Thanks! It's not really about the length, it depends the scope of the actual edits. It helps to keep the "how do I do the edit" and "what do we want to change" as two separate contexts. So at minimum we'll have two agents if there's no thinking step.
The "supervisor" agent is careful not to spawn concurrent agents when the edits are at risk of conflicting, so the exact number is just a matter of how "broad" the changes you're asking for are, not how long the document itself is.
Very cool work and writeup. Why do we need a swarm for shorter docs not just 1 agent? And does the number of agents scale with the doc size?
Thanks! It's not really about the length, it depends the scope of the actual edits. It helps to keep the "how do I do the edit" and "what do we want to change" as two separate contexts. So at minimum we'll have two agents if there's no thinking step.
The "supervisor" agent is careful not to spawn concurrent agents when the edits are at risk of conflicting, so the exact number is just a matter of how "broad" the changes you're asking for are, not how long the document itself is.