How AI coding agents changed the way I work

Since early 2025, there’s no denying it: AI-assisted coding has completely changed the way I work.

For the less technical readers: IDEs (the development software you see on developers’ black screens) now ship with very powerful assistants built on LLMs. You may have heard of VSCode with Copilot, Cursor or Windsurf.

It happened in three big steps:

Smart autocomplete (from 2021), which guesses the end of a line or a small block of code so you don’t type it all by hand. Already handy.
ChatGPT and other assistants built right into the IDE, in 2023. That saved countless trips to Stack Overflow and let you generate or rewrite whole chunks of code. A big time saver.
“Agents”, from 2024 (though for me the revolution really hit in early 2025). They can move between the files of a project, understand the dependencies and complete complex tasks on their own.

From where I sit as a data scientist, agents are astonishing. They completely change my relationship with development.

These days, as soon as I step outside my area of expertise in code, I switch to “prompt” mode: I describe what I want, let the agent propose an implementation, then check and adjust.

A concrete example: I’ve built several front-end interfaces, perfectly presentable, to make applications usable by anyone. Almost entirely by typing English into a chat box. That used to take me days. Now it’s wrapped up in two hours on a Friday afternoon.

In short, for adjacent roles writing non-critical code (that is, code not headed to production for the general public), it multiplies productivity in an almost magical way.

In the next post, I share my detailed recipe for getting the result you want 😉