MOC: Agent Workbenches

Map of Content for the emerging category of environments that run, coordinate, and make visible one or more coding agents. These products sit between a coding-agent harness and a task manager: they provide a place to launch work, inspect progress, retain context, and sometimes let agents manage other agents.

Entries

• bb — local-first, open-source IDE/orchestrator whose agents can build plugins that reshape the workbench itself.
• pi.dev — minimal, terminal-first harness with deep extension and package support.
• paperclip — agent-team management built around goals, roles, budgets, and governance.
• claude-code-agent-teams — Claude Code's native, experimental team mode: a lead plus teammates sharing a task list and mailbox, for one bounded job at a time.
• claude-projects — Anthropic's own coordinator-plus-threads workbench for Claude Code: one conversation delegates to parallel cloud sessions.
• hn-agent-control-planes-survey — what Hacker News has upvoted in this category since 2025, grouped by kind, and the gap nobody fills: a shared work graph and control plane.
• delta-zed — Zed's multiplayer environment: teammates share one live agent thread, with conversation anchored to code via its own VCS (DeltaDB).

Why this category matters

The question is no longer only which coding agent to use. As several agents and long-running tasks become ordinary, the surrounding environment determines how work is queued, observed, resumed, customized, and handed off. The category is still crowded and its boundaries are unsettled: a tool may be an IDE, a harness, an orchestrator, or a task system at once.