An AI agent has booked and paid for a ride to Singapore’s Changi Airport in a live, authenticated transaction. The journey itself was ordinary. The payment chain behind it was not.
Mastercard completed the transaction with DBS and UOB, using mobility provider hoppa and an agent supplied by CardInfoLink. It marks the company’s first live agent-based payment in Singapore and moves autonomous commerce one step beyond controlled demonstrations.
An agent can now complete the purchase
The agent identified the transport service, made the booking and initiated payment. Mastercard Agent Pay provided the framework, assigning a Mastercard Agentic Token to the agent. Consumer consent was captured explicitly, while Payment Passkeys authenticated the user and confirmed the purchase.
Those layers address the central problem in agentic commerce: how to let software act without making financial authority ambiguous. Tokenisation avoids handing the agent a reusable card credential. Passkeys verify the consumer. Explicit consent establishes that the agent has permission for the transaction.
The result is more than a chatbot finding a taxi. It is a full chain in which software interprets a need, selects a provider and completes payment. Removing the final manual confirmation can make an experience smoother, but it also raises the importance of clear limits. Consumers need to know what an agent may buy, how much it may spend and when it must ask again.
Singapore becomes a live test market
Mastercard has carried out authenticated agentic transactions in Australia, New Zealand and India, but Singapore occupies a strategic position. The company is establishing a regional AI Centre of Excellence there and deploying dedicated agentic-commerce teams across Asia-Pacific.
Singapore’s banks are also testing competing networks. DBS completed a separate Visa pilot in February 2026 in which agents executed food and beverage purchases using DBS and POSB cards. Its participation in milestones from both Mastercard and Visa within weeks indicates that banks do not want to wait for one standard or provider to define the market.
Mastercard plans to expand Agent Pay into transportation, travel and retail—areas where repeated searches and checkout steps create friction. These categories are also easy for consumers to understand: book a ride, arrange a journey or replenish an item within a stated budget.
The ride to Changi therefore functions as a deliberately simple proof. It shows that several independent organisations can coordinate identity, consent, merchant access and settlement around an agent’s action.
Trust will determine whether convenience scales
The technical ability to pay does not settle questions about responsibility. What happens if an agent selects the wrong journey, misunderstands a constraint or spends outside the user’s intention? A usable system needs receipts, dispute paths and a visible record of what was authorised.
Banks and payment networks already manage fraud, authentication and merchant acceptance. Agentic payments extend those controls to a new actor: software operating on a consumer’s behalf. Per-agent tokens offer a way to identify that actor and potentially limit its scope without confusing it with the human account holder.
Competition between Mastercard and Visa may accelerate deployment, while banks such as DBS can learn across both ecosystems. The winners will not simply process an AI-initiated transaction first. They will make the experience understandable enough that customers are willing to delegate.
Mastercard’s Singapore transaction is modest in value but important in structure. The agent did not merely recommend a ride or fill a checkout form. It completed an authenticated payment through live financial infrastructure.
Autonomous commerce has therefore entered a more demanding phase. The next milestones will be measured not by novelty, but by whether consent remains meaningful, errors are recoverable and customers feel in control even when they no longer press the final button.
