Build a production-ready multiplayer AI workspace that lets a whole team collaborate with AI agents in real time — the way Figma turned design into a shared, live experience.
Core vision:
Today AI sessions are single-player: one person prompts one agent in isolation. Build the multiplayer version for engineering teams, where long-running agent tasks become shared team experiences that anyone can watch, redirect, or take over.
The system must support real-time, multi-user collaboration around agents:
1. Start a long-running agent task in a shared workspace
2. Let teammates see the agent's progress live
3. Allow anyone with access to redirect or steer the agent mid-task
4. Support clean hand-off of a running task from one person to another
5. Keep a shared history of what agents did and why
6. Turn individual sessions into a team knowledge base
Core capabilities:
Shared agent sessions:
- Real-time view of an agent's plan, steps, and outputs as they happen
- Live presence and cursors showing who is watching or steering
- Comments and inline direction on any step
- The ability to pause, redirect, or approve an agent mid-run
- Hand-off controls to transfer ownership of a running task
Team workspace:
- Projects and channels grouping related agent work for engineering
- A team activity feed of running and completed agent tasks
- Roles and permissions (owner, editor, viewer)
- Templates for common engineering workflows
Long-running task management:
- A queue and status board for in-flight agent tasks
- Notifications when a task needs human input or finishes
- Automatic capture of context so a teammate can jump in cold
- Retry, fork, and branch controls for agent runs
Knowledge and reuse:
- Searchable history of past agent sessions and outcomes
- Save successful runs as reusable playbooks
- Metrics on agent throughput, hand-offs, and interventions
Design:
Fast, collaborative, and alive — real-time presence indicators, smooth updates, and a layout that makes a running agent feel like a shared document the whole team is in together.
Builds a working MVP of a real-time collaborative workspace for teams working alongside AI agents. Includes shared agent sessions, live presence, the ability to watch/redirect/hand off long-running agent tasks, and a team activity feed. Real-time sync is implemented at the app level with an agent task simulator.