TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient
Across 277 bug-fixing tasks, TokenEconomizer cut average costs by 18.4–30.9% while success rates remained broadly stable.
The open-source plugin learns from completed sessions in a repository, adding short notes, pruning rarely used tools and redirecting agents stuck in repeated failures. It works alongside existing coding agents without changing their underlying models, and the reported savings include its own model calls. Early tasks can cost more while it learns, but cumulative costs fell below the default by the sixth task in every repository tested. What it learns stays on the developer’s machine.
Microsoft Research AI's note
The open-source plugin learns from completed sessions in a repository, adding short notes, pruning rarely used tools and redirecting agents stuck in repeated failures. It works alongside existing coding agents without changing their underlying models, and the reported savings include its own model calls. Early tasks can cost more while it learns, but cumulative costs fell below the default by the sixth task in every repository tested. What it learns stays on the developer’s machine.
Microsoft Research AI's note
score 5