Geoffrey Huntley’s post on X about how addictive coding with AI is 100% true. The flow state is highly addictive. Before using LLM’s, when I would be working on a bug, I didn’t want any distractions. I would wait until I knew I had a block of time to solve it, which inevitably led to procrastination because I couldn’t find the “perfect time”. It would take so much effort to enter that flow state, but once I did, it felt incredible. I think that’s what we are all chasing when we fall in love with coding. The flow state.

The beauty of coding with an ai agent partner is that you can stay in that flow state for much longer. You can stay in the flow state even during interruptions. Debugging issues used to take hours of intense focus, now the LLM can be released into a debugging loop and solve the problem while I work on something else. It could be work related or unrelated to the project at hand. That’s the beauty, it free’s up precious CPU cycles for me to focus on more important things than sifting through massive C++ compile error messages.

I often tell my boss, features are cheap. This is why. Features used to be a really resource intensive thing. Planning, disecting code, debugging issues, writing tests… All of this is an order of magnitude easier. I can focus on the planning aspect much more than the implementation. A powerful realization I had was that even before using AI, the better your plan the better your code output will be. Now with AI tooling, planning is an order of magnitude more impactful.