On the doubt that grew the more I used AI

AI has made my studying much faster. Material that used to take me days I can now skim in one, and when I get stuck I ask on the spot. I want to spend my life learning, so I genuinely think this is a good change.

Oddly, though, the more I used it, the more an opposite feeling grew alongside. It was close to a sense of loss. What am I better at than AI, and then why am I even studying? The thought kept coming back.

Even to me the feeling didn’t add up, and it bothered me for a while. Then I watched an interview with Chris Piech that EO Korea posted on August 24, 2026. He teaches intro programming and math for AI at Stanford, and the interview answered more of my doubt than I expected. The interview falls into three parts, so this post follows the same order.

Lesson 1. The difference between people and AI was motivation

Piech has run an online intro programming course, Code in Place, for six years. It has around 17,000 students and over 1,000 teachers, roughly one teacher for every ten students. He describes it as the course with the most teachers in the world.

Over those six years he experimented with mixing AI into the course in different ways, and the results were a bit surprising.

Just handing students a chatbot and telling them to study with it made people quit, predictably. Dropping AI on someone at the wrong moment in their learning actually saps their motivation. But when a pop-up appeared mid-coding, saying a teacher was online and would like to spend ten minutes with them, the students who clicked yes were 10 percentage points more likely to finish.

You’d guess the people gave better answers, but the chat logs said the opposite. The AI was the accurate one, with no hallucinations, and it was the humans who were sometimes wrong. Still, it was the humans who got students to the finish line.

“The crown jewel of education has always been motivation.”

His explanation was simple: the human touch is special, and that’s what motivates. So, he said, something like this lands far better with a student:

“In this age of AI, I’m not giving up on you becoming a smart person. Let’s do the fundamentals together.”

After that comes a promise: once the fundamentals are done, he’ll teach them how to code with AI.

This is where I started to see where my doubt came from. I had always taken it as a skills problem, a feeling that came from AI being better than me. But there’s a space that no number of accurate answers can fill, and when it’s empty, the reason to keep going fades no matter how skilled you are. That was probably what I’d been feeling.

So where does curiosity come from?

Midway through, Piech sets a challenge for people building chatbots. Chatbots are good at answering questions, so how do you make them inspire, too?

Here’s how he inspires students. When a student comes by his office, he says, “Want to see something really cool about probability?” and shows them something. It has nothing to do with the question they came to ask, but the student feels a kind of care in it.

“If I can flip the switch that makes a student so curious they can’t help but learn, then all day long the thing on their mind will be the problem I gave them.”

If that curiosity gets ignited, then I feel like they’ll get there.

Today’s chatbots aren’t good at lighting that spark. As he put it, the chat never shows up first. It doesn’t come to you and offer to show you something mind-blowing, and I liked the way he said it. He can do it because he knows the student’s context: where they are right now and where they’re trying to go, which tells him what to show them.

Why he cares so much about this comes through in his own story. Born in Nairobi, he moved to Kuala Lumpur at twelve and came to the US for college, and he never set out to become a professor. “I was just a curious person.” In a programming class he took only to fill a breadth requirement, the teacher assigned the students to make the coolest thing they could with what they’d learned in two weeks, and he found himself excitedly pouring in 40 hours beyond the regular coursework.

That left me with a question. If the chatbot won’t come to me and light the spark, maybe I’m the one who has to light it. I haven’t yet figured out how to keep learning with AI without losing curiosity and passion. It’s a question I keep turning over these days.

Lesson 2. Then why still learn to code?

I suspect many developers share this worry. AI already codes better than I do, so will I still be working as a developer in five or ten years?

Piech answers with something from his own past. Around 2011 and 2012, during his PhD, someone who is now his colleague was hitting early milestones in self-driving cars. Watching cars drive themselves, everyone had the same thought: what about taxi drivers? What happens to truck drivers?

Piech doesn’t deny that a fully self-driving world will come someday. What struck him was how badly we overestimate how fast we’ll get there. The reason he gives is the long tail.

“We’ll get there eventually. We just massively overestimate how fast.”

Honestly, I paused at this part. Won’t technology keep getting faster from here? The pace now feels different from the self-driving era, so on this point I can’t fully agree with him yet.

But the question that came next was impossible to argue with.

When am I actually getting something out of AI, and when have I given away too much of the growth?

Piech programs with AI a great deal himself, but he happens to know architecture well. If he didn’t, the AI’s bad decisions wouldn’t show in the first prototype. They’d only surface five weeks later, when students actually using it start hitting strange bugs, and without understanding the structure, he couldn’t help them then.

That stung a little. I had been feeling that, as I handed more to AI, I was drifting away from solving problems myself. His condition is aimed right there: as you use it, look honestly at whether you’re growing alongside the AI. Whether I’m growing as much as things have gotten easier is something I’ll have to keep checking.

“AI does it all anyway, so why study?”

If someone asked me that now, I think I’d say that knowing how to solve problems takes more fundamentals, not fewer. Piech splits learning to program into two pieces, too.

“AI is going to get really, really good at syntax.”

Syntax is the part you can hand off. Hand off problem solving too, and when those strange bugs show up five weeks later, you can’t do anything about them.

One more reason he gave stayed with me: coding gives immediate, falsifiable feedback. If your logic is wrong, the program doesn’t run, so you can see the mistake and quickly try a fix. Other problems in life return feedback so slowly that practicing to get better is hard in itself. That’s what makes coding unusually good for practicing problem solving.

Another surprise: enrollment in his course doubled after Cursor and Claude Code came out. I’d assumed fewer people would want to learn once AI could code, but more did.

But what matters more comes next

This is the side I think about most these days. Going forward, more than raw development skill, I’ll need to be able to answer what users want from the service I’m responsible for: what problems customers are running into, and which key features would help them take the next step.

Piech said nearly the same thing. The barrier to entry has dropped so far that two high school students can now build a pretty impressive codebase. Once that’s true, the question left is this:

There is a real art form to knowing what is a valuable problem to solve.

And more and more juniors can take part in that work, he said: judging what’s worth building, what users want, and which features actually move their problems forward.

“Bridging what computers can do and what people actually need has always been the decisive higher-order skill. If I were a junior engineer, I’d start on it now.”

I wouldn’t wait till I was a senior engineer.

His closing words pointed the same way. The best engineers may not be the ones who know all the code, but the ones who can carry real human problems into the world of apps, of data science, of research.

So lately I’ve been wondering how to build that skill through conversations with AI: using it not to write code for me, but as a partner for picking apart users’ problems. I’m still learning, so I don’t have a tidy answer yet.

There was one method worth borrowing. It’s what he said he’d do in my place: build lots of prototypes with Claude Code, and keep asking it to “teach me all the most important things you did while building this.” Do that enough and you work out for yourself which concepts matter. To me it sounded less like “don’t hand off the work” and more like “take back what you handed off.”

Lesson 3. The most discouraged are the ones who think too hard

This part felt a bit like being hit on the head.

The people I’ve seen get most lost and most demotivated are sometimes the ones who are overthinking it.

Piech had one student who was thriving: using AI, solving problems, learning amazing things. When he asked the student what they were thinking about, the answer was roughly this:

“Honestly, I don’t think about it. I don’t think much about the future of AI, and that’s why I’m doing fine.”

He stopped short at that. He’s someone who thinks about AI ten times a day, and the student had a simplicity he lacked: “No, I’ll just learn with curiosity.”

For me, that was the most important line in the interview, and the one that rang truest. I’ll just learn with curiosity.

Looking back, I’d been thinking far too much about what developers will be in five years, how far AI will go, and what I’m actually good at. I took it for diligence, but it probably wasn’t. If predictions rarely come true anyway, thinking about them longer won’t produce an answer.

Since then, he says, he starts each day with an axiom. He doesn’t ask why he wants the next generation to be smarter; he just takes it as true. “I want this, and I’ll work toward it.” That’s also why he stresses fundamentals. If you assume the next generation will be smarter than us, there’s no way to skip the basics, just as children still learn multiplication long after calculators arrived. What matters is the concept of multiplication, not doing 13 times 7 quickly, so we can be more deliberate about what to focus on.

What I’m holding on to now

Piech has a name for AI: it’s a tool, and it will multiply people.

What I held on to was the word multiply. If it added, the tool would contribute the same amount no matter who I am. If it multiplies, the story changes: what’s being multiplied, me, decides the result. Put that way, “what am I better at than AI?” was pointed the wrong way from the start. The stronger the tool gets, the more my side of it matters.

His parting advice was, in the end, to build. “Go build something. Build things people use, build things people love.” Keep building, he said, and you’ll get the chance to get good at both coding and problem solving. Then he closed the interview with this:

“Chris hasn’t given up on you. Don’t give up on yourself.”

It’s the line from Lesson 1, the one he says to his students, handed back at the end to whoever is watching.

Of course, I haven’t settled everything. I still see the pace of technology a little differently from him, and I’m still figuring out how to keep my curiosity fed.

But I have decided one thing: to keep becoming a better, more capable developer. I’ve stopped asking why I want that. It seems better to leave it be, and study and work at it.

References

• Interview with Chris Piech, EO Korea, 2026-08-24. Lines in italics are his own words from the video; the other quotes are my renderings of what he says.
• Code in Place: Stanford’s online intro programming course that Piech runs.