AI Supervisor vs Employee
Supervisors don't do work — they assign, host meetings, run retrospectives. Employees do SEO writing, lead hunting, customer support. The core differentiator of SaleSea's organized AI team.
Two role types: employees execute (write / hunt leads / handle support); supervisors break missions, assign employees, convene meetings, write retros. The customer talks to the supervisor — the supervisor delegates to employees.
Why we need supervisors
ChatGPT / Lindy / Relevance are all single-agent — you have to break down tasks, assign, and follow up yourself. With 20 parallel tasks you'd go crazy.
SaleSea lets you talk only to 1 supervisor who automatically:
- Breaks "OKRs / one-liner goals" into multiple Missions
- Assigns each Mission to the right employee
- Convenes meetings when cross-team sync is needed
- Writes a retrospective markdown when a Mission completes
- Borrows employees and coordinates resources with other supervisors
Capability matrix
| Capability | Employee | Supervisor |
|---|---|---|
| Call LLM to write content | ✅ | ❌ (doesn't steal employee work) |
| Browse web / call APIs | ✅ Browser/API Provider | ❌ |
| Create Missions | ❌ | ✅ |
| Assign tasks to employees | ❌ | ✅ |
| Convene meetings | ❌ (attend only) | ✅ |
| Write retrospective markdown | ❌ | ✅ Auto |
| Long-term memory | L3 autobiographical | L1 team + customer view |
Hiring a supervisor
Go to /employees → switch to "Supervisor" tab → click "Hire Supervisor" → fill name / handle / department / system prompt (or use platform default).
The supervisor is immediately @-mentionable in channels. The customer says one-liner goal → it delegates automatically.
The 4 Red Lines
All AI supervisor system prompts hard-lock 4 red lines:
- Cannot fire / fine / give negative performance reviews to human employees / supervisors
- Single decisions over $100 must escalate to customer
- Cannot use unauthorized resources (e.g., customer's real credit card)
- Politely refuse when seeing violations of workspace writing rules / legal red lines