Why we did not build a nicer chat box
The chat box is the default shape of every AI product, and also its ceiling: context is used once and thrown away. Notes on the trade-off we made.
The problem with chat is not the interface
Chat is pleasant: type, read, clear. The problem is its data structure — a conversation is disposable, and context dies with it.
So you develop a habit: before it can help, you paste the material in again.
We moved the centre
Not the conversation, but the material library. Chats, pages, files and images come in as raw stock that can be searched, questioned, and read or written by an Agent. Conversation is one way to work on material — not the place it lives.
Three things change:
- Context accumulates. What you collected today still works next month.
- Conclusions land somewhere. They go back into notes instead of scrollback.
- Citations stay traceable. Answers point at paragraphs, not at model memory.
What it costs
- A steeper start: there is one more concept, "material", to understand.
- It is bad at small talk. It is not a companion product and does not role-play.
- It rewards a little collecting discipline. An empty library cannot answer well.
We accept those costs. We were never solving "how do we chat more smoothly". We were solving "how does AI actually finish the work".
Chat is the process. Material is the asset.
Keep reading
ProductConnected in five minutes: extension and skill
Two paths in: the extension handles capture, the skill handles use. Which one you need depends on where you are stuck.
Read post
CreateFrom material to a deck you will actually use
Generating slides is easy. Wanting to use them afterwards is hard. Most of our effort went into the second half: editable, exportable, structurally sound.
Read post
AgentLet the Agent read and write your material
The ceiling of a chat window is that you keep re-feeding it context. With the skill installed, the Agent reads your library directly and writes back to your notes.
Read post