UCLA Study Finds 18% of AI Agent Output Tokens Are Wasted
A UCLA and Northeastern University paper introduces a proxy-based audit that uses an LLM judge to mark removable spans in agent output, then reruns tasks to verify nothing breaks. Across six frontier models and four benchmarks, roughly 18% of output tokens were waste; pruning repetition and filler from Kimi-K3's reasoning on SWE-bench Verified cut 41% of tokens while success held at 89.5% versus 89.0%. The takeaway: cut agent costs by locating waste in your task and harness, not by capping tokens.
Related Entities
Full Report
Rohan Paul (@rohanpaul_ai) · 2026-11-09 12
New UCLA+ Northeastern University paper finds that about 18% of AI agent output tokens are removable waste, and cutting it barely hurts task success.
Agent bills are usually measured as total tokens. That number can't tell you which tokens did useful work and which just repeated a policy, refetched a file, or went down a dead end.
They route agent calls through a proxy and have an LLM judge mark spans that can go without losing useful work. Then they delete those spans and rerun the agent to check nothing breaks.
Across 6 frontier models and 4 benchmarks, about 18% of output tokens were waste. Pruning repetition and filler from Kimi-K3's reasoning removed 41% of it on SWE-bench Verified, and success stayed at 89.5% versus 89.0%.
To cut agent costs, find where your task and harness create waste instead of just capping tokens.
