In august 2026 I couldn’t resist the temptation anymore and bought a Claude Pro license for 18€/month. I’m very cautious about using AI in my work. Previous to this “upgrade” I only used Gemini and ChatGPT since early 2026 via the browser UI to ask questions about certain technologies, software concepts and asked for short, mostly trivial but otherwise google-heavy snippets of code that I then manually added to my codebase. Letting a real “agent” loose around my codebase was a profound experience.

As of september 2026, the leading models are incredibly capable. They familiarize themselves with extensive codebases in the matter of minutes, can understand the difference between intended behavior and actual behavior, can point out cases where potential bugs can be hiding and can suggest advanced performance optimizations. Their capabilities are truly amazing.

I see some very real issues with using coding agents but I’m excited about the opportunities. I have a strong belief that there are “Wrong Ways” and “Right Ways” of using them, each leading to vastly different outcomes both in terms of the quality of the software produced and in terms of the impact on the person’s skills using the agent. I believe the way we use them can lead to huge long term gains or huge long term losses.

I’ll try to go into more detail about that in a later post, but for now, I just want to list the things that a coding agent already helped me learn, things that probably would have taken much longer without them:

  1. OAuth2, SSO and Keycloak
  2. setting up a remote debugger with VS Code