There are really two very different ways of coding with AI. One is to quickly iterate with small changes. The other is letting an agent go off and work on a hard problem or high-level goal, where it churns for minutes or hours. For the first, model speed is almost everything.
@manton Which do you prefer? I have been doing both, depending on the type of work. I mostly prefer small changes because it keeps me connected to the work in a much better way.
@st3fan I mostly prefer small changes too. Sometimes I’ll have AI go off and do a first draft of something, and that takes longer, then I iterate and tweak endlessly after that.
@manton I believe there is a third model which I sometimes use which is empowering the agent to make continual small improvements under the banner of a larger task. That obviously requires an agentic approach that not all tools can do.