I'm on the fences on Context Isolation as the absolute rule/cure... don't take me wrong, sounds pretty clean, and I see the value but in practice it can either reduce bias/noise or destroy helpful continuity depending on where you draw the boundary.... there is some tension for fast-pacing prototyping tasks for instance... regardless, good read, thanks!
For exploratory coding, I also often prefer long single-sessions jams, especially with large context models like Opus, because we can go back and forth from short planning to implementing to planning again.
Good question. Five has been the answer more or less for 25 years so far, so I think we're covered on that ground. We are rediscovering this workflow in the age of AI agents because few things take more advantage of structured problem solving.
Great post Alejandro. I'm going to give it a go.
I'm on the fences on Context Isolation as the absolute rule/cure... don't take me wrong, sounds pretty clean, and I see the value but in practice it can either reduce bias/noise or destroy helpful continuity depending on where you draw the boundary.... there is some tension for fast-pacing prototyping tasks for instance... regardless, good read, thanks!
For exploratory coding, I also often prefer long single-sessions jams, especially with large context models like Opus, because we can go back and forth from short planning to implementing to planning again.
Obvious question: is five the end point, or is N+1 always better than N?
Before there was the five-point plan there were weekly or monthly meetings.
Good question. Five has been the answer more or less for 25 years so far, so I think we're covered on that ground. We are rediscovering this workflow in the age of AI agents because few things take more advantage of structured problem solving.
There's an old saying that each generation of technology rediscovers the lessons of the previous one.