Agno claims its agents start in 2 microseconds

Agno says its runtime spins up a new agent in under two microseconds, using about 3.75 kilobytes of memory per instance, according to the project's GitHub page.

That matters once a workflow is running thousands of agents instead of one. Startup time and memory footprint turn into a real cost line, not a rounding error next to which model you picked.

The catch: the numbers come only from Agno's own repo. There's no independent benchmark cited, no load test run by anyone outside the project, and no head-to-head against other agent frameworks doing the same job. Until someone outside Agno reproduces it, this is a single-source claim, not a verified spec.

The framing behind it is worth separating from the number itself. Most agent framework pitches compete on which model wins a benchmark. Agno is arguing the bottleneck moves somewhere else once you're spawning agents by the thousand: the cost of spawning itself becomes the constraint, not model quality.

If nobody outside Agno reproduces that microsecond figure under real load, it's a marketing number, not an architecture advantage. I'd rather see one team running Agno at swarm scale publish their own measurement than take GitHub's word for it.