Claude code madness is in the air. It is phenomenal to see how far we have come in the last couple of years. From smarter autocomplete to full blown apps being built purely from terminal commands, it truly looks like the age of the idea guy.
As a design technologist, I had the unique opportunity to sit between non-technical folks like designers, PMs and the engineering team. I've been asking them how they use tools like claude code and codex in their workflow and found an interesting pattern.
The use of AI tools can be broken down into - What gear are you operating your agent in? Similar to a manual car, the higher the gear the higher the speed range and would require more skill from the driver at the wheel.
On one end of the spectrum you have the highest gear going full throttle at the problem. Submit a prompt for the agent to build the whole thing, take a long break, check bunch of other apps and see what is up after 30mins. All gas no brakes.
On the other end is meticulously breaking down problems for the agent and moving step by step, checking small changes every 30sec and moving deliberately in detail and realigning the agent to the strategy if it goes astray
Not surprisingly, the latter is what the skilled programmers use and the former by the non-tech folks