A call of intelligence

It's crazy to think that you can call an API and get intelligence in return.

Not too long ago, intelligence was something you had to find in people. It came with years of experience, intuition, memory, and everything a person had lived through. Now you can summon “something resembling it” with a few lines of code.

I don't think we have fully absorbed how absurd this is.

For centuries, we've built machines to extend our physical capabilities. Then came computers, extending our ability to calculate, store, and process information. And now, for the first time, we have machines that can recognize patterns, reason through unfamiliar problems, and make judgments in ways that were previously reserved for humans.

Yes, they make silly mistakes. Sometimes embarrassingly silly ones. But I find it fascinating that we can steer something this capable with language. A few words can change how a system behaves. A paragraph can give it a role, a perspective, almost a personality.

Language is becoming an interface to intelligence.

And I wonder how much of the world is going to change because of that.

I think we're approaching an inversion in how software works.

For decades, software was built for humans. We learned to navigate interfaces, fill forms, click buttons, and understand how different systems worked. Entire companies were built around making these interactions easier.

But what happens when the user of software is no longer human?

I can imagine a world where SaaS products become the underlying infrastructure, serving agents, while agents serve humans. The software doesn't disappear. It just moves further away from us.

The interface becomes intelligence.

And perhaps, eventually, we won't care which software is doing the work underneath.

I find myself in conversations these days where someone proposes building something, and inevitably the discussion reaches a point where we say, "Let's put an LLM in between."

And sometimes, that's genuinely all it takes.

Something that would have required a complicated set of rules, edge cases, and deterministic logic can suddenly be approached with an intelligence layer.

It feels almost unfair.

Years of engineering abstractions, and now a model can sometimes reason its way through the same problem.

Of course, that introduces a different class of problems. Unpredictability, reliability, hallucinations. And hence we are seeing the rise in verification softwares

But what interests me more is the shift in how we think about building things.

AI today feels like an incredibly powerful autocomplete. It completes our code, our sentences, sometimes even our thoughts.

And everyone seems to be talking about how anyone can now build software.

I agree, to an extent. But there's something about that statement that bothers me.

How do you build something when you don't know what's possible?

I think there are two kinds of possibilities.

The ones you already know exist, and the ones you discover by going deep into something.

The first kind is easy to describe. You have an idea, you know roughly how it should work, and now an agent can help you build it much faster.

The second is different.

When you've spent enough time working with a system, you begin to see things that aren't obvious from the outside. You understand its limitations, its peculiarities, the things it allows you to do that nobody explicitly designed it for.

Sometimes you start building one thing and end up discovering something entirely different.

And I think that kind of understanding comes from being involved in the process, not just observing the outcome.

Maybe AI will eventually be better at this too. I wouldn't bet against it.

But I wonder if we're creating a generation of builders who can produce almost anything they can describe, without necessarily knowing what else could have been possible.

There's a difference between knowing how to build something and knowing what can be built.

And I don't think we've figured out how to bridge that gap yet.

There's another thing I've been feeling lately.

AI fatigue.

My workflow has slowly changed from writing code to supervising its creation.

An agent asks me a question. I make a decision. It does something. Then another question. Another decision. Another review.

At times, it feels like being interrupted by someone every few minutes, except that someone is remarkably capable and never gets tired.

I don't think most of us were trained to think this way.

To receive a constant stream of bullet-point questions, make judgments, and move on to the next one.

Thinking isn't always a sequence of discrete decisions. Sometimes you need to sit with a problem. Wander around it. Write something, delete it, try something stupid, and eventually stumble into an answer.

There's a certain continuity to that process.

And I feel like I'm losing some of it.

I wonder what flow state means now for people who used to enjoy writing code at 3 in the morning, headphones on, music playing, completely lost in something.

I was one of them.

There was something deeply satisfying about writing code by hand. Not just the outcome, but the act itself. Understanding why something worked. Finding a simpler abstraction. Spending hours on a problem and finally seeing it come together.

These days, I sometimes spend more time reviewing code than writing it.

And the strange part is, I can't even argue against it.

It's faster. Often good enough. Sometimes better than what I would have written myself.

From a company's perspective, spending hours writing something an agent can generate in minutes doesn't make much sense.

I understand that.

But I also think we're making a mistake when we assume that because implementation has become cheap, thinking should become fast too.

Implementation is getting cheaper. Understanding isn't.

In fact, I think understanding is becoming more valuable precisely because implementation is becoming so accessible.

When everyone can build, knowing what to build matters more.

Yet I see an increasing pressure to produce more, ship faster, and keep agents running. Almost as if every minute not spent generating something is a minute wasted.

But some of the best ideas I've had came when I wasn't producing anything.

Just thinking.

And I wonder whether we're giving ourselves enough time for that anymore.

There's a strange sadness in watching something you deeply love become less practical.

I don't think writing code by hand will disappear. People still paint after the invention of photography. People still play instruments when music can be generated.

But it's different when your craft is also your profession.

You don't always get to choose how the work gets done.

Maybe the joy will move somewhere else. Maybe the next generation of engineers will find the same satisfaction in designing agent systems, orchestrating intelligence, or building things we can't yet imagine.

I hope they do.

I just hope we don't forget the value of going deep into something, even when there's a faster way around it.

And then there's the bigger picture.

I think personal AI agents are going to become a significant part of our lives.

Not just chatbots that answer questions, but systems that know our context, remember our intentions, and act on our behalf.

Imagine an intelligence that knows what you're working towards, what you've been putting off, what matters to you, and what you've changed your mind about.

Something that doesn't need to be told everything from scratch.

We've spent years putting pieces of ourselves into different software. Our conversations, documents, calendars, photographs, decisions, and memories.

Perhaps personal agents will be the layer that connects all of it.

And that raises some interesting questions.

If something knows your preferences, remembers your past, and can reason about your future, how much of your thinking will you eventually outsource to it?

And when something becomes good enough at anticipating your decisions, how do you know whether a choice is yours or one you've gradually learned to accept?

I don't have answers to these questions.

But I think the implications go much deeper than productivity.

We're not just changing how work gets done. We might be changing how humans relate to knowledge, effort, and eventually, to themselves.

For most of history, what you knew was a significant part of what made you valuable. Knowledge took time to acquire. Experience took years to accumulate.

Now the distance between not knowing and getting an answer is shrinking.

But an answer and an understanding are two different things.

I keep coming back to that distinction.

I turned 27 today.

The last year has been significantly different from all the years before it.

I've spent a good part of my life learning how computers work, how to build software, how to make machines do exactly what I want.

And now I'm watching machines become capable of figuring out parts of that themselves.

It's exciting. And a little scary.

Sometimes I feel incredibly lucky to be alive at this point in history. Other times, I wonder if we're moving too fast to understand what we're leaving behind.

Maybe every generation feels this way about the things that shaped them.

I don't think software engineering is dying. I think it's changing into something we don't fully understand yet.

And perhaps that's what makes this moment so interesting.

We're building systems whose possibilities we haven't completely discovered, using intelligence we don't completely understand, to shape a future we can barely predict.

I find that both beautiful and unsettling.

I don't know what programming will look like five years from now. I don't know whether the things I spend my days learning today will still matter in the same way.

But I hope curiosity remains valuable.

I hope we still find reasons to understand things instead of merely getting them done.

And somewhere, I hope there's still someone sitting at 3 in the morning, headphones on, music playing, building something for no reason other than the joy of figuring it out.

Maybe that someone will still be me.