Skip to content
Personal essay·Software engineering·5 min read·Sep 5, 2026

Who Holds the Leash

Code got cheap. What's rare now is someone who can handle everything around it. A case for a much bigger definition of “software engineer.”

Software engineers today remind me of border collies working for AI agents. Border collies are fast, smart, and obedient — but they don't choose where the flock goes. Someone else holds the leash.

On learning

People talk a lot about coding skills. But if you think about it, knowing how to code really mattered before LLMs. Now software engineers are people who write prompts and make sure nothing breaks — a kind of QA.

Is learning to code hard today? I don't think so, not the way people say. I remember nights before a deadline, spent building something an LLM can now do in twenty minutes.

I can't say things are better now. Back then, you knew every corner of your own project. Today, nothing stops you from studying your project just as closely. But I believe most people won't — people are lazy.

That's exactly why this needs to be learned on purpose now. In university, I had Cisco Packet Tracer, courses on building a startup from zero — this was before LLMs existed. A thought, an opinion, had its own value back then. Now even a message to a colleague gets written by an LLM, and my job is just to check that it said what I actually meant.

It's about time “software engineer” meant something bigger than someone who writes code and closes tickets — bigger than just writing code. Right now, I think the skills that matter are the ones that help the business and the project actually get things done.

On markets

Economics became my hobby a while back — reading books, thinking about business, studying the history of the stock market. That habit is exactly why this bothers me: business decisions are more and more handed to a model's answer, the same way code is. A model can read every market report ever written, but it has never actually lost money in a crash, and it has no real stake in the outcome. You can only judge how good those decisions really were once the next crisis hits — and by then, it's too late to undo them.

On security

In university, we learned how security works — not just how to build systems, but how someone tries to break them. It might seem like LLMs can do that job too now. But trying to break someone else's work isn't natural for a model — it just writes code based on the code it already learned from. It doesn't think like an attacker.

That's why you still see the same old problems in new, AI-written code: an API key typed straight into the file, no check on what a user actually types into a form, a database query built by gluing strings together instead of doing it safely. These are old, well-known holes. A model can avoid them — but only if someone actually checks for them. The model itself has no reason to look.

There was already code with holes in it before LLMs. Now there's just more of it, and it isn't getting better.

So review the code before you ship it. Don't trust it just because it runs.

On automation

When you're starting out, it's simpler to automate things you could already do in code — it's faster and easier at first. That's actually how I started with n8n too: the backend behind my English-learning app and the backend behind my portfolio site were the first things I built with it. Later I found n8n's API, and things changed: I could build a workflow that builds other workflows.

The English app splits a book into pages, caches each page after the first read, and translates words or phrases with an LLM using the full sentence for context, then caches those too. On the other side, I built a pipeline that pulls from RSS feeds and posts to Telegram and YouTube on its own, almost no manual work needed. That's the real skill: connect enough small pieces, and you're managing a system, not single tasks.

Closing

Put this together, and “software engineer” isn't just someone who writes code anymore. It's someone who can understand and handle almost every part of a system — the code, the business side, the security side, the automation — because AI tooling makes that possible now. That's the bigger definition.

And let's be honest about how business actually works: no company is going to pick a developer who refuses to use AI over one who ships ten working features with it. The quality of those features is a fair question. But the company gets paid either way, and that's what actually decides who gets hired.

Border collies are bred to be the fastest, most obedient thing on the field. They never choose the field.

The interesting question this decade isn't how to run faster than the model next to you.
It's how to stop being the dog.

And in case I don't see you: “good afternoon, good evening, and good night” — Truman had the right idea, walking out that door.

Comments
  • Be the first to comment.