When Intelligence Stops Being Scarce
I used to open a chat window the way I'd message a smart coworker: ask a question, get an answer, move on. That habit doesn't fit anymore. What answers back isn't one clever person. It's something you can copy as many times as you want and leave running all night.
That single fact keeps changing how I think about work.
The picture I had was too small
For a long time I pictured AI as a genius living in my pocket: helpful, fast, and basically like talking to a person.
That picture undersells what's actually happening. Chatbots and flashy demos get the attention, but the real focus inside the top labs is different: building AI that can help design the next version of itself. Coding agents made this real for me. I used to write code by hand. Then I started copying suggestions from a model. Then I pointed an agent at a whole module and reviewed what came back. Now entire pieces of software get built through long agent runs, and my job is mostly deciding what "finished" should mean.
I hear the same shift from people who aren't engineers at all. Someone shows up with an idea and asks where to find a team to build it. My answer lately is almost always the same: describe the problem clearly enough that an agent can build you a first version. Typing the code stopped being the scarce skill. Knowing exactly what you want, and why it matters, is what's scarce now.
What actually worries me
"AI will take all the jobs" isn't what keeps me up at night. Jobs will change. Mine already has. What worries me more is quieter: a life comfortable enough that you no longer need to work to survive, and you still have to find your own reason to get up and try hard.
I go to the gym even though nothing forces me to lift anything heavy. I have an elevator. I have food delivery. I go because I chose it, on purpose, as a habit. I think more of life is going to work that way. If I were sixteen again, I wouldn't try to be the smartest person in one narrow field. I'd practice being fast: comfortable picking up a new subject, quick to drop an approach that stopped working, quick to notice when the tools around me changed the rules again.
Foundations still feel worth years of study, the kind that teaches you how to think, not just what to know. Tools matter too. I just stopped treating the tools as the goal.
Where I actually spend my time
Here's the part fewer people talk about. Models can already write and check each other's work around the clock. A hospital, a bank, or a team with a real deadline still has to trust that output, change how they work, train people, and stay inside the rules that apply to them. None of that moves at demo speed.
That gap, between "the model can do it" and "our organization can actually run it", is where I spend my days: retrieval, evaluation, cost, safety, and the unglamorous work that turns a good demo into something a team can rely on every Monday. The more impressive the models get, the more that unglamorous work is worth.
I feel two things about where this goes. I'm optimistic long term: cheaper access to good advice and real help for more people. I'm also uneasy about the road there. The scary version of this story doesn't need AI to "wake up" and turn against us. It just needs powerful tools to end up in the wrong hands, and we already know that happens: software is guiding weapons today. Safety filters help until someone finds a way around them, and someone usually does. At some point I expect governments to do something uncomfortable: slow this down on purpose and treat the biggest computing power the way they treat other things that are dangerous at scale.
On top of that, it got a lot cheaper to fake the truth. I remember when a video still felt close to proof of what happened. That feeling is gone now. So more and more, the question I ask myself isn't "what will AI become." It's "what am I becoming while I use it": which habits I'm keeping, which ones I'm quietly dropping, and whether I still know why I'm building any of this in the first place.
If you're stuck somewhere between "we play with chatbots" and "we need this running in production," that's the conversation I keep having. The tools will keep changing. Someone still has to aim them.
— Evgeni Altshul