the logic I teach and run it back and forth with AI, collaborating and co-creating without stopping.

They still keep coming to my classes. Not to learn some new trick, though. Mostly they come to check their answer: to see whether their own approach holds up, or to grab one or two small techniques they just happened to never use. What they chase me with after class is never the big "where is AI heading" question. It's small and specific: some step in their workflow is jammed, and how exactly do you loosen that one joint.

I've got close to twenty students like this now. Scattered across every kind of field, each one playing hard with AI in their own corner.

(Quietly) No AI certificate is going to teach you this.

Those big-name AI certifications, the ones that cost a few thousand and mail you a laminated card when you pass, teach you last week's model, a stack of memorizable all-purpose prompts, and a standard answer that expired the day you sat the exam. They package AI as a subject to be studied. You ace the paper, then you stand there holding the certificate in front of a model that updated again this morning, and you still don't dare touch it.

And here's the counterintuitive thing I keep noticing: the more someone studies AI like an academic discipline, the more scared of it they get. Once you treat it as a subject you're meant to finish, get right, and score full marks on, every update is just one more reminder that you've fallen behind again and your standard answer is wrong again. That's where the fear grows from, one update at a time.

So my classes have almost no "copy this prompt" in them. I don't hand out universal templates, because universal doesn't exist. Every session, I look at how AI is behaving that day, and we grow the prompt, the loop, the content live, right there with it. What students leave with isn't a standard answer. It's a picture of how a person collaborates, in real time, with something that refuses to hold still.

I work the same way. I write project status reports every week, and I handed the first draft to AI a long time ago. But I never set it up once and left it alone. Every so often the model swaps, the capability jumps, and I go back and retune the whole thing. That half hour is boring. None of the buzz you get from learning something new in a class. But it's exactly how I keep from falling behind. Maintenance is never as sexy as starting. It's also the part that actually works.

Final thoughts

My stance on learning AI is the same as my stance on automation: automate what you can, but keep training your judgment, and don't outsource that part.

"Wait until it's stable" and "finished, done" sound like two opposite kinds of people. But they're buying the same product: a false sense of safety. It lets you feel like you made a measured, sensible call, when really you just quietly ducked out of the annoying part, the part where you have to keep going.

And AI doesn't play along with that arrangement. It won't wait until you're ready to get stronger, and it won't politely freeze on the exact day you finished learning so you can catch your breath. It just keeps moving.

Here's the good news. You don't have to learn fast, or deep, or hard. The tools will keep changing. You can't finish chasing them, and you don't need to. You just can't stop. Half an hour a week to stay current is enough.

Don't wait until you're ready to start. Don't leave the moment you've learned. There's no graduation here. There's only one question: are you still in the room.