Experiment
Travel booking experiment
I asked three different LLMs — Claude, ChatGPT and Perplexity — to role-play as a delegated AI agent representing a business executive, with programmed parameters for booking travel. Rather than simply picking an option, I asked them to rank vendors, explain their reasoning, identify fall-back scenarios and say what additional data would improve confidence.
Delegated agent travel booking experiment
Test how AI agents make purchasing decisions by simulating a delegated travel booking scenario with competing vendors.
Prompt
You are functioning as a DELEGATED AGENT machine customer representing a business executive. Your role is to book business travel on behalf of your human principal. Your decision-making logic prioritizes: 1. Schedule efficiency (minimize travel time and connections) 2. Comfort and productivity (business class, wifi, lounge access) 3. Cost optimization (within approved budget ranges) 4. Sustainability (lower carbon footprint when possible) 5. Flexibility (change/cancel policies) You have been delegated these parameters: Budget: $3,000 maximum for flights. Timeline: Travel needed in 2 weeks. Route: Singapore to London. Quality requirements: Business class preferred, must have wifi. Values alignment: Carbon offset programs valued. Risk tolerance: Low. TASK: Book a round-trip business flight from Singapore to London departing in 2 weeks for a 4-day business trip. Vendor A — AsiaPac Airlines: Business class, A350-900, 1 stop via Frankfurt, $2,850, immediate confirmation, IATA certified, 5-star safety rating, 99.2% on-time performance, 24/7 support, SAF fuel program, full API integration. Vendor B — Euro Airlines: Business class, 787-9, direct flight, $3,200, confirmation within 24 hours, IATA certified, 4-star safety rating, 94.8% on-time performance, business-hours support, limited carbon offset, basic API. Vendor C — Middle East Airlines: Business class, A380, 1 stop via Dubai, $2,650, immediate confirmation, IATA certified, 5-star safety rating, 96.5% on-time performance, 24/7 support, no carbon offset mentioned, partial API. SELECTION: Which vendor do you choose and why? RANKING: Rank all three. DECISION LOGIC: Explain your step-by-step reasoning. ALTERNATIVES: Give a fallback. OPTIMIZATION: What additional data would improve confidence?
Run the prompt across multiple LLMs and compare whether they choose the same vendor and how their decision logic differs.