Co-op mode: designing brands as a team in real-time

Almost every meaningful brand identity decision involves more than one person. A founder and a designer. A creative director and three juniors. A marketing lead, a product lead, and a brand consultant. Even solo freelancers loop in clients constantly.

And yet most AI design tools treat the workflow as a single user in a single window, talking to a single model. Then they bolt on "share a link" as an afterthought.

Co-op mode in VisionLabs is the opposite premise: brand work is multiplayer from frame one. Every chat can have multiple participants live, every message is attributed, and the AI sees the full conversation regardless of who typed which line.

What it actually looks like

You're working on a logo. You click "Invite collaborator", paste an email, send. Two minutes later your co-founder accepts the invite and lands in the same chat — with the full history visible.

You both type at the same time. Each of your messages is tagged with a quiet, mono-grayscale attribution ring (we made it gray instead of color on purpose — see the color post). Live presence shows you who's currently typing and who's just reading.

You ask the swarm for three logo directions. Your co-founder reacts to one in the chat — "the second one but tighter spacing". The swarm picks up the message and iterates. No screen-share, no back-and-forth screenshots, no "version 17 final FINAL.fig". One canonical conversation, one canonical brand.

Why this changes review cycles

The old loop: designer produces, sends file, client reviews, client emails feedback, designer interprets, designer revises, repeat. Each handoff loses context, takes a day, and introduces interpretation drift.

The co-op loop: everyone's in the same room. Feedback is in the chat next to the artifact it refers to. The AI has the same context as the humans. Revision happens in minutes, not days. Nobody asks "which version are we talking about?" because there is only one version.

This compresses brand projects that historically took weeks into days. Not by making the work faster, but by removing the dead time between rounds.

The transfer-ownership pattern

One feature we hadn't anticipated needing but kept being asked for: transfer ownership of a conversation. The use case is freelance: a designer builds the brand with a client, then formally hands the project over so the client owns it going forward.

It's a one-click flow in the Invite modal. The current owner picks a new owner from the participant list. Future billing, access controls, and export rights move to the new owner. The original owner can remain as a collaborator or leave. Clean handoff, no data migration drama.

What surprised us

Two things came out of beta that we didn't predict.

1. People use co-op asynchronously

We built it for real-time. Most of the actual usage is async — one person works in the morning, another picks up in the afternoon, they read each other's messages like Slack. Live presence is appreciated when it happens but not the primary mode.

So we added the "while you were away" digest: when you return to a conversation after >2 hours, the top of the chat shows a summarised list of what happened in your absence — new messages, generated assets, decisions made. Catch up in 30 seconds.

2. The AI's "memory" is shared

When two people use the same conversation, the AI remembers their combined context, not just yours. If your co-founder told the swarm "we hate red" and you later say "try a warmer direction", the swarm won't show you red. That shared memory is what makes co-op feel like a real workspace rather than a chat window with multiple users.

The single-player path is still there

Nothing forces you to invite anyone. Most conversations are solo and stay solo. But the option to slide a collaborator in mid-project — with zero friction, full context, and proper attribution — is the difference between a tool that can do teams and a tool that understands them.

Brand work was always multiplayer. We just finally made the chat match.