The Swarm

When Single Agents Are Not Enough

I asked a single researcher agent to evaluate a product launch — it took 25 minutes and covered 2 angles. Then I deployed 5 agents in parallel on the same question. They returned in 8 minutes covering research, copy, planning, technical risk, and competitor analysis. 3x faster, 5x more perspective. That is the swarm.

For your business as a solo operator, the swarm lets you simulate having a full team even if you are working alone. Multiple AI agents analyze your problem from different angles and synthesize their findings.

Deployment Modes

Not every task needs a full swarm. Match the deployment mode to the task scope.

| Mode        | Agents | When to Use                           |
|-------------|--------|---------------------------------------|
| Single      | 1      | Clear task, one domain                |
| Hive-Micro  | 2-5    | Small task, 1-2 day scope             |
| Hive-Sprint | 5-15   | Feature or campaign, 1-2 week scope   |
| Hive-Full   | 15+    | Venture-level work, multi-week scope  |

Decision: How many DOMAINS does this task touch?
- 1 domain → Single agent
- 2-3 domains → Hive-Micro
- 4+ domains → Hive-Sprint or Hive-Full

How to Run a Swarm

The key insight: send multiple agent spawns in a single message. This triggers parallel execution — all agents work simultaneously instead of sequentially.

SWARM CONTEXT:
- Task: [Describe the task or decision]
- Business: My Business (business)
- Constraints: [Any constraints]

AGENT 1 (researcher):
Research [specific angle]. Report findings with sources.

AGENT 2 (task-planner):
Break down [the task] into phases. Include acceptance criteria.

AGENT 3 (code-debugger):
Review [codebase/system] for potential issues. Report risks.

---
After all agents return, SYNTHESIZE:
1. Group findings by theme
2. Note where agents agree (high confidence)
3. Note where agents disagree (needs resolution)
4. Present unified recommendation

Synthesis Protocol — Combining Results

The swarm builds on everything you have set up so far. Each agent in the swarm was defined in Chapter 3: Building Your First Agents, each receives context injection from your Life Repository (Chapter 2), and the rules from Chapter 4: The Rules System govern how agents behave even in parallel.

The most important part of a swarm is what happens after agents return. The Architect (main AI) synthesizes all outputs into a unified recommendation.

1. 1Group findings by theme — what topics did multiple agents address?
2. 2Identify consensus — where did agents agree? These are high-confidence findings.
3. 3Surface conflicts — where did agents disagree? These are the interesting parts.
4. 4Do not force consensus — some disagreements are the insight. Present both sides with evidence.
5. 5Prioritize by impact — rank findings by how much they affect the decision.
6. 6Present a clear recommendation — one sentence, with caveats noted.