Multiplayer.
The unsolved bet. Shared cursors solved co-presence. Coordination is a different problem. Whose judgment wins when multiple humans and multiple agents are working the same project at the same time, with overlapping intent and unequal authority. Co-presence is the easy half. Authority is the half nobody has shipped.
human ─┐ ┌── agent · tests
human ─┼──▶ ┌─────────────────┐ ◀─┼── agent · copy
human ─┘ │ the project │ └── agent · design polish
│ (one surface) │
└────────┬────────┘
│
▼
┌─────────────────┐
│ authority │ whose judgment wins
│ layer │ when the room disagrees?
└─────────────────┘Look closely at what's still broken. Three tensions, none of them solved by sharing a cursor.
Generation vs. iteration.
Two humans prompt the same component. The editor treats the second prompt as the next instruction, not as a conflict to resolve.
Velocity vs. visibility.
An agent makes a structural change while a human is editing visually. The human discovers it after the diff lands.
Authority.
A founder running the Bench has no priority queue and no veto. The bench runs as if no one is watching.
Four patterns the multiplayer UX will lean on. Not the solution. The vocabulary the solution will be built from.
DRI.
Single-name ownership. The room can disagree; one person carries the call.
Suggesting mode.
Non-authoritative input made visible. A change you can see before it lands.
Pull requests.
Approval as a UX surface. Required reviewers, blocking comments, CODEOWNERS.
Permissions + roles.
Who can edit, comment, view. Authority encoded into the artifact itself.
Engineering ships the plumbing. Design ships the model. The team that ships the model first owns the category. Not the team that ships the most agents. The one that ships the interface for trusting them.