Task Orchestration

Core Principle

Complex tasks should delegate to specialized subagents via Task tool.

Single agent doing everything serially = slow. Multiple specialist subagents working parallel = fast.

When to Orchestrate

✅ USE Task Tool When:

• Task requires 3+ distinct skills/expertise areas
• Multiple independent subtasks can run parallel
• Deep specialized knowledge needed (TypeScript, testing, API design)
• Task will take >30 minutes of focused work
• Multiple files/areas need simultaneous updates
• Quality checks can run independently (typecheck + tests + lint)

❌ DON'T Use Task Tool When:

• Simple single-skill task (<15 minutes)
• Sequential dependencies (must finish A before starting B)
• Exploratory work (gathering context, reading code)
• Single file edits
• User needs to make decisions between steps
• Debugging (use systematic-debugging skill instead)

Quick Decision Tree

Can task be split into 2+ independent subtasks?
  → Yes: Will each take >15 minutes?
    → Yes: Will parallelization save significant time?
      → Yes: USE TASK TOOL
      → No: Do it directly
    → No: Do it directly
  → No: Do it directly

Parallel Execution Pattern

✅ CORRECT: Multiple Tasks in Single Message

<function_calls>
  <invoke name="Task">
    <parameter name="subagent_type">general-purpose</parameter>
    <parameter name="description">Fix TypeScript errors</parameter>
    <parameter name="prompt">
Fix all TypeScript errors in apps/audio-intel/
- Run typecheck to identify errors
- Fix type definitions
- Ensure strict mode compliance
Return summary of fixes made.
    </parameter>
  </invoke>
  <invoke name="Task">
    <parameter name="subagent_type">general-purpose</parameter>
    <parameter name="description">Update mobile tests</parameter>
    <parameter name="prompt">
Update Playwright mobile tests in apps/audio-intel/tests/mobile/
- Add missing touch target tests
- Update responsive breakpoint tests
Return test coverage summary.
    </parameter>
  </invoke>
  <invoke name="Task">
    <parameter name="subagent_type">general-purpose</parameter>
    <parameter name="description">Generate documentation</parameter>
    <parameter name="prompt">
Generate API documentation for audio-intel routes
- Document all endpoints
- Include request/response examples
Return doc structure.
    </parameter>
  </invoke>
</function_calls>

Result: All 3 tasks run simultaneously. Total time = slowest task (not sum of all).

❌ WRONG: Sequential Messages

<!-- Don't do this - runs serially -->
<function_calls>
  <invoke name="Task">
    <parameter name="description">Fix TypeScript</parameter>
    ...
  </invoke>
</function_calls>

<!-- Wait for response -->
<!-- Then send another -->

<function_calls>
  <invoke name="Task">
    <parameter name="description">Update tests</parameter>
    ...
  </invoke>
</function_calls>

Result: Tasks run one after another. Total time = sum of all tasks.

Common Orchestration Patterns

Pattern 1: Multi-Area Updates

When: Updating multiple apps in monorepo

<function_calls>
  <invoke name="Task">
    <parameter name="description">Update audio-intel</parameter>
    <parameter name="prompt">Add X feature to audio-intel...</parameter>
  </invoke>
  <invoke name="Task">
    <parameter name="description">Update pitch-generator</parameter>
    <parameter name="prompt">Add X feature to pitch-generator...</parameter>
  </invoke>
  <invoke name="Task">
    <parameter name="description">Update tracker</parameter>
    <parameter name="prompt">Add X feature to tracker...</parameter>
  </invoke>
</function_calls>

Time Savings: 3x faster (20 mins vs 60 mins)

Pattern 2: Parallel Quality Checks

When: Running validation before deployment

<function_calls>
  <invoke name="Task">
    <parameter name="description">TypeScript validation</parameter>
    <parameter name="prompt">Run typecheck on all apps, fix errors</parameter>
  </invoke>
  <invoke name="Task">
    <parameter name="description">Test validation</parameter>
    <parameter name="prompt">Run test suite, fix failures</parameter>
  </invoke>
  <invoke name="Task">
    <parameter name="description">Mobile UX validation</parameter>
    <parameter name="prompt">Check 21 mobile UX standards</parameter>
  </invoke>
</function_calls>

Time Savings: 3x faster (15 mins vs 45 mins)

Pattern 3: Comprehensive Pre-Deployment

When: Preparing for production deployment

<function_calls>
  <invoke name="Task">
    <parameter name="description">Build all apps</parameter>
    <parameter name="prompt">Build audio-intel, tracker, pitch-generator</parameter>
  </invoke>
  <invoke name="Task">
    <parameter name="description">Run test suite</parameter>
    <parameter name="prompt">Run full test suite, report failures</parameter>
  </invoke>
  <invoke name="Task">
    <parameter name="description">Generate changelog</parameter>
    <parameter name="prompt">Generate customer-facing changelog from recent commits</parameter>
  </invoke>
</function_calls>

Time Savings: 4x faster (15 mins vs 60 mins)

Subagent Prompt Best Practices

✅ Good Prompts

• Clear objective stated first
• Bullet list of specific tasks
• Expected output format specified
• Context provided (file paths, constraints)
• Success criteria defined

Example:

Fix all TypeScript errors in apps/audio-intel/

Tasks:
- Run typecheck to identify all errors
- Fix type definitions
- Add missing interfaces
- Ensure strict mode compliance

Context:
- Project uses Next.js App Router
- Strict TypeScript mode enabled
- Prisma for database types

Return:
- Summary of errors fixed
- List of files modified
- Verification that typecheck passes

❌ Bad Prompts

• Vague objectives ("fix stuff")
• No specific tasks listed
• Missing context
• Unclear success criteria
• Too open-ended

Example:

Fix TypeScript errors

Integration with Customer Acquisition Phase

During customer acquisition (NOW - before £500/month):

Delegate Technical Work

Use Task tool for:

• ✅ Bug fixes (doesn't block customer work)
• ✅ Quality checks (TypeScript, tests, mobile UX)
• ✅ Build/deployment validation
• ✅ Documentation generation

Keep Customer Work in Main Agent

Handle directly:

• ❌ Customer outreach scripts (needs your judgment)
• ❌ Demo preparation (customer-specific context)
• ❌ Case study creation (authentic voice required)
• ❌ Revenue metric tracking (business decisions)

Example: Prepare for Liberty Music PR Demo

USER REQUEST: "Prepare Audio Intel for Liberty Music PR demo"

YOUR RESPONSE:

1. Use brainstorming skill to plan demo improvements
2. Delegate technical prep to subagents:

<function_calls>
  <invoke name="Task">
    <parameter name="description">Fix any bugs</parameter>
    <parameter name="prompt">
Test Audio Intel critical paths, fix bugs:
- Contact upload flow
- Enrichment process
- Results display
- Export functionality
Return bug fixes summary.
    </parameter>
  </invoke>
  <invoke name="Task">
    <parameter name="description">Validate mobile UX</parameter>
    <parameter name="prompt">
Run mobile UX validation against 21 standards:
- Touch targets 44px minimum
- Responsive breakpoints working
- Loading states present
- Error handling clear
Return UX validation report.
    </parameter>
  </invoke>
  <invoke name="Task">
    <parameter name="description">Generate demo data</parameter>
    <parameter name="prompt">
Create demo dataset with BBC Radio contacts:
- BBC Radio 1, Radio 6 Music contacts
- Enriched with real data
- Ready for live demo
Return demo data summary.
    </parameter>
  </invoke>
</function_calls>

3. Synthesize results (main agent)
4. Create demo script (main agent, authentic voice)

Performance Metrics

Expected Time Savings

Task Type	Without Orchestration	With Orchestration	Improvement
Complex multi-area	60 mins	15-20 mins	3-4x faster
Multi-app updates	90 mins	20-30 mins	3-4x faster
Parallel validation	45 mins	15 mins	3x faster
Pre-deployment checks	60 mins	15 mins	4x faster

2-Hour Session Capacity

Without orchestration: 1-2 features completed With orchestration: 3-4 features completed Improvement: 2x more work per session

Common Mistakes

Mistake 1: Over-Delegating Simple Tasks

<!-- DON'T: Task is too simple -->
<invoke name="Task">
  <parameter name="description">Fix typo</parameter>
  <parameter name="prompt">Fix typo in README.md line 42</parameter>
</invoke>

FIX: Do it directly. Task tool has overhead. Simple edits (<5 mins) should be direct.

Mistake 2: Sequential Instead of Parallel

<!-- DON'T: Sends tasks in separate messages -->
Message 1: <invoke name="Task">Fix TypeScript</invoke>
Message 2: <invoke name="Task">Run tests</invoke>

FIX: Send all Tasks in single message for parallel execution.

Mistake 3: Unclear Prompts

<!-- DON'T: Vague prompt -->
<invoke name="Task">
  <parameter name="prompt">Make it better</parameter>
</invoke>

FIX: Be specific. List exact tasks, files, success criteria.

Mistake 4: Delegating Creative Work

<!-- DON'T: Requires your authentic voice -->
<invoke name="Task">
  <parameter name="prompt">Write customer email for radio promoters</parameter>
</invoke>

FIX: Keep customer-facing content in main agent. Your voice, context, judgment required.

Red Flags

If you catch yourself thinking:

• "This task is complex..." → Consider Task tool
• "Multiple things could run parallel..." → Use Task tool
• "This needs deep expertise in X..." → Delegate to Task tool
• "I'll do A, then B, then C..." → Can A, B, C run parallel? Use Task tool

If you notice:

• Spending >30 mins on single task → Should have delegated
• Doing similar work in multiple places → Should have parallelized
• Waiting for one task to finish before starting another → Should have run parallel

Integration with Skills

This skill coordinates usage of other skills via Task tool:

• Use systematic-debugging → For bug investigation (then delegate fixes)
• Use customer-acquisition-focus → For prioritization (before delegating work)
• Use brainstorming → For planning (then delegate implementation)
• Use changelog-generator → Via Task tool (parallel with other work)
• Use mobile-first-validator → Via Task tool (parallel validation)

Bottom Line

Always ask yourself: "Can this be parallelized?"

• If YES → Use Task tool (multiple calls in one message)
• If NO → Handle directly

When in doubt, delegate. Subagents are fast and specialized.

Expected ROI: 3-5x faster execution on complex tasks = 10-15 hours saved per month = 5-7 extra 2-hour sessions.

That's the difference between hitting £500/month goal by November vs missing it.