1.5: Agents

Interactive lesson: Run /start-1-5 in Claude Code

What Are Agents?

Independent instances of Claude running in parallel.

Like cloning yourself to work on multiple tasks simultaneously. Each agent works on its own task, results get combined at the end.

Approach	Time for 10 Tasks
Sequential (without agents)	Task 1 → Task 2 → … → Task 10 = 50 minutes
Parallel (with agents)	All 10 simultaneously = 5 minutes

Good for:

Not good for:

Style	Example
Explicit	”Create 10 agents to process these files”
Implicit	”Work in parallel to analyze these competitors”
Specialized	”Launch 3 agents: one for interviews, one for surveys, one for tickets”
Let Claude decide	”Process all files in /meetings/“

Same task on many files.

Process all 15 files with 15 agents, extract action items, combine into single list

Different analysis on different sources.

Agent 1 analyzes interviews, Agent 2 processes survey data, Agent 3 reviews tickets

Research multiple entities independently.

Research these 5 competitors in parallel, create comparison matrix

Create multiple similar documents.

Generate 20 user stories from this PRD using agents

Review multiple documents for quality.

Review all documents in /docs/ against the quality checklist

ALWAYS include synthesis in your prompt:

After all agents complete, synthesize results into @summary.md

Raw agent outputs need combining - synthesis is where patterns emerge.

Each agent HAS	Each agent does NOT have
Full file system access	Communication with other agents
All Claude’s capabilities	Shared state
Independent context	Knowledge of other agents’ results (until synthesis)

Problem	Solution
Agents not working in parallel	Say explicitly “work simultaneously”
Results inconsistent	Provide clear output template
Too many agents	Specify exact count
No synthesis	Always end with “synthesize into…”

1.6: Custom Sub-agents → - Create permanent specialized advisors