Someone asked me recently if I could still write code without AI tools. I said yes, of course. I'd just go back to Google and Stack Overflow, the way I did a few years ago.
Halfway through the answer, a thought hit me. Why would I go back?
It took me back about six years, to when I studied frontend development at a yrkeshögskola, a Swedish vocational college. One day our teacher asked the class if we remembered everything from the lecture the day before. The room went quiet.
Then he said something like, you can forget about all of that. Most of the time you'll be able to Google it or find it some other way. You don't have to remember everything by heart.
He wasn't only a teacher. He worked as a developer too, so he knew what was actually happening out there. And he was right. Since then I've more or less never been in a situation where I couldn't search for how something works or how the syntax looks.
That's what stuck with me about the question. I know I could do it. But it assumes the old way is the real test, and that the tools I use every day are something I have to prove I can live without.
My teacher was preparing us for the world we're living in now. The question was testing for the one before it.
Tests for a world that's already gone
Some coding tests back then still said no internet. Write it from memory, or from a book. Even then it felt odd. Nobody at work coded without Google. The test measured a way of working that was already gone.
Today I hear about schools banning AI and computers in tests, because they're afraid students will cheat. I understand the worry. But the life those students are heading into will have these tools built into almost everything, whether we like it or not. At some point I think the tests have to reflect that too.
Build for where things are going
Boris Cherny, who leads Claude Code at Anthropic, said something like this about how they build it. They don't build for the model they have today. They build for what they think the next one will be able to do. Otherwise they'll always be a step behind.
I think that way of thinking fits a lot more places than AI products. If we write tests, job interviews and training based on how things worked years ago, we'll always be at least one step behind.
Sweden already did a small version of this. You can take your driving test for an automatic-only licence now. More and more of us drive automatics, so the test followed.
Plenty of people made a big deal of it, and some still defend the old way. But those voices get a little quieter every year.
And cars keep changing. Cameras, sensors, more and more of the driving done by the car itself.
So should we keep testing drivers the old way, or adjust? I sometimes throw this out just to see how people react. I don't have the answer. I just think it's a fair question.
So should we stop learning to read?
This one is a thought experiment, and I use it when I want a real discussion going. Why learn to read and write if an agent can do it for you?
You hold down a button on your phone and ask. It finds the information, reads it, does the task. For a lot of things, that's already possible today.
Now think about what Boris said, and what the next model will be able to do.
I'm not saying that's where we should end up. I hope we never do.
Reading and writing help us in a hundred ways, from reading a sign to lying on a beach with a book just because you enjoy it. I'd like to think there's still room for an old-fashioned love letter too. And if the power goes out or the network is down, you'll be glad you can do it yourself.
But taking it all the way is useful, because that's where the line shows up. Some skills are the foundation, and everything else gets built on top of them. Others were the tools of their time.
Reading and writing are foundation. So is understanding what you're doing, and being able to tell when something is wrong. Remembering syntax by heart, or finding the answer in a book instead of a search, made sense when they were the only option.
If that's still your way and you enjoy it, go for it. It's still an option. It's just not the only one anymore.
I'm still a software engineer. I deliver what I'm supposed to, and I get paid for it. Agents write most of my code now, and I still decide what gets built, review it, and answer for everything that ships. The title didn't change. The tools did.
So if you want to test me, test the part that matters more now, not less. Can I tell when the output is wrong? Do I understand what I'm asking for? Do I know where a human needs to check before something happens?
We should learn from the past, for sure. Some things are fundamental and we take care of them. The rest we can build around what we actually have, and what's coming.
Which of the things you test for today are the foundation, and which were just the tools of their time?