Back in 2019, before AI in software development became part of everyday engineering work, I attended the London SC conference. It was the first time I saw code fully generated by a machine.
During Frances Buontempo’s talk, “What’s Machine Learning Got to Do with It?”, a computer was given a simple task: implement the FizzBuzz kata. It worked. The solution was correct.
And the whole room burst laughing.
The code was painful to read.
That moment planted a question that stayed with me:
If computers can write code themselves, will humans still need to read it at all?
Fast forward to now
Several years later, AI tools are part of our daily work. Despite their massive impact, there’s still a large group of laggards – people who say AI is “just a junior”, that it can’t handle complex problems, and that they are faster and better without it.
That reaction isn’t new. It’s classic innovation adoption curve behavior.
The difference is this: our SDLC has already changed – and it’s not going back.
What I’m seeing in real projects
Just yesterday, in one of the projects I am involved in, the delivery flow looked like this:
- A task is created in the ticket management system
- The description is written (or refined) using AI
- A software engineer copies the description into a coding agent
- The agent implements the changes, pushes code, and opens a PR
- Another agent reviews the PR and leaves comments
- The developer mentions yet another AI agent to apply fixes
- The PR gets merged
Needless to say, this process can be improved. But that’s not the point.
Back in 2019, generating poor-quality code with AI took hours. In 2025, you can ship a small feature in minutes. That shift already happened.
Developers write less code – and that’s a fact
For sure, developers today write less code than they used to.
They prompt, observe, prompt again, review, adjust. They guide more than they type. Writing code is no longer the dominant activity – decision-making is.
We need to accept that.
Let’s be honest: software development as we know it today will look very different in just a few years. Many people predict that our role will turn into that of a technical supervisor. In reality, we are already there.
AI writes the code. We review it.
The uncomfortable question is:
Will we need as many engineers to do that?
What happens to junior developers?
Traditionally, juniors differentiated themselves by learning syntax, patterns, and frameworks faster. With AI, that advantage almost disappears.
That doesn’t mean juniors are doomed – but it does mean the old growth path no longer works. The value won’t come from how fast you write code, but from how well you understand problems, constraints, and trade-offs.
And that’s a much harder skill to learn.
Maturity matters more than ever
The 2025 DORA report suggests that AI boosts performance mainly in mature environments. Strong platforms, clear processes, and well-defined standards are no longer nice to have – they are mandatory.
Without them, AI doesn’t accelerate delivery. It amplifies chaos.
With them, something interesting happens.
We become more outcome-driven. Business brings problems. We propose solutions. AI helps implement them. Features ship faster, often imperfect on day one – and that’s fine.
They will be improved. Iterated. Fixed.
So… will humans still read code?
Maybe less than before. Maybe they won’t at wall.
Understanding systems, making decisions, asking the right questions, and taking responsibility – those things won’t be automated anytime soon.
The craft is changing.
The role is shifting.
And pretending otherwise is the real risk.