From the source
Amp's team argues that a 200k token context window is sufficient for coding, because short threads produce better results and cost less than long ones.
The post describes how the team breaks features into small, single-task threads and shares context between them using thread mentions.
