Together AI's agent eval workflow lifted task completion from 67% to 96%

togethercompute · x · 2026-11-17

Together AI shares its agent eval methodology: define 10 representative tasks, run Claude Code agents in sandboxes, test 3 configs per task (baseline, MCP, skills), then analyze trajectories to improve docs and APIs. Task completion improved from 67% to 96%, with the chart showing baseline at 82%, MCP at 96%, and skills at 97%.

Original post →

More from coding & agent

coding & agent channel →