Summary
According to a post on X by @Marktechpost, Microsoft Research released Agensh under the MIT license. Every worker runs the same claim, build, verify and merge loop and coordinates through a shared Git repo and chat channel instead of an orchestrator, with support for up to 1,024 agents. On the five hardest ProgramBench tasks, the mean pass rate rose from 19.31% with one agent to 28.78% with 128. Going from 1 to 8 agents added only 1.37 points, and token costs were not disclosed.
AI-assisted summary based on the listed source.
What happened
Microsoft Research released Agensh under the MIT license; every worker runs the same claim, build, verify and merge loop and coordinates through a shared Git repo and chat channel instead of an orchestrator. On the five hardest ProgramBench tasks the mean pass rate rose from 19.31% with one agent to 28.78% with...
Why it matters
Teams building multi-agent coding systems may find Agensh a concrete reference for coordinating many agents without a central controller. The reported gains appear at higher agent counts, while the small early improvement and undisclosed token costs leave the cost-benefit trade-off unclear. Readers should watch for independent benchmark results and published cost figures before drawing conclusions.
Signal Intelligence
Signal Strength 91%
Technical label WATCH
Public Interest 35
Category OPEN SOURCE
Reader Depth TECHNICAL
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
Public Interest components
Recognizable Entity Score 0
Practical Impact Score 26
Novelty Interest Score 94
Consequence Score 30
Curiosity Score 16
Shareability Score 50
Why this is here
VQV surfaced this signal because it is recent, relevant to AI Coding Tools, connected to Scobleizer News Pulse (X).
