2.2 AI employees, work groups, and task teams
You are responsible for goals, resources, and final decisions. AI employees take on lasting responsibilities and collaborate with you in work groups. Task teams help complete the current task. Your twin (My twin, at the bottom left) keeps observations about your personal habits.
AI employees: lasting identities
An AI employee is defined by its responsibilities, business scope, skills, employee memories, knowledge, and tool permissions. Its identity remains after a task ends. A research employee and a marketing employee may use the same skill to summarize material, but sharing a skill does not give them the same data access or write permissions.
In the current product, “distilling skills” means organizing methods, configuration, and experience into reusable capabilities. It does not mean training new model weights.
Why AI employee
Agent is the underlying technical category; AI employee is a lasting work role people can understand. It has a role, responsibilities, knowledge, skills, tools, memory, and permissions, and takes assignments, collaborates, and delivers within a team. You grant access and review important results.
“Collaborating as colleagues” describes the relationship without renaming the object AI colleague. “Expert” can describe domain capabilities, but it is not the name for every role. Assistant, Expert, and Agent are not fixed stages in an ability hierarchy. Everyday interfaces should show specific job names such as Researcher or Content strategist and explain their deliverables.
Work groups: business collaboration with employees
A work group is where you assign work, discuss it, follow progress, and receive results. In a marketing work group, Alice can own content and Bob can own data analysis. You follow and review their respective work. Joining a work group does not expand any employee's permissions. See the work group guide.
Task teams and task assistants: completing the current task
A task team organizes work around the current task. Either an ordinary conversation's lead assistant or an AI employee can start one. Task assistants are its temporary members: they do not automatically become AI employees or gain additional access to materials, tools, or actions. An ordinary conversation can use a task team without creating a work group or hiring an employee first.
In the marketing example, Alice can organize a task team to gather material and review copy, while Bob organizes another to check data. Task assistants return their work to the responsible employee. Alice and Bob summarize their results and report back to the work group. A member finishing does not mean the whole task is complete.
AI employees and work groups describe lasting responsibilities and business collaboration. Task teams and task assistants describe how the current task is carried out. See the native capability guide for availability, configuration and usage in the current release.
My twin focuses on you
The twin's private memories are stored separately from an employee's business memories. Habit observations that Auto Dream saves automatically are visible only to you and can serve as reference when you chat privately with My twin; they are not instructions or authorization to act. Regular conversations, groups, and other employees do not use them automatically. See Automatic memory for the precise boundary.
Why the distinction matters
| Item | Where it belongs |
|---|---|
| The steps an employee follows to write a report | Skills and employee configuration |
| Recently confirmed experience within a business area | Employee memories with sources |
| Observations about your personal preferences | The twin's private samples |
| Sending externally or approving a budget | Your explicit decision and the corresponding authorization |
Community is designed for one human user. Having several employees in a group does not mean enterprise accounts for multiple people or organizational permissions are already available.