Software is dying, you are not

Hey {{first name | there}}.

Software is changing, no doubt. A lot of past issues have touched on various portions of this, from everyone being able to code now to handing your agent the keys.

This weekend I was reading Alex Pliutau's post on the future of software engineering, and I have an updated perspective. Software as we knew it might be dying, but you are not.

In today's technical notes:  

Alex has been programming since around 2005, and his post is titled "The future of Software Engineering is boring". He lists three problems with how we build software now.

The first is learning. You used to hit a problem, go read a book or a Stack Overflow thread, think about it, and then arrive at a solution. That middle step is where the understanding came from. Now you can go from problem to solution in one prompt.

The second is what he calls codebase erosion. AI adds code far more often than it removes it, review gets skipped or handed to another model, and the codebase grows faster than anyone's understanding of it. He uses an old reviewer joke to make the point: "Give a reviewer a 10-line Pull Request and he will find 10 issues, give him a 500-line one, and he will say LGTM."

The third is noise. Generated emails and documents that someone else now has to read, and managers who expect everything faster because AI exists.

I agree with most of this. The line from his post that stayed with me was "we've become a caretaker of code that nobody really understands."

So yes, the way software gets written is changing, and some of it is getting worse.

I see it a little differently from Alex. You are still the center of driving software. A model can write the function, but you decide what gets built, what gets merged, and what happens when it breaks in production.

What did change is that nobody is going to make you learn anymore. Before, the research step was forced on you because there was no other way to get to a solution. Now it is optional, which means your learning and upskilling depend entirely on you.

Anthropic ran a study on this earlier this year. 52 mostly junior engineers learned Trio, a Python async library they had never used. The group that used AI scored 50% on a follow-up quiz, against 67% for the group that coded by hand, and they were not significantly faster.

In the AI group, the people who asked conceptual questions, or asked for explanations alongside the code, scored between 65% and 86%. The people who handed over the whole task scored between 24% and 39%.

The craft Alex misses might be dying, but whether you keep growing is still a choice you make every time you open a prompt.

Ask why the code works, read the diff before you merge it, and pick up the book when you have the time.you already have. The win is optional, which is exactly why I am willing to poke at it.

How AI assistance impacts the formation of coding skills by Judy Hanwen Shen and Alex Tamkin. The study behind the quiz numbers above, including the six ways people used AI and which ones kept learning intact.

The 70% problem: Hard truths about AI-assisted coding by Addy Osmani. Why AI tools help experienced engineers more than juniors, and how to use them as a teaching aid instead of a shortcut.

AI Copilot Code Quality: 2025 Look Back at 12 Months of Data by GitClear. The data on cloned code rising and refactoring falling as AI assistants spread.

That’s it for this issue.

Know an engineer who will find this helpful? Share this link with them

Jubril Oyetunji
CTO, EverythingDevOps