A coach's note, on the morning of the AP CS A exam.
Today is the AP Computer Science A exam. A parent emailed me last night. Her son sits for it in a few hours. She asked: setting aside the 5, setting aside the college credit — will any of this still matter in five years, when AI writes the code?
Fair question. The short answer is yes. The longer one follows.
What hasn't changed
The Raspberry Pi Foundation said it well last summer: the difficulty of programming isn't a bug to engineer away — it's the part doing the work. Writing code by hand teaches things that don't form any other way. How to decompose a problem. How to hold a model in your head. How to predict what a machine will do before it does it. A kid who has spent a term doing this thinks differently afterward. That doesn't go away because a model can autocomplete a function.
What AI actually is, for a learner
The conversation gets pulled two ways: hype on one side, fear on the other. Neither helps you plan your kid's year. Two writers have steadied me, and I'll pass them on.
Dario Amodei — who runs Anthropic, and is about as close to the frontier of this technology as anyone — has described the current moment as an adolescence of technology. Powerful, growing fast, capable of surprising even the people who built it, but not yet grown. That framing feels closer to the truth than either "AI is solved" or "AI will replace everything." It is going to keep changing for a while, and we have a real say in how it changes.
A programmer who writes under the name Geek Orthodox makes the second point. The people who get the most out of these tools are the ones who understand the fundamentals well enough to direct them. "We're still the ones," he writes, "asking the questions, coming up with the concepts, and directing and tweaking the final shape." A kid who knows what a loop is can build something five times bigger with AI than alone. A kid who doesn't copies what the model gives her and hopes it runs. The difference is small at the start. Enormous by year three.
The pace of all this has been overwhelming even for me, and many of my former colleagues who have been in this industry for more than 10 or 15 or 20 years. If you have felt that as a parent, you are not alone — and I want to sit beside that feeling, not talk you out of it. What steadies me is that the overwhelm is on the side of possibility. The recent shift to agentic programming — AI that takes initiative across a whole task, not just a line — is what changed what I can offer a beginner. A single person, working at home, can now build software that sat inside a professional engineering team a year ago.
What we're changing
What students and parents wanted most was something real — a project the student could point to and say I made this. A 3D game. A flashcard app for the language she was studying. Iterating on a real project teaches more than programming: how to scope an idea, how to recover from a wrong turn, how to keep going when nothing on the screen looks right yet. But setting one up well took a full weekend per student, and I could not afford that. So I fell back on generic exercises. They worked. They were not what I wanted to be teaching.
That's the part AI changes first. A starter project tuned to a student's interests now takes an evening, not a weekend. The bigger change is one level up, and it's the part I am most excited about. One of my favorite books is Josh Waitzkin's The Art of Learning — it is about reflecting on how you learn, and learning to put what you notice into words. That habit is what separates a student who improves quickly from one who only puts in the hours. Until now, it took a coach sitting beside her to develop.
Picture a student finishing the first working version of her flashcard app. Instead of jumping straight to the next feature, she takes ten minutes with her study partner: what did she try, what worked, where did she nearly take a wrong turn? She is the one thinking; the AI catches her notes and files them into a record that gets sharper week by week. Cognitive, metacognitive, and resource-management strategies — names most teenagers never hear — start attaching themselves to things she has already lived through.
I am using these same tools on my own learning right now — rebuilding the program with them, in the open, as I go. The agentic systems that surprised me this spring are scaffolding the curriculum, the tutor, the project library; this is work I would not have shipped alone. Later this year I will also redesign the website itself with these tools, so that what you see when you visit reflects how the program actually works now, not how it worked three years ago. We are starting. More is coming this year, and I would like to build it where you can see it, with your child in mind.
Closing
In 2021, parents asked whether this would help their child get into a good college. Now they ask whether it is still worth her time. I feel that worry — I have it about my own work some weeks. Amodei's framing steadies me: an adolescence, not an endgame, and we still have a choice in the matter. Kids who learn to think clearly with a computer in the room will be the ones deciding what we do with it.
This is a long transition. I'd like to walk it with you.
Further reading
- Raspberry Pi Foundation, Why kids still need to learn to code in the age of AI (June 2025).
- Geek Orthodox, Choose Your Own Programming Adventure.
- Dario Amodei, The Adolescence of Technology.
- Josh Waitzkin, The Art of Learning (Free Press, 2007).