1.4: Agents for Parallel Work

• Time to Complete: 25 minutes
• Prerequisites: Module 1.3 (File operations and @ references)

Start this module in Claude Code: Run /start-1-4 to kick off the interactive experience.

📖 Overview

Module 1.4 introduces agents - independent instances of Claude that run in parallel. Instead of processing 10 meeting notes sequentially (50 minutes), spin up 10 agents to work simultaneously (5 minutes).

Key takeaway: Agents transform Claude Code from a single assistant into a scalable team. This isn’t just faster - it unlocks entirely new workflows.

🤯 What Are Agents?

The Core Concept

Agents are independent instances of Claude that run in parallel.

Think of it like cloning yourself to work on multiple tasks simultaneously:

Sequential work (without agents):

Task 1 (5 min) → Task 2 (5 min) → Task 3 (5 min) → ... → Task 10 (5 min)
Total time: 50 minutes

Parallel work (with agents):

Task 1 (5 min) ┐
Task 2 (5 min) │
Task 3 (5 min) ├─→ All happen simultaneously
...            │
Task 10 (5 min)┘
Total time: 5 minutes

Agent Capabilities

🎯 When to Use Agents

Decision Framework

Use Agents For:

• Batch processing: Same type of work on many items
• Parallel research: Research multiple entities independently
• Multi-source analysis: Different analysis on different sources
• Time-sensitive work: Need results in hours, not days

Skip Agents For:

• Single tasks (just prompt Claude directly)
• Sequential work (tasks that depend on each other)
• Quick tasks (simple requests taking seconds)
• Context-dependent work (needs to build on previous work)

🔧 How to Prompt for Agents

Pattern 1: Explicit Agent Request

Create 10 agents to process the 10 meeting notes in /meetings.
Each agent should extract:
- Key decisions
- Action items
- Blockers
- Next steps

Combine all results into @weekly-summary.md

Pattern 2: Let Claude Decide

I have 15 competitor websites to research for features and pricing.
Work in parallel to get this done quickly.

Pattern 3: Specialized Agents

Launch 3 specialized agents:
- Agent 1: Analyze all interview transcripts in /interviews
- Agent 2: Process survey data in @survey-results.csv
- Agent 3: Review support tickets in /support

Synthesize findings into @user-research-insights.md

💼 Real-World Examples

Example 1: Meeting Processing

Scenario: Monday morning, 10 meeting transcripts, standup in 1 hour.

Prompt:

I have 10 meeting transcripts in /meetings/last-week/.

Create 10 agents to process each meeting simultaneously.

For each meeting, extract:
- Key decisions made
- Action items (with owners and due dates)
- Blockers or risks raised

Synthesize ALL findings into @monday-standup-prep.md with:
- Section 1: Critical blockers
- Section 2: Key decisions
- Section 3: Action items by team member

Result: 5 minutes vs 60 minutes sequential work.

Example 2: Competitive Research

Scenario: CEO needs competitive analysis on 5 competitors by end of day.

Prompt:

Launch 5 agents to research these competitors simultaneously:
- Asana, Linear, Monday.com, ClickUp, Notion

For each competitor, research:
- Product features and capabilities
- Pricing tiers and packaging
- Target market and positioning
- Recent updates (last 6 months)
- User reviews and sentiment

Create individual files: @competitor-[name]-research.md

Then synthesize into @competitive-analysis.md with:
- Feature comparison matrix
- Pricing comparison
- Positioning map
- Gaps and opportunities

Result: 1 hour vs 5.5 hours sequential work.

🎨 Agent Orchestration Patterns

Pattern 1: Fan-Out Processing

Use for: Multiple similar tasks, need all results combined

Process all 15 files in /meetings/ with 15 agents.
Extract action items from each.
Combine into single action item list grouped by owner.

Pattern 2: Specialized Roles

Use for: Different types of analysis on different sources

Launch 3 specialized agents:
- Agent 1 (Interview Expert): Analyze qualitative data
- Agent 2 (Data Analyst): Process survey numbers
- Agent 3 (Support Analyst): Categorize ticket themes

Synthesize into comprehensive user research report.

Pattern 3: Parallel Research

Use for: Research multiple entities independently

Research these 5 competitors in parallel.
Each agent creates detailed profile.
Synthesize into comparison matrix.

Pattern 4: Batch Document Generation

Use for: Create multiple similar documents

From @prd.md, generate 20 user stories with acceptance criteria.
Use agents to create stories in parallel.
Ensure consistency in format and quality.

Pattern 5: Validation Pipeline

Use for: Review multiple documents for quality

Review all PRDs in /docs/prds/ against our quality checklist.
Each agent reviews one PRD.
Create summary report of issues and recommendations.

💡 Best Practices

Do:

• Use for time-sensitive batch work - When you need results fast
• Let Claude decide agent count - “Process these files in parallel” works great
• Provide clear individual task instructions - Each agent needs to know what to do
• Request synthesis - Tell Claude to combine agent results meaningfully
• Specify output format - Standardize agent outputs for easier synthesis

Don’t:

• Don’t use for single tasks - Just prompt Claude directly
• Don’t create agents that need to communicate - They work independently
• Don’t use for sequential work - Tasks with dependencies should be done in order
• Don’t forget synthesis - Raw agent outputs need combining
• Don’t over-agent small tasks - If it takes 30 seconds, don’t use agents

🐛 Troubleshooting

Agents aren’t working in parallel?

• Say explicitly: “Create agents to work in parallel” or “Simultaneously process”
• Use phrases like “at the same time”

Agent results are inconsistent?

• Provide clear template:

  Each agent should output markdown with these sections:
  ## Summary
  ## Key Findings
  ## Recommendations

Too many agents created?

• Specify exact count: “Create exactly 5 agents”
• Or provide constraint: “Use as many agents as needed, max 10”

No synthesis after agents finish?

• Always end with: “After all agents complete, synthesize results into @summary.md”

One agent failed?

• Check error message for specific issue
• Ensure all referenced files exist
• Verify instructions are clear

🚀 What’s Next?

You now understand ad-hoc agents - temporary agents created on-the-fly for specific parallel tasks.

Module 1.5 introduces custom sub-agents - permanent team members with personalities, specializations, and visual identities (emojis and colors!). Instead of “act like an engineer” every time, you’ll have a 👨‍💻 Engineer sub-agent you can call anytime.

The difference:

• Module 1.4 (ad-hoc agents): Temporary clones for parallel processing
• Module 1.5 (custom sub-agents): Permanent specialized team members

Interactive track: Type /start-1-5

About This Course

Created by Carl Vellotti. If you have any feedback about this module or the course overall, message me! I’m building a newsletter and community for PM builders, check out The Full Stack PM.

Source Repository: github.com/carlvellotti/claude-code-pm-course