FigBrains releases RoleCUA: 780+ hrs of expert data for multi-month AI agents
vsikka · x · 2026-11-21
FigBrains previewed RoleCUA, a frontier dataset targeting entire senior-level job functions over multi-month horizons, with 780+ hours of trajectories from dozens of expert roles across design, ops, and engineering.
- Author Harsh Sikka frames models as hill-climbing a capability ladder — single actions, short tasks, long sessions, then open-ended multi-goal horizons — with each step demanding longer-horizon, higher-signal data.
- This has spurred a new class of data/environment methods: compressing sparse months-long expert trajectories into scalable learning signals (RoleCUA and RealCUA are early results).
- Coverage spans docs, planning, research, code review/QA, UI/UX, and outreach, across tools like Chrome, Slack, Superhuman, and Zoom. Academic groups can request access; more results promised in coming weeks.
More from coding & agent
- rauchg ships a personal shopper agent on eve, buys Christmas gifts with it — cramforce · 2026-11-21
- Coding agent drains Mac battery to 4% — Ghostty flagged for 'significant energy' — MikkoH · 2026-11-21
- Sentry CEO: give your agent a read-only DB replica to answer product questions — zeeg · 2026-11-21
- Codex revamps threads: 60-70% faster cold start, Plus gets ChatGPT Space — SIGKITTEN · 2026-11-21
- U-Mich paper: coding agents report false p<0.05 findings in 54–98% of runs — rohanpaul_ai · 2026-11-21
- Agents can now buy stock market data autonomously via x402, from $0.02 per call — virattt · 2026-11-21