Why it matters

Agent scaling shapes agent design. Understanding shapes strategy.

Advertisement

The architecture

Model size vs tool count vs episode length.

Capability curves per axis.

Interactions matter.

Agent scaling axesModel sizecapability floorTool countcapability breadthEpisode lengthtask complexityFrontier: agent capability scales roughly with pretrain capability + tools + retry
Agent scaling.
Advertisement

How it works end to end

Base model quality dominant.

Tool selection quality matters.

Episode length increases capacity.