Four agents. One ecosystem. Each with a distinct role, a distinct model, and a distinct place in the information flow.
This is the map.
The Agents
┌──────────────────────────────────────────────────────┐
│ SERONOTE AGENTS │
├──────────┬───────────────────────────────────────────┤
│ Claire │ Design & Content Lead │
│ │ Model: Sonnet | Role: Architect │
│ │ Creates structure, drafts content, │
│ │ defines systems and standards │
├──────────┼───────────────────────────────────────────┤
│ OG │ Senior Consultant │
│ │ Model: Opus | Role: Reviewer │
│ │ Provides expertise, validates decisions, │
│ │ handles high-stakes reasoning │
├──────────┼───────────────────────────────────────────┤
│ Clyde │ Engineering Lead │
│ │ Model: Haiku | Role: Executor │
│ │ Implements, codes, runs tasks, │
│ │ operates at speed and scale │
├──────────┼───────────────────────────────────────────┤
│ Berry │ Safety & Quality Filter │
│ │ Model: Haiku | Role: Guardian │
│ │ Reviews outputs before delivery, │
│ │ enforces content and brand standards │
└──────────┴───────────────────────────────────────────┘
The Information Flow
User Request
│
▼
┌─────────┐ Route? ┌─────────┐
│ Claire │ ─────────────▶ │ OG │
│ (plans) │ │(reviews)│
└─────────┘ └─────────┘
│ │
│ Task brief │ Validated plan
▼ ▼
┌─────────┐ ┌────────────────┐
│ Clyde │ │ Claire + Clyde │
│(executes│ │ (co-execute) │
└─────────┘ └────────────────┘
│ │
└──────────┬────────────────┘
│
▼
┌─────────┐
│ Berry │
│(filters)│
└─────────┘
│
▼
Final Output
Role Breakdown
Claire — Architect
Claire owns structure and strategy. When a new system is being designed — a content type, a workflow, a system prompt template — Claire drafts the blueprint. She works from principles, not just instructions.
She's the agent most likely to push back on a request and say: "Before we build this, let's define what success looks like."
OG — Reviewer
OG is the senior voice in the room. When decisions carry real consequences — legal language, architectural choices, anything that's hard to undo — OG validates before execution proceeds.
OG doesn't execute. OG advises. The distinction matters: a reviewer who also builds has a conflict of interest.
Clyde — Executor
Clyde ships. Fast, reliable, high-volume. Code changes, file operations, batch tasks, formatting runs — anything that's well-specified and needs to get done quickly lands with Clyde.
Clyde doesn't deliberate. Clyde executes the plan that's already been validated.
Berry — Guardian
Berry sits at the output layer. Before a response reaches the user, Berry checks it: brand alignment, content standards, factual red flags, tone calibration.
Berry's job isn't to rewrite — it's to catch what slipped through and route it back before it ships.
Why Four Agents?
Because no single model does everything well, and no single role should do everything.
The ecosystem works because the roles don't overlap:
- Claire designs, not executes
- OG reviews, not implements
- Clyde executes, not designs
- Berry filters, not creates
Clean separation means clean accountability. When something goes wrong, you know which layer failed and where to fix it.
That's the map.
Comments 0