Berkeley and Microsoft paper cuts coding-agent token costs 18.4-30.9% with a black-box external controller
rohanpaul_ai · x · 2026-11-17
A new UC Berkeley and Microsoft paper introduces TokenEconomizer, a black-box external controller that tackles the token costs coding agents rack up across sequences of real tasks, not isolated ones.
- It learns from past runs which operational micro-skills to inject, which persistently irrelevant tools to prune, and when to bail out of unproductive failure loops — changing only prompts and tool settings.
- Across Claude Code, Codex, and GitHub Copilot CLI on 277 SWE-Bench tasks, it cuts mean per-task token cost by 18.4–30.9% with comparable success, improving Pass@1 in 4 of 5 agent-model settings; savings generalize to web development and security auditing (6.4–21.2%).
- Key takeaway: measure agent cost over a whole project and adapt per repository with an external controller rather than fixed plugins.
Code is open-sourced at github.com/microsoft/TokenEconomizer.
More from coding & agent
- A one-line prompt fix for "AI brain fry": limit focus to 1-3 items — gerardsans · 2026-11-17
- Park, a Canvas-Based Mac Browser From Arc's Lineage, Launches Public Beta — round · 2026-11-17
- Vision AI watches a garage door for ~$1.50/year, replacing a hardware sensor — SuB8u · 2026-11-17
- Ex-fintech insider: agentic payments UX is an open market waiting to be owned — SimonBalmain · 2026-11-17
- He automated his writing with a Claude Skill — then used it to write about itself — danshipper · 2026-11-17
- PostTrainBench one year on: AI agents now run complex post-training pipelines — maksym_andr · 2026-11-17