Episode 2 — Chat, in depth
The daily surface, used well — the same question asked ungrounded and grounded, model choice with evidence in Arena Mode, big deliverables in the canvas, and where deep research lives.
2 min read
Episode 1 toured the workspace; this episode moves into the room your team will actually live in. Its core beat is a controlled experiment: the same onboarding question asked twice — once with no context, once with the Q2 support review attached — so you can watch a fluent answer and a grounded answer stop being the same thing.
What the episode shows
| At | Scene |
|---|---|
| 0:13 | The composer: agent, model, and context are three separate choices |
| 0:28 | Asking without context — fluent, confident, generic |
| 0:45 | The same topic grounded: the Q2 review attached, sources named |
| 1:02 | The one habit that matters: fluent is not evidence, sources are |
| 1:22 | The model picker and what Auto decides for you |
| 1:41 | Arena Mode: one prompt, two models, a recorded verdict |
| 2:05 | The canvas: a brief lands as a file, not a wall of chat text |
| 2:24 | Deep research: the Researcher agent and where it lives |
Where to go next
Chat basics covers the composer in reference depth; attachments explains what riding along as context really means. For the model side, read models and Arena Mode; for the workbench, the canvas pane; and for research with cited sources, deep research.