A product organized around an AI team naturally develops specialist entry points: employees, projects, tasks, materials, and business work. More complete specialization can make the first request harder: who should receive it, and does it already require a task?
The squirrel provides a relationship at that stage. People can begin by describing the problem. Short questions stay close to the current work; requests that need sustained execution become independent work. The team product retains a stable, recognizable counterpart.
Its design therefore concerns both a character and a working relationship. The small window, full main chat, and background task cannot be three merely similar interfaces. They need to explain how one discussion continues and when responsibility for execution has actually been assigned.
Change the window without starting another discussion
Suppose you have just discussed the outline of a talk with the main assistant. While checking project material, you want to add, “Start with the problem we encountered, then explain the method.” If the page assistant has a separate history, you have to guess whether it knows what you just decided.
The small window and full main conversation in Teloa use the same conversation. Short replies can be read beside the page; long content and the full process can use a larger view. Switching changes the space while preserving the earlier discussion.
When you leave the page and return, reopening the assistant brings back the same main conversation. You can carry on from the last exchange without working out which identically named window remembers it.
Sharing the main chat answers where the previous discussion belongs. It does not give the assistant automatic knowledge of everything on screen. We deliberately keep those concerns distinct: the window can sit beside a project, while its input basis remains explicit. Moving a window should not make the source of an answer inexplicable.
The character makes the entry recognizable; the shared conversation makes the discussion continue. Connecting a floating character to existing discussion and subsequent work gives it a reason to remain in the working environment.
This also shapes the small window's information hierarchy. Brief questions can be answered in place; longer work needs access to its full process. Putting all task management inside the window would obscure the work it was intended to stay close to.
A quick question and an independent assignment
Two requests call for different ways of working:
- A quick question: “What does this paragraph mean?” suits a direct answer.
- An independent assignment: “Prepare an explanation from this material; I'll read it later” needs a record of the goal, process, and deliverable.
Treating both as just a message would let long work crowd the conversation and make it difficult to follow separately.
Community currently supports starting independent general-purpose work from the main conversation or the squirrel. After explicitly assigning it, the user handles confirmation in the full main conversation. Only after approval does the system create and start a separate task, using the general assistant without requiring an employee to be created.
Continue the talk preparation above. You can first compare two openings in the main chat and settle on “problem, then method.” When you need a draft, explicitly assign “Prepare a ten-minute talk from the material I provided.” The small window currently leaves tool confirmation to the full main conversation, where you review and approve that execution.
The main chat then retains how the request developed, the independent task carries responsibility for the draft, its execution conversation handles the work, and the deliverable provides content to obtain and review later. You can discuss another problem without letting a long assignment occupy the whole conversation. Producing one talk also does not require creating a lasting presentation role.
This lets the effort of starting collaboration grow with the work. First clarify a problem; review the goal and execution when assigning responsibility; maintain a specialist role when the work recurs. Switching from the small window to the full chat is the current confirmation arrangement. What needs to carry forward is the requirement already developed through discussion.
What the main assistant and professional employees retain
The main assistant provides an ongoing place to talk; a professional employee works around a category of responsibility. A research employee retains source standards, a content employee retains writing methods, and the main conversation receives goals that the user raises informally and develops over time.
Both approaches can coexist. Users who know who should handle something can go directly to that role. Those who first want to clarify a problem can stay in the main conversation. The product need not route every assignment through a single intermediary or disguise every independent assignment as a new employee.
Hiding the character gives back screen space
A readily accessible assistant also needs to allow people to stop looking at it. When reading a long document or checking a table, they can move, collapse, or hide the squirrel. Display preferences and actual execution are maintained separately.
That distinction affects trust. Hiding the character should free space; stopping a task changes execution. Showing the entry point again should not reassign the work. Understanding these consequences makes it easier to keep the assistant in the working environment.
People can begin by explaining a problem to a familiar assistant or go directly to a specialist who knows the work. The squirrel deserves to remain readily accessible because starting a collaboration should not require understanding the whole team. When the conversation becomes a serious assignment, the requirements already developed have somewhere to continue.