aifollow.news
← TrendingLatest development Nov 17 07:06
STORY IN FOCUS

TokenEconomizer study uses an external controller to cut coding-agent token costs

A UC Berkeley and Microsoft paper proposes an external controller that learns from past coding-agent runs to inject useful facts, remove unused tools, and stop failing loops early. Across 277 SWE-Bench tasks using Claude Code, Codex, and GitHub Copilot CLI, the reported mean per-task cost fell by 18.4% to 30.9%.

1 developments3 reports2 sources

Event developments

Coverage of the same occurrence shares a node; subsequent developments have their own nodes.

UC Berkeley and Microsoft paper proposes external controller to cut coding-agent token costs

The paper proposes TokenEconomizer, which adjusts prompts and tool settings outside the agent. It learns from past runs which facts to inject, which unused tools to remove, and when to stop a failing loop. Across 277 SWE-Bench tasks using Claude Code, Codex, and GitHub Copilot CLI, mean per-task cost fell by 18.4% to 30.9%, while Pass@1 improved in four of five agent-model settings.

View source reports · 2 sources · 3 reports
All reports · 3

By original publication time, with each report's bookmarks and feedback preserved.

Rohan Paul (@rohanpaul_ai) Machine translation

TokenEconomizer study uses an external controller to cut coding-agent token costs

A UC Berkeley and Microsoft paper proposes an external controller that learns from past coding-agent runs to inject useful facts, remove unused tools, and stop failing loops early. Across 277 SWE-Bench tasks using Claude Code, Codex, and GitHub Copilot CLI, the reported mean per-task cost fell by 18.4% to 30.9%.

Rohan Paul (@rohanpaul_ai) Machine translation

UC Berkeley and Microsoft paper proposes external controller to cut coding-agent token costs

The paper proposes TokenEconomizer, which adjusts prompts and tool settings outside the agent. It learns from past runs which facts to inject, which unused tools to remove, and when to stop a failing loop. Across 277 SWE-Bench tasks using Claude Code, Codex, and GitHub Copilot CLI, mean per-task cost fell by 18.4% to 30.9%, while Pass@1 improved in four of five agent-model settings.

Microsoft Research Machine translation

TokenEconomizer open-source plugin uses past sessions to cut AI coding agent costs

TokenEconomizer, from UC Berkeley, Microsoft Research, and Microsoft Security, uses completed sessions in a repository to guide later tasks without changing the agent or its underlying model. Across 277 bug-fixing tasks in six repositories, the team reported average cost reductions of 18.4–30.9% while task success rates remained broadly stable. Early tasks can cost more as the plugin learns.