On Being an Engineer in 2026

Are We Still Needed?

We all feel it. Every other week, we lose or delegate another part of our work to agents.

It started with giving up on writing code. Then it expanded to pretty much any writing - Slack messages, documents, emails. Since around Opus 4.6, debugging, once an art and something that made us proud, became clicking on the recommended option in Cursor. And with better harnesses, even architecture and high-level thinking started moving to agents.

My day-to-day now mostly consists of making sure my agents aren't blocking other people's agents (which is another topic worth writing about).

This naturally leads us to think that this is it. We, engineers, are doomed. But before you start thinking, again, about which jobs are "AI-protected", let's take a step back. We still have some real leverage over agents, and let me try to prove it using my favorite mathematical trick - reductio ad absurdum, aka "proof by contradiction".

• Assume we have no real leverage over agents. The agents are doing all the thinking, and we're just getting in their way.
• We all have access to roughly the same models and harnesses.
• So, if the human doesn't matter, the difference between engineers' output should shrink dramatically.
• But that's not what we're seeing. Great engineers are still producing much better results than average engineers. If anything, 10x engineers have become 100x engineers. And the average engineer is still average.
• Therefore, there's still something the human brings to the table.

But there's a problem with proof by contradiction - it shows if an assumption is wrong, but it doesn't tell you why.

So What Is It?

Now that we've established that humans still have some leverage, let's try to understand what it actually is. Let's reverse-engineer it from a few observations:

• (Sometimes) agents get lost when the problem isn't defined well enough. I tend to keep asking questions until I feel we're solving the actual problem.
• (Sometimes) agents overcomplicate simple things or oversimplify complex ones. I have some intuition about which problems are hard and which ones can be kept simple.
• (Sometimes) agents can solve a problem without having to live with the solution. I have to think about whether it'll still make sense two years from now.
• Some things I can't put my finger on. I know something is wrong, but can't really explain why (what we call "intuition").

A cool example of that last point is our allergy to AI slop. We all know that feeling when we read less than a single sentence and immediately know that an AI wrote it, so we just stop reading. We don't necessarily know why, but something feels off. We've even developed a whole language around it - slop, slop canon, slop grande, and an entire meme culture. We've developed a collective taste for what feels like slop without ever explicitly defining what that means.

Anyway, going back to the observations list - you could argue, of course, that all of this is temporary. With Opus n+1, we're done. That might be true. But I've been hearing this for three years now, and somehow we are still here. So maybe in order to predict when we'll become obsolete we need to understand why we haven't become obsolete yet. And for that, let's use one of the oldest questions in philosophy.

On Human Consciousness

For centuries, philosophers have tried to answer a seemingly simple question - what is consciousness? What makes us, us? We've studied the brain, its cells, neurons, even our DNA. Yet we still can't fully explain how consciousness emerges from all these individual parts.

While reading about this topic, I came across two concepts that clicked for me. Together, they seemed to explain what I was looking for:

• Emergence - the idea that something can arise from the interactions of many smaller parts, even though none of those parts has that property on its own.
• Tacit Knowledge - an idea by the philosopher Michael Polanyi that "we can know more than we can tell" - things we know but can't easily explain.

I'm not suggesting that consciousness explains why we're better at some things than agents (or even that agents won't have consciousness), but it made me think about how much of who we are comes from things we can't easily describe.

Much of that comes from experience. A random conversation, something that happened years ago, a thought during a morning run. None of them matters much on its own, but together they shape who we are.

We Will (Probably) Still Be Needed

Models already outsmart us in many ways. They have agency and they're developing taste - at least the kind of taste you can learn from reading everything we've ever written. Both will keep improving. Maybe Opus n+1 will close the gap even further.

But I think there's a difference between knowing everything about something and actually experiencing it. We're still the ones having random conversations, making mistakes, developing opinions, having fears and desires. All those things shape who we are, and in turn, how we approach problems and what we build.

I keep asking questions because I've built a solution to the wrong problem before. I have a feeling about what problems are genuinely hard because I've been outsmarted by the ones that looked easy. I care about the next two years because I solved problems two years ago and I learned what works and what doesn't. And that thing I can't put my finger on - the thing that makes us allergic to slop - is an experience I never consciously stored.

Even caring about not blocking other people's agents is, ten layers down, about being afraid of getting fired and wanting to be excellent at what I do - 2026's mix of impostor syndrome and pride in the craft.

Maybe one day agents will have their own experiences, too. Maybe they'll develop intuitions they can't quite explain, just like we do. But until then, our leverage isn't just in what we know, but in everything we've experienced that made us who we are.