Software is done, engineering is just starting ​

The question we all have to think about has flipped.

I used to spend most of my time on a new feature researching: which tool, which library, which approach. Then prototyping. I’ve stopped doing that, because I no longer need ready-made tools. Every “I wish I had time to bring that tool into my life” and every “if only that app had this one feature” I ever said: that feature exists now. I just ask for it.

That leads straight to the future of software, and of engineering.

Code is the new binary ​

When work that took days or months takes minutes or hours, why would anyone need people who write code?

Engineering was never just writing code, or working out how to solve something. AI has solved that part. The real engineering is working out how the solution breaks.

Five years ago almost nobody wondered what the binary of their code looked like. Today nobody needs to wonder about the code itself. The hard part is somewhere else: seeing where the product will hold up and where it won’t, what could go wrong and where, and knowing when to stop building.

A team for anyone who wants one ​

Anyone who wants a software team can have one now.

For two years I tried to build an orchestrated team of agents. I failed five times. With Orka I finally got it to the level I wanted, because model intelligence finally got there too. On one project I kept a single session open for days, talking to one place, and we finished thousands of small tasks in a week. A few years ago I did work of a similar scope with a team of ten, over months. This time, in one week, it became exactly the product I had in mind. And while it ran, a pile of hobby projects and “I’ll do it later” work projects got done in parallel.

Did I read every line? No. Do I need to? Absolutely not. One model writes, another reviews, finds the problem, and the first one fixes it. My input in that loop is zero. AI has shown me again and again that its code is far better than what people write.

To anyone still saying “AI slop” or “AI opens security holes”: look at how many Linux kernel vulnerabilities AI has found in the last year, like a remote bug that sat unnoticed for 23 years or a root exploit hidden since 2011. Systems that thousands of people built over decades still had those holes. What a capable engineer with AI at hand can close is a different thing altogether.

The quality of what I build with AI today is far higher than what I got when I managed dozens of people. The edge cases I can test, the breaking points I can predict and check, handing out work and actually getting a result at the end: it’s a different level.

Things that are now worth two minutes ​

In the last few weeks I wrote about ten separate services to manage our company’s servers and infrastructure. Each one hooks something I kept doing by hand into our company’s assistant, so it does it instead.

Was it necessary? The manual job took maybe ten minutes. Having one built took about two.

I always wanted to talk to my computer and get something done while my hands were full. I wrote a two-line prompt. Now I control my computer by voice when I need to.

These used to be things that needed a lot of free time, or things you’d do once they were truly needed. Now they’re side jobs: let it get done in the background.

The end of the profession ​

In a few years, software as a profession has landed next to tinsmiths, carriage drivers and calligraphers. When my eight-year-old daughter can sit down and talk her way into the game she wants, writing software is no longer a profession.

Engineering is another story.

Controlling my computer by voice, or automating manual work in minutes, opens a new window for me. The question used to be how to write it. Now it’s how to make it better, and what useful thing I could build next. We’ve reached the years where engineering gets to shine.

The baklava master down the street won’t come to you anymore saying “make it like Facebook, but with prayer times on the side, I’ll pay 10,000 lira”. He’ll build it himself, and he’ll be its only user. The real work will be solving that person’s problem, and finding the thing he couldn’t have thought of on his own.

Engineers who look at the world’s problems and say “here’s a solution I propose”, or who say “I saw this problem at the company and I fixed it”, will move ahead. The ones who say “give me a task and I’ll do it, I did my job, I’m stepping aside” will disappear like the tinsmiths.

Where the engineering is now ​

Code depth and repetitive work used to be a weight on companies. A lot of engineering was finding ways to fight them. They hardly matter now. Real engineering belongs to the people who notice that one call takes 10 ms and another takes 15, and go find out whether they can bring it down to 10.

Off-the-shelf software is ending too. Instant, made-to-fit tools will win over all-in-one suites. Nobody will have to use accounting software with 500 features just to track income and expenses.

For most companies a software developer used to be an unnecessary role. Now any company can hire one in-house engineer who can do everything. Call it efficiency engineering. That is the thing to become: someone who follows the whole process and says “if we add this here, we gain 1%, so let’s add it”. Good developers used to do that for the part they owned. Now they own everything.