Module 3 The Digital Org Chart
Status: Complete · How one operator conducts a team of AI personas.
An SAE does not manage people — they conduct computational personas. The operator sits at the apex of a digital org chart, routing context across specialized AI sub‑agents.
Text version
┌──────────────────────────────┐
│ Human SAE Conductor (You) │
└──────────────┬───────────────┘
│
┌───────────────────────┼───────────────────────┐
▼ ▼ ▼
┌─────────────────┐ ┌──────────────────┐ ┌─────────────────┐
│ The Visualizer │ │ The Engineer │ │ The Guardian │
│ (Generative UI) │ │ (Core IDE Engine)│ │ (Autonomous QA) │
└─────────────────┘ └──────────────────┘ └─────────────────┘
The supreme orchestrator. Responsibilities are exclusively high‑leverage:
- Auditing overall system topology
- Reviewing AI output for subtle structural drift
- Verifying edge‑case conditions
- Making strategic product calls based on business rules
You guide the context window; the machines handle the execution bytes.
The persona focused on UI design, component structure, design tokens, and interaction mechanics. Feed it product requirements; it compiles clean, semantic component code — eliminating manual styling churn.
Typical tools: v0, Shadcn generators, design‑token systems. See Approved Zero‑Ops Tooling List.
The core codebase engine with direct repository access (e.g. Cursor, Claude Code, terminal orchestration agents). It reads the system‑wide .cursorrules, parses repo files, handles migrations, and writes application logic.
The safety net whose sole objective is to break what the Engineer builds. Running continuously in a containerized sandbox, it reviews changes, writes end‑to‑end integration tests, simulates hostile payloads, and blocks regressions from reaching the delivery layer.
| Persona | Receives from Conductor | Produces | Hands to |
|---|---|---|---|
| Visualizer | Product intent, design constraints | Semantic UI components | Engineer (for wiring) |
| Engineer | Spec + repo state | Schema, routes, logic, migrations | Guardian (for attack) |
| Guardian | The generated slice | Tests, adversarial findings, regressions | Conductor (for ruling) |
The Conductor closes every loop — the personas never sign off on their own work.
Reality check. These "personas" may be three different tools or one model wearing three hats across three prompts — the separation of concerns matters more than the tool count. The critical rule: the agent that wrote the code must not be the sole judge of whether it works. Independent adversarial review (a fresh context tasked to refute, not confirm) catches what self‑review rationalizes away. If you collapse the Guardian into the Engineer, you lose the framework's only built‑in check on hallucinated correctness.
▶ Putting it into practice: persona prompt templates in Agent Setup & Configuration · the "one persona, one job" and adversarial‑review moves in Operator Techniques.