AWS Builder Center

the one-mouse problem: why multi-agent systems can't actually multitask

everyone wants multi-agent systems. nobody talks about what happens when all those agents need the same physical computer.

the one-mouse problem

multi-agent systems usually sound like a concurrency problem.
spawn more agents. split the task. let them work in parallel.
then you put those agents inside a real desktop and discover something slightly embarrassing:
there is only one mouse.
that became one of the more interesting engineering problems while building klip, a screen-aware desktop companion that can reason about tasks and actually operate a computer.

the idea

klip can take a request such as:
"open chrome and find the latest pricing for this product, then put the results into excel."
a single agent could attempt the entire task sequentially.
but some desktop tasks naturally contain independent pieces of work.
so klip has a workforce architecture.
a lightweight commander first decides whether a request genuinely contains independent subtasks. if it does, the task can be divided between named workers.
if it doesn't, klip simply uses a single worker.
that fallback is important.
multi-agent isn't automatically better.
sometimes the best architecture is one agent doing one thing.

where the problem starts

imagine three workers:
commander
|
+--- worker: research
|
+--- worker: browser
|
+--- worker: spreadsheet
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article