My 2025 Ai Retrospective
2025 was by far the most interesting year of my tech career. The industry moved so fast that I’m convinced we’re living through a once-in-a-lifetime change in how we build software and work with it.
The emergence of agentic coding
At the end of 2024, when I was working at Optibus, all we talked about was how handy Copilot’s autocomplete was. I quickly got into the habit of writing a comment describing what a function should do and waiting for Copilot to suggest an implementation. It felt like magic!
In February I started using Cursor, which let me drop the “comments hack” and just type what I wanted into the chat bar. Back then it only worked within the open file or its tests.
Write a function that will read the format of the Authentication Headers of the received request and return which authentication mechanism should be used.
It still felt like piloting a machine. I was telling it exactly what to do.
Around May, I think, I tried Claude Code for the first time. My employer was paying for Cursor and ChatGPT licenses, but I was curious about a tool built around the command line, so I paid for it out of my own pocket.
A couple of weeks later I was reviving old, forgotten projects, doing big refactors in some of them, and finally starting ideas I’d never had the courage to begin.
From coder to architect
The more I used Claude Code, the more I trusted it, and the more room I gave it to figure out on its own how to get where I wanted. Hallucinations were common, and I had to correct course often, but I still felt more productive than before.
As I got better at using AI, I spent less time typing code and more time planning and describing solutions to high-level problems.
A few months later I joined Mighty Networks, a company that had fully adopted AI-augmented development. I already had my own set of MCPs, a curated CLAUDE.md, and a pretty good sense of how to get the most out of the tool. Still, being somewhere that fully supported and encouraged exploring AI is what finished converting me.
That support meant:
- My employer paid for enough tokens that I could run several development sessions in parallel
- An internal community shared experiences and best practices
- AI was built into different steps of the development workflow (development, code review, troubleshooting…)
Looking back at the last few months, I’m far more productive than I’ve ever been, and, paradoxically, I’ve never written so many lines of code.
The question for 2026
Will I miss writing code myself?
That’s a topic for another post. :)