Great new paper from AWS on agent handoff tax.

If you build agents today, you need to understand the so-called handoff tax.

(bookmark it)

Escalating to a stronger model mid-run is usually the resort when a cheap agent stalls.

New work from AWS AI Labs measures how much that switch actually costs.

Coding agents run for dozens of model calls, so teams escalate when a weak model struggles and downshift once the hard reasoning is done. Every switch forces the receiving model to continue a trajectory another model wrote.

Across pairs of Claude and GPT models, full-trajectory escalation recovers less than half the quality gap between the weak and strong model while adding a substantial cost premium. The authors call that penalty the handoff tax. Downshifting lands at a much better cost-quality point.

Cutting the weak model's trajectory information improves escalation quality, while removing the strong model's trajectory hurts downshift quality.

Paper: https://t.co/59ozKgugB8

Track more trending AI papers in our academy: https://t.co/1e8RZKs4uX