I started developing in 2012, went through a developer training program, then a two-year technical degree, then a four-year program. During those years and into my early career, one debate came up constantly: specialist or generalist. Symfony against Laravel, Angular against Vue, React against everything else. Two camps, and I was always on the side that said you should adapt rather than lock yourself onto one stack.
That cost me in interviews more than once. Not knowing a library inside out is enough to make a recruiter uncertain. And some technologies genuinely take time before you are actually productive with them. NestJS, for example, is not something you improvise if you show up without the right reflexes. But I held the line anyway.
What actually matters is not the technology
What I understood over the years as a freelancer and then as a CTO is that the technology is rarely the real subject. The subject is the need. What makes the real difference is being the person who cares about what the client does for a living, who understands why a feature is being requested before thinking about how to build it.
A developer who receives a spec and executes without asking questions will deliver exactly what was written. And usually what was written is not what the client needed. This sounds obvious when you say it out loud, but in practice very few people actually take the time to understand the business before opening their editor.
I had been convinced of this for a long time. What I did not anticipate was that AI would make it even more true, and much faster than I expected.
Copilot first, without much conviction
I started using Copilot around 2022-2023. Mostly as autocomplete, sometimes letting it generate blocks of code I would then rework. Honestly, at that point it did not really change my pace. The gain was real but marginal. I was correcting about as much as I was gaining.
About five months ago I changed my approach. I started using AI tools seriously, Claude in particular, and it was a different experience.
What it actually changes
Working with these tools daily, I noticed that the developer's value is shifting. The code itself becomes almost secondary. What matters is how precisely you express a need, how well you catch when the machine is wrong, how quickly you redirect it when it misses the point.
Doing that well still requires solid foundations. Not to write code instead of the tool, but to recognize when its output is incorrect, incomplete, or solving the wrong problem. Machines misread problems that feel obvious to us. They feel obvious because we have years of context behind them. Remove that context and the AI delivers something technically valid that solves nothing.
I had a former colleague who had spent years becoming genuinely expert in one specific technology. That depth gave him a clear market value. Over the past few months that value has taken a serious hit, not because he became a worse developer, but because what he knew how to do is now accessible via a prompt. The generalist who always reasoned at the level of the need does not have that problem.
Choosing Product Management
Since 2026, I have been working toward a Product Management master's degree. It is the logical conclusion of everything I just described.
Understanding users, translating their needs into features that actually make sense, deciding between what is urgent and what is important: that is exactly what I spent 12 years learning to do without ever formally naming it. AI accelerates code production. But someone still has to decide what to build, why, and for whom. That is the PM's job. And that is where I want to be.