Dewy Resort Demo

Live Demonstration (15 minutes)

What You’ll See

During this live demonstration, your facilitator will show the Dewy Resort application in action. Watch for these key concepts:

Compositional MCP Tools in Action

Instead of calling many small API wrappers, you’ll see how a single compositional tool handles entire workflows:

Scenario What Happens
Guest Check-In One tool call updates booking status, room status, and opportunity stage atomically
Checkout + Maintenance Two intents in one message handled by independent tools
Edge Case Handling Structured errors enable graceful recovery without hallucination

Key Takeaways to Watch For

  • Single tool calls instead of 6+ API wrappers
  • Fast execution (3-4 seconds vs 8-12 seconds)
  • Higher reliability (~94% vs ~73% success rate)
  • Natural conversation - no complex prompt engineering required
  • Observability - every step traceable in Workato logs

Demo Scenarios

Your facilitator will demonstrate three scenarios:

  1. Guest Check-In - A guest arrives and checks in using their email
  2. Checkout with Maintenance Request - A guest checks out while reporting an issue
  3. Edge Case - Requesting service for a non-existent room

After each scenario, pay attention to how the LLM selects tools and how errors are handled gracefully.

Facilitator Demo Script

Pre-Demo Checklist

  • Dewy Resort app running locally or on demo server
  • Workato workspace open in separate tab
  • Salesforce org open in separate tab
  • Fresh test data (or reset seed data)

Scenario 1: Guest Check-In (5 min)

User Input:

“I’m Sarah Johnson, checking in. My email is sarah@example.com.”

Tool Called: check_in_guest

What to highlight:

  • Single tool call (not 6+ API wrappers)
  • 3-second execution time
  • Atomic state transitions:
    • Booking: Reserved -> Checked In
    • Room: Vacant -> Occupied
    • Opportunity: Confirmed -> Checked In

Show in Workato:

  1. Open Job History
  2. Find the check_in_guest execution
  3. Expand to show:
    • Input parameters (guest_email)
    • Step 1: Search Contact (300ms)
    • Step 2: Search Booking (400ms)
    • Step 3: Validate Room (200ms)
    • Steps 4-6: Update operations (500ms each)
  4. Highlight transaction-level tracing

Show in Salesforce:

  1. Open the Contact record
  2. Show Booking record status change
  3. Show Room status change

Scenario 2: Checkout with Maintenance Request (5 min)

User Input:

“Beth Gibbs is checking out, and she mentioned the toilet in room 302 is broken.”

Tools Called:

  1. process_guest_checkout
  2. submit_maintenance_request

Contrast with Naive Approach (show slide or whiteboard):

Approach Tool Calls Latency Success Rate
Naive 11+ 8-12 sec ~73%
Compositional 2 3-4 sec ~94%

What to highlight:

  • LLM correctly identified TWO intents in one message
  • Each tool handled its workflow independently
  • Maintenance request includes room context automatically
  • No complex prompt engineering required

Scenario 3: Edge Case Handling (3 min)

User Input:

“I need extra towels in room 404.”

(Room 404 doesn’t exist in the system)

Tool Called: submit_guest_service_request

Error Returned: 404 - Room not found

LLM Response:

“I couldn’t find room 404 in our system. Could you double-check the room number?”

What to highlight:

  • Structured error codes enable graceful recovery
  • LLM doesn’t hallucinate or retry endlessly
  • User gets actionable guidance
  • Error logged for debugging

Behind the Scenes: Observability (2 min)

Show Logs:

  1. In Workato, navigate to Tools > Logs
  2. Click on the most recent logs (displayed in descending order by default)
  3. Show:
    • All three executions visible
    • Correlation IDs (idempotency tokens)
    • Input/output for each step
    • Execution duration breakdown

Demonstrate Idempotency:

  1. Copy the idempotency token from a successful check-in
  2. Attempt to re-run the same check-in
  3. Show: Returns success without duplicate operations
  4. Explain: “Safe to retry. No duplicate bookings.”

Q&A Transition

“Any questions about what you just saw? We’re about to take a break, and then you’ll set up your own environment to build these patterns yourselves.”

Common questions to prepare for:

  1. “How does it know which tool to call?”
    • MCP tool descriptions guide selection
    • LLM reasoning in context window
  2. “What if both tools fail?”
    • Each tool independent
    • Partial success is possible
    • Saga pattern for rollback (advanced topic)
  3. “How do you test these tools before going live?”
    • Workato provides test mode for individual recipes
    • Use recipe versioning to iterate safely
    • MCP tools can be tested directly via API before connecting to an LLM
    • Mock payloads can be built into recipe designs for testing

Transition to Break

“We’ll take a 15-minute break. When we come back, you’ll set up your own environment - Salesforce metadata, Workato recipes, and the local application. Grab coffee, stretch, and we’ll dive into hands-on work.”

Display on screen during break:

  • Wifi credentials
  • Restroom locations
  • Slack/Discord channel for questions
  • “Back at [TIME]”