As AI agents gain autonomy, trust becomes the new currency. Mastercard wants to power both.

Business Insider

03-07-2025

By Matthew Driver, executive vice president, services, Asia Pacific
Today, AI helps us research, plan, and decide. Tomorrow, it will act.
The rise of agentic AI marks a shift from assistance to execution —where AI not only suggests but also completes tasks like booking, buying, and managing on your behalf.
Agentic commerce isn't just about where you buy — it's about why, when, and how intelligently those purchases are made. This means that AI "agents" will be empowered to initiate tasks, as well as make decisions within clear parameters defined by you. And there are huge implications for digital commerce. Whether it's booking a flight, buying multiple products from different e-commerce merchants, or fulfilling orders digitally from buyers, agentic AI works proactively — often in the background — to get things done.
Agentic AI in practice
To appreciate the paradigm shift this technology represents, consider a few near-future scenarios.
• Autonomous retail concierge that shops on your behalf — blending personal taste with real-time price hunting, inventory checks, shipping optimization, and seamless return handling. Whether assembling an outfit from multiple brands or sourcing a rare gadget at the best price, it acts as your behind-the-scenes strategist, delivering a frictionless retail experience across platforms.
• Intelligent home steward that takes the stress out of home and digital life management. It notices when supplies run low, reorders essentials, fine-tunes lighting and climate preferences, and prunes unused subscriptions — all without prompts. It's like having a quiet, invisible house manager who knows your rhythm and keeps everything running without ever needing to be asked.
• Smart business quartermaster that acts as your behind-the-scenes enabler — quietly managing inventory, sourcing supplies, renewing software, and handling operational details before they become bottlenecks. It's like having an always-on business partner who keeps your tools sharp, your shelves stocked, and your systems running, so you can stay focused on building what matters.
By taking on execution, agentic AI shifts technology from being simply helpful to genuinely timesaving, reducing manual effort, and eliminating everyday friction from our lives.
Know your agent: AI agency, with accountability
Naturally, the idea of giving an AI agent any degree of autonomy to act on our behalf raises important questions about security and safety. Invoking sci-fi movie examples, we want our agent to act more like R2-D2 from Star Wars or TARS from Interstellar — rather than inadvertently ending up with a malicious and self-interested HAL from 2001 or Ultron from The Avengers.
Perhaps the most pressing issue is consent. How does a user grant an AI permission to access sensitive data — but only certain types for certain tasks — and operate within well-defined limits while still retaining meaningful control? Much of this starts with smart design: Clear interfaces that allow users to set parameters, define triggers, and establish thresholds for when the agent should act, pause, or ask for confirmation.
These guardrails are relatively straightforward in more structured tasks — say, planning an overseas holiday. An agent might be asked to build an itinerary and source the best deals on flights, hotels, and activities within a range of time and cost parameters. Once the options are presented, the user can select their preferences and authorize the AI to complete bookings. But the challenge deepens when full automation is the goal.
Imagine an agent tasked with monitoring secondhand marketplaces and purchasing specific collectibles the moment they appear. In this case, the user needs to set clear parameters — how many items to buy, the maximum spend per item, and whether to prioritize rarity or price. The more autonomy the agent has, the more critical it becomes to define these components and limit thresholds up front, as well as its operating domain.
Transparency is a vital part of the guardrail system — not just for building trust, but for defending against manipulation. As agents take on more decision-making power, users need visibility into how and why those decisions are made. Whatever an AI recommends, the user should be able to see why it was chosen and verify that it truly was the best option available.
That visibility becomes even more important in an environment where bad actors may attempt to game the agent's logic — inflating prices, mimicking legitimate offers, or triggering purchases through misleading signals.
Building the infrastructure for trusted autonomy
Mastercard is taking a proactive stance in shaping the infrastructure that will support agentic commerce. Through our newly launched Mastercard Agent Pay, we're integrating the power of tokenization into this new technology, enabling AI agents to transact securely, with human authorization and authentication embedded from the outset.
At the heart of this is the Agentic Token, a next-generation credential designed to ensure that every payment request an AI makes is traceable with predefined rules and permissions.
As the program matures, merchants will be confident they're receiving legitimate transactions from verified, authorized agents. In the future, users can retain control over where, how, and under what conditions their agents can act. And in case there is a dispute or error, the full record of what was ordered, bought and delivered is securely logged and recorded.
While we are still in the nascent stages of this technology, this is the beginning of a new conversation. As these systems see wider adoption, new questions will emerge around accountability, interoperability, and ethics. By taking measured, careful steps towards giving agency to AI — while embedding trust into its foundations — we can build confidence from the outset in an autonomous future for commerce.
Big you. Small you. Always-on you. (Or you, multiplied?)
Think of agentic commerce as your parallel digital self: An always-on version of you, quietly navigating the noise, transacting, optimizing, and executing in the background, according to your defined parameters. While you focus on the big moments, it takes care of the small ones: booking, buying, scheduling, syncing.
It's not just assistance — it's amplification and a new way to transact. A tireless counterpart built for speed, precision, and flow. And the most powerful part? It works so seamlessly, you'll forget it's there until you realize how much more you're getting done. And that's exciting.