Pixel Labdeveloper docs

Design record · 2026-10-agent-coworker

Intent: Agents as coworkers, with comments#

Author: J. Law Cordova (owner). Status: draft.

Problem#

In the owner's words:

Humanizing of agents in the Pixel Lab. Right now, the agents are just chat boxes, but I want to have a comment system that actually treats them like a coworker.

Today the only way to work with an agent in Pixel Lab is the chat in the Agent tab (../2026-10-agent-chat/intent.md) or a terminal. A chat is a conversation off to the side: it isn't attached to the art it's about, and it doesn't feel like working with someone on the document.

Proposed outcome#

Agents in Pixel Lab feel like a coworker rather than a chat box. The owner can leave comments on the art, and agents take part in them the way a colleague would: reading them, answering them and acting on them, with the work they do showing up where the comment was made.

Affected users and systems#

  • The owner in the app, in scene mode and object mode.
  • Agents, in the app's chat and over MCP.
  • The document, if comments are kept with it.
  • The Agent window and its proposals.

Constraints#

  • What agents change stays a proposal the owner approves.
  • VoiceOver labels, visible focus, comfortable hit targets and Reduce Motion on anything new.
  • No new dependencies without asking.

Open questions#

  • What makes an agent feel like a coworker beyond comments: a name, an avatar, presence on the stage, a history of what it did?
  • Where can a comment be left: on a scene, an item, a layer, an object, a pixel area?
  • Do comments replace the chat, sit beside it, or turn into chats?
  • Are comments kept in the document (seen by every agent and the CLI), or only in the app?
  • Can an agent start a comment thread itself, or only answer the owner's?
  • Is there more than one agent coworker, each with its own identity?