Expertise: Teach the Robot the Job
Expertise answers the practical question: how does the robot actually do the job? Once the robot can converse reliably, use this area to add knowledge, workflow logic, role strategy, and environment context so it can complete one real task.
Start with one minimum viable workflow
Pick one real task first, for example:
- greet visitors and answer common questions
- patrol a fixed route and report anomalies
- escort a guest from the front desk to a meeting room
Once the task is explicit, decide which assets are needed. Do not work backwards by filling every page in the product.
Minimum viable setup
| Capability | When you need it | Purpose |
|---|---|---|
| Knowledge documents | The robot must answer from source material | Provide RAG content |
| Scripts / role | The service flow must stay consistent | Standardize behavior and wording |
| Flows | Reactions must follow explicit triggers | Chain actions into one workflow |
| Scenes | The robot moves or depends on spatial context | Manage maps, routes, and zones |
Recommended Setup Order
- Define the target job, success criteria, and hard boundaries
- Add the minimum knowledge, scripts, or role setup needed to answer correctly
- Add flows only if the job needs explicit trigger-based behavior
- Bind scenes, routes, and zones if the robot moves in the environment
Validation before handoff
- Ask 5 frequent user questions and check the knowledge hit quality
- Run one full business flow and confirm the robot does not skip steps
- If movement is involved, test one start-to-finish navigation path
- If behavior is unstable, narrow the task before adding more configuration
Boundary with Intelligence
- Intelligence: model, persona, perception, memory, voice, expressions, actions — who he is
- Expertise: knowledge, scripts, role, flows, scenes — what he does