← Back to research

Research Brief · 2026-11-09 Core

AutoBuddy: Evaluating Proactive, Always-On Agents on Task Discovery and Execution in a Simulated Workplace

The idea, in brief

An always-on workplace agent has to find which emails, chats and records hold work the user owes, then do it. Seventeen models in one simulated company.

Editorial summary · not the author abstract. Read the linked paper for the complete abstract and claims.
Figure 3: read-only sessions build a work profile and mine the workspace for owed tasks, separating direct requests from others' work, answered threads and automated notices; each match is then executed in its own session.
Figure 3: read-only sessions build a work profile and mine the workspace for owed tasks, separating direct requests from others' work, answered threads and automated notices; each match is then executed in its own session. Original figure · paper authors ↗

Release evidence

The paper links no code or benchmark and states no release plan; a GitHub search found only unrelated projects.

Last checked · 2026-11-15