Chat is a terrible UI for hard questions. We replaced it with a canvas where 3 agents investigate, then debate.

Spent the last few months talking to people running heavy questions through chat( investments, career moves, technical decisions). And I kept hearing the same thing, chat threads get lost in the weeds, multiple tabs of model comparison are super tedious, and deep research is too much to read and unreliable. So we built a canvas mode. You ask one question. Three agents (normally chat 5.5, opus 4.7 and gemini 3.1) instances investigate different angles in parallel different framings, different evidence bases. Then they actually debate each other. You watch the disagreement and steer.