Nothing Will Replace Thinking for Yourself
Every few weeks I hear a variation of the same question.
Should my kids still learn math?
Should they study computer science?
Does it still make sense to learn how to code if AI can already write software?
These are reasonable questions. I've been asking myself the same ones—not only as an engineer building AI products, but also as the father of young children. The pace of progress is unlike anything we've seen before, and it's natural to wonder which skills will still matter ten or twenty years from now.
After thinking about it for a long time, I've arrived at a simple conclusion:
> Nothing will replace thinking for yourself.
And learning—not simply knowing facts, but going through the process of learning—is how we develop that ability.
Learning is not about information. One of the biggest misconceptions about education is that its purpose is to fill your brain with information. It isn't. The real purpose of learning is to train your mind. When you study mathematics, physics, computer science, history, philosophy, or music, you're certainly acquiring knowledge. But something much more important is happening underneath: you're learning how to reason. You're learning how to break down complex problems. You're learning how to build mental models. You're learning how to recognize patterns. You're learning how to learn. Those abilities stay with you long after you've forgotten the details.
I rarely use what I studied. I use what it taught me
I studied computer science. I learned how processors work, how memory is managed, how operating systems schedule processes, how networks move packets, how compilers are built, how databases are implemented, how assembly works, and countless algorithms and mathematical concepts.
Do I use assembly language every day? Of course not. Do I design CPUs? Not even close.
Most of the software I build today sits on top of layers upon layers of abstractions that other brilliant engineers have already created. But that's always been true. Every generation builds on higher levels of abstraction. Open-source software meant we stopped rewriting common libraries. Cloud computing meant we stopped managing physical servers. Frameworks meant we stopped writing boilerplate.
AI simply represents the next abstraction layer. The tools changed. The value of understanding didn't.
Because even if I don't implement a database anymore, understanding how databases work helps me know when to use one, which one to choose, why a query is slow, or why a design won't scale.
The knowledge isn't valuable because I reproduce it every day. It's valuable because it shaped how I think.
AI shouldn't replace judgment
I use AI constantly, it writes code with me every day. Sometimes I'll ask it to build an entire prototype. Sometimes I'll let it generate hundreds of lines of code. But that doesn't mean I blindly accept everything it produces. If I'm building something that matters—a production system, a product, or critical infrastructure—I read the code, I question the architecture, I challenge the design. I often ask AI to rewrite large parts because I disagree with its approach. Not because the AI is bad, implementation isn't the hardest part anymore. Judgment is.
Knowing *what* to build, *how* to structure it, *which trade-offs matter*, and *when a solution is elegant versus merely functional* are still deeply human skills.
AI can generate answers. It shouldn't own your judgment.
## Great leaders understand enough to ask better questions
Over the years I've worked with many engineering leaders. The best ones weren't necessarily the best programmers on the team.
But they understood technology deeply enough to exercise good judgment. They knew enough to recognize when a proposal wouldn't scale. They knew enough to challenge assumptions. They knew enough to ask better questions.
Unfortunately, I've also seen the opposite. Leaders without sufficient technical depth often direct talented teams toward the wrong problems or insist on solutions that shouldn't exist in the first place. The result isn't just bad software. It's frustrated engineers, unnecessary friction, lost time, and many times even losing great people.
Now imagine replacing those engineers with AI agents. Unlike great engineers, AI won't naturally push back. It won't tell you that your framing is wrong. It will faithfully execute the direction you provide. Which means the quality of the outcome depends even more on the quality of your thinking.
If you don't understand the problem, you probably won't ask the right questions.
If you don't understand the trade-offs, you probably won't choose the right architecture.
If you don't have taste, AI can't invent it for you.