Hub-and-Spoke is a coordination pattern where a central coordinator (hub) delegates tasks to multiple specialist agents (spokes), waits for all responses, then synthesizes results into a unified recommendation.
It's a delegation pattern where the hub orchestrates specialists, each with deep expertise in their domain.
Each specialist works independently on their piece, but all report back to the hub.
This pattern is comprehensive and expert-driven. The hub ensures quality by consulting specialists.
Hub (Coordinator) → Specialist A + Specialist B + Specialist C + Specialist D → Hub synthesizes → User
When to use: Results must be INTEGRATED, hub needs expertise to synthesize, conflicts must be resolved, outputs must work together coherently.
When to Use It
Use Hub-and-Spoke when:
Examples:
When NOT to Use It
Don't use Hub-and-Spoke when:
Wrong use cases:
Frontstage: What the User Sees
The user sees a single comprehensive recommendation that integrates multiple specialist perspectives.
Example: Financial Portfolio Planning
What user experiences:
Backstage: What the Agents Do
Behind the scenes, a hub coordinator delegates to four specialist agents, waits for all expert opinions, then synthesizes into coherent strategy.
Hub Agent (Portfolio Coordinator)
Role: Orchestrates/delegates specialists and synthesizes recommendations
Specialist A: Stocks Agent
Domain expertise: Equity markets, growth stocks, dividends
Input from hub: User age 45, risk tolerance medium, 20-year time horizon, ÂŁ200K portfolio, retirement goal
Actions:
Duration: 8 seconds
Returns to hub: 40% stocks allocation, 9-11% expected annual return, moderate risk
Status: COMPLETE
Specialist B: Bonds Agent
Domain expertise: Fixed income, government bonds, corporate bonds
Input from hub: User age 45, risk tolerance medium, 20-year time horizon, ÂŁ200K portfolio, retirement goal
Actions:
Duration: 6 seconds
Returns to hub: 30% bonds allocation, 4-5% expected annual return, low risk
Status: COMPLETE
Specialist C: Real Estate Agent
Domain expertise: REITs, property investment, real estate markets
Input from hub: User age 45, risk tolerance medium, 20-year time horizon, ÂŁ200K portfolio, retirement goal
Actions:
Duration: 7 seconds
Returns to hub: 20% real estate allocation, 6-8% expected annual return, moderate risk
Status: COMPLETE
Specialist D: Tax Agent
Domain expertise: Tax optimization, retirement accounts, tax-loss harvesting
Input from hub: User age 45, risk tolerance medium, 20-year time horizon, ÂŁ200K portfolio, retirement goal, UK location
Actions:
Duration: 5 seconds
Returns to hub: 10% cash allocation, ISA/SIPP strategy, ÂŁ2K annual tax savings
Status: COMPLETE
Hub analysis
Total allocation: 100%
Risk balance: 40% growth, 30% stability, 20% diversification, 10% liquidity equals Medium risk
Conflict resolution: No conflicts (all allocations compatible)
Tax optimization: Integrated throughout
Expected return: Weighted average 7-9% annually
Hub synthesizes unified recommendation
Portfolio Strategy:
Tax Strategy: Maximize ISA, SIPP contributions, projected ÂŁ2K annual savings
Risk Level: Medium
Expected Return: 7-9% annually
Rebalancing: Quarterly
Each specialist's expertise contributed to this balanced approach.
Critical Coordination Data
Hub dispatches to all specialists:
All specialists return to hub:
Hub waits for slowest (Stocks: 8 seconds), then synthesizes:
Service Blueprint
The service blueprint shows frontstage (what users see) above the line of visibility, and backstage (hub-and-spoke coordination) below the line.
Key elements:
Accessibility Considerations
1. Specialist Contributions Made Visible
The challenge:
Screen reader users need to understand WHERE recommendations come from. Which specialist said what?
Design solution:
Make each specialist's contribution explicit and navigable.
Use expandable sections so screen reader users can navigate through each specialist's reasoning independently.
2. Hub Synthesis Explained
The challenge:
How do you explain to users (especially those with cognitive disabilities) how the hub combined specialist opinions?
Design solution:
Show synthesis logic transparently.
3. User Can Challenge Specialists
The challenge:
What if user disagrees with one specialist's recommendation? Can they adjust?
Design solution:
Allow users to interact with individual specialist recommendations.
Each specialist recommendation is individually adjustable
Changes announce to screen reader
Risk level recalculates visibly
User maintains control
4. Specialist Conflicts Made Clear
The challenge:
What if two specialists disagree? How do you communicate conflict resolution accessibly?
Design solution:
Show conflicts and how hub resolved them.
Use alert role for conflict notifications
Clearly explain resolution logic
Allow user to see original specialist recommendations
Common Failure Modes
Failure 1: Specialists Contradict Each Other
Stocks Specialist recommends: 50% stocks allocation
Bonds Specialist recommends: 40% bonds allocation
Real Estate Specialist recommends: 30% real estate allocation
Tax Specialist recommends: 20% cash allocation
Total: 140% (impossible)
Hub doesn't know how to resolve.
User sees: "Error: Unable to create portfolio."
Design conflict resolution rules:
Priority order: Tax compliance first (legal requirement), then risk tolerance (user preference), then return optimization
Proportional reduction: Scale all recommendations down to total 100% while maintaining ratios
User choice: Show conflict to user, let them decide
Design transparent communication:
"Specialists recommended more than 100% total allocation.
We adjusted proportionally to fit your medium risk tolerance:
Stocks: 50% → 40% (scaled down 20%)Bonds: 40% → 30% (scaled down 25%)Real Estate: 30% → 20% (scaled down 33%)Cash: 20% → 10% (scaled down 50%)
Total: 100%
Review adjusted allocation?"
- Simulate conflicting specialist recommendations
- Verify hub resolves conflicts transparently
- Check user can see original vs. adjusted recommendations
- Confirm resolution logic is explained clearly
- Test with screen reader (conflict resolution announced)
Failure 2: Hub Can't Synthesize Specialist Outputs
Stocks Specialist returns: Complex technical analysis, 47 data points
Bonds Specialist returns: 12-page report on fixed income markets
Real Estate Specialist returns: REITs ranked by 23 criteria
Tax Specialist returns: Tax code references and optimization strategies
Hub receives: Mountain of specialist data
Hub attempts: Synthesis
Hub fails: Too complex, no clear integration logic
User sees: "Processing..." forever
Design specialist output format:
Each specialist MUST return:
- Recommendation: Clear single number or action
- Rationale: One sentence explanation
- Supporting data: Limited to 3 key points
- Risk assessment: High/Medium/Low
- After timeout, treat as failure.
Example structured output:
Stocks Specialist returns:
Recommendation: 40% allocation
Rationale: "20-year horizon supports growth strategy"
Key points: Strong historical returns, volatility acceptable at this age, diversification across sectors
Risk: Moderate
Design hub synthesis framework:
- Step 1: Collect all recommendations
- Step 2: Check total equals 100%
- Step 3: Assess combined risk level
- Step 4: Calculate weighted expected return
- Step 5: Apply tax optimization
- Step 6: Present unified recommendation
- Verify all specialists return structured outputs
- Check hub can process all output formats
- Test synthesis with conflicting data
- Confirm hub produces coherent recommendation
- Test with various specialist response combinations
Failure 3: One Specialist Fails, Hub Blocks
Stocks Specialist: Returns recommendation (8 seconds)
Bonds Specialist: Returns recommendation (6 seconds)
Real Estate Specialist: Times out (30 seconds, no response)
Tax Specialist: Returns recommendation (5 seconds)
Hub: Waiting for Real Estate Specialist...
User sees: "Consulting specialists..." forever
Design partial synthesis:
- If 3+ of 4 specialists respond: Create portfolio with available recommendations
- If 2 of 4 specialists respond: Ask user if they want partial recommendation
- If 1 or 0 specialists respond: Clear failure message
Design timeout communication:
"Real Estate Specialist taking longer than expected.
Would you like to:
- See portfolio recommendation without real estate (stocks, bonds, cash)
- Wait 30 more seconds for real estate analysis
- Cancel and try later"
Design graceful degradation:
3 specialists responded:
- Stocks: 40%
- Bonds: 30%
- Tax: 10% cash
Hub recalculates without Real Estate:
- Stocks: 50% (increased from 40%)
- Bonds: 40% (increased from 30%)
- Cash: 10% (unchanged)
- Total: 100%
User sees:
"Created portfolio based on 3 of 4 specialists. Real estate allocation unavailable. Portfolio adjusted for stocks and bonds."
- Simulate each specialist timing out
- Verify partial synthesis works
- Check user informed about missing specialist
- Confirm portfolio still coherent without one specialist
- Test with screen reader (partial results announced)
Real-World Examples
Example 1: Financial Portfolio Planning
Pattern: Hub delegates to Stocks + Bonds + Real Estate + Tax specialists
Scenario: User says "Create investment portfolio for retirement"
Hub execution:
Specialists work:
Hub synthesizes:
Requires multiple types of expertise (stocks ≠bonds ≠real estate ≠tax)
Each specialist has deep domain knowledge
Quality comes from combining expert perspectives
Hub synthesizes coherent strategy from different domains
"Consulting specialists..."
"Recommended portfolio: 40% stocks, 30% bonds, 20% real estate, 10% cash. Tax optimized for ÂŁ2K annual savings. Risk: Medium. Return: 7-9%. Approve?"
Example 2: Medical Diagnosis
Pattern: Hub delegates to Symptoms + Lab Results + Imaging + Medical History specialists
Scenario: Patient presents with chest pain
Hub execution:
Specialists work:
Hub synthesizes:
Multiple data sources required
Each specialist interprets different type of evidence
Diagnosis requires integration of all findings
Hub (diagnostic AI) synthesizes into clinical recommendation
"Analyzing symptoms, labs, imaging, and history..."
"Assessment: Findings suggest cardiac event. Immediate cardiology consultation recommended. Emergency services contacted."
Example 3: Legal Case Preparation
Pattern: Hub delegates to Case Law + Contract + Compliance + Precedent specialists
Scenario: Lawyer preparing employment discrimination case
Hub execution:
Specialists work:
Hub synthesizes:
Legal research requires multiple specialties
Each specialist searches different legal domain
Quality legal strategy integrates all research
Hub (legal AI) synthesizes coherent case strategy
"Researching case law, contracts, compliance, and precedents..."
"Case assessment: Strong claim supported by 3 statutes, contract violation, and favorable precedents. Recommended strategy: Settlement negotiation. Detailed report ready."
When implementing Hub-and-Spoke, ensure:
Planning:
User Experience:
Accessibility:
Error Handling:
Testing:
Comments are closed.