Factory released Droid 1.0, a model-independent coding agent harness designed to make programmatic tool calling the default interaction mode. On Factory’s Terminal-Bench 4 benchmark, Droid achieved 65.2% success rate with Claude Opus 5.5 at a cost of $4.73 per passed task, demonstrating a practical framework for integrating coding agents across different model backends (MarkTechPost).
In a complementary development, Vercel Labs released agent-desktop, an Apache-2.0 Rust CLI that enables AI agents to perform computer use on macOS, Windows, and Linux by reading OS accessibility trees and acting on stable references rather than pixel coordinates (MarkTechPost). Microsoft Research and UC Berkeley also published TokenEconomizer, an open-source plugin that reduces coding agent token costs by 18.4% to 30.9% on SWE-Bench by learning from past session data, requiring no changes to the agent itself (MarkTechPost).