UC Berkeley Supervisor Rebuilds Coding Agent Teams to Beat Expert GPU Kernels

1 reports archived1 independent sourcesupdated 11/10/2026, 05:22:55

Synthesis & Latest Updates

1 Sources Cross-Validated

A new UC Berkeley paper introduces Evolving Agent Teams, in which a supervisor diagnoses plateaus in coding-agent teams and rebuilds roles such as profiler, debugger, and knowledge integrator, with a shared file system carrying lessons across teams. On the MLA decode kernel it reached 0.74x FlashInfer speed under the same token budget versus 0.31x for a fixed four-role team, and 1.21x with more budget, winning 36 of 47 production workloads.

LATEST/A new UC Berkeley paper introduces Evolving Agent Teams, in which a supervisor diagnoses plateaus in coding-agent teams and rebuilds roles such as profiler, debugger, and knowledge integrator, with a shared file system carrying lessons across teams. On the MLA decode kernel it reached 0.74x FlashInfer speed under the same token budget versus 0.31x for a fixed four-role team, and 1.21x with more budget, winning 36 of 47 production workloads.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

Rohan Paul (@rohanpaul_ai)T2Nov 10, 05:22·78 pts
UC Berkeley Supervisor Rebuilds Coding Agent Teams to Beat Expert GPU Kernels
- A supervisor that diagnoses plateaus and rebuilds roles outperforms fixed agent configurations
- 0.74x FlashInfer under equal token budget versus 0.31x for a fixed four-role team
- 1.21x FlashInfer with more budget, winning 36 of 47 production workloads
View Related Reports & Analysis Original Source