Skills
Skills define how specialized work is performed.
The Digital Employee owns the job. The skill provides the method required for the current task.
A skill can contain
- purpose
- required inputs
- process or methodology
- decision criteria
- output structure
- quality checks
- escalation conditions
- approved tool usage
Load skills when needed
A Digital Employee should not carry every possible method in permanent context. The execution environment loads the relevant skill when the current job requires it.
This keeps personas smaller, reduces duplicated instructions, and lets the same method be reused or improved independently of the employee definition.
Token-efficient execution
AI Teams use progressive context loading:
- Load the small team and job context.
- Load the required skill.
- Retrieve only the customer or business context needed for the task.
- Invoke approved tools only when required.
- Produce a structured artifact for the current stage.
- Pass the artifact, rather than the entire working conversation, to downstream work.
The Customer AI Control Plane can tell the workforce where authorized context is located and which sources are authoritative without placing the underlying information into every prompt.
Next: AI Team Orchestration coordinates when these jobs and skills run.