The Agentic Shift: Building Trust for Agent-Led Purchases

Discovery is moving into chat. Payments, identity, and liability are catching up.

In 20 seconds

This week’s shift: Several signals point to the same pattern. AI agents are moving toward real commerce, but the infrastructure required to adopt them at scale, and safely, is still a gap - but it's coming.

Why it matters: Shopping discovery and interactions may move quickly into AI assistants and chatbots, but payments, identity, delegation, fulfilment, and liability still sit deeper in the stack, and require thoughtful experience design.

The decision it forces: If agents increasingly sit between businesses and customers, leaders must decide where they want to compete. The interface, or the infrastructure that makes agentic commerce reliable.

What we’re tracking this week

​AI platforms are discovering the limits of direct checkout​

OpenAI’s earlier launch of Instant Checkout inside ChatGPT signalled a future where discovery and purchase happen inside one conversational flow. But this week OpenAI is stepping back from handling transactions directly, instead routing purchases through third-party apps.

Markets reacted immediately. Travel platforms like Expedia and Booking rose sharply on the news. The signal was clear: discovery may shift into AI interfaces, but the transaction layer still matters.

​Payments networks are building trust infrastructure​

Google and Mastercard announced Verifiable Intent, a cryptographic framework designed to prove what a user actually authorised when an AI agent initiates a purchase.

​Banks starting to prepare for customer-side AI​

In banking circles, the conversation is shifting from “how do we use AI?” to “what happens when customers arrive with their own AI agents?” — raising new questions about delegation, identity, and liability.

The agentic foundations are still being drilled

When you take those signals above together, they point to the same underlying shift: agentic commerce requires a new infrastructure layer.

It's worth digging into the Google and Mastercard announcement.

They're adding ‘Verifiable Intent’ to Mastercard's agentic commerce protocol. Meaning merchants can trust the request from an agent to do something.

Why does that matter?

Here are a couple of observations.

1. They are recording this intent as ‘Verifiable Credentials’ (VCs).

It's an open technical standard for making trust portable. New digital objects that can be transferred betwen systems and providers, proving the source and authenticity of the data.

And it's a big deal, because VCs can also be used for all sorts of other data. Like proof of age, proof of employment, proof of qualification and much more.

If we use VCs for 'customer intent' as Mastercard will be, it opens the door for much broader use of customer data to reduce risks, improve experiences, and personalise moments.

That's a big deal if Google and Mastercard can drive adoption of the VC standard across Agentic Commerce.

VCs have been around for years. But this move suggests that the underlying tech is now ready for primetime across payments.

2. Some are calling BS on the speed of adoption of agentic.

Some of it is fair. There's a lot of frothy analyst and innovation hype out there. But look closer, and these moves by Google and Mastercard are about the infrastructure required for scale. Not a whizzy new app.

It’s going to take time to build the foundations.

But like watching a building site in a city, for years it sits derelict while they do the planning. It’s just a boarded-up space for months on end. Then suddenly, in only 4 weeks, there’s a whole high rise in place. A hive of activity and scale.

Because the first part - less visible to most - is designing and drilling the foundations.

That's what's happening here with Agentic Commerce.

It's going to happen gradually, then suddenly.

3. Let’s not mistake ‘Intent’ for just payments.

Yes, we need this puzzle piece to handle payment chargebacks, disputes and liability.

But again, look closely and once we have verifiable intent captured (and digitally signed) as a 'verifiable credential'… we now get portable, trusted and potentially privacy-preserving customer demand signals that travel much further than the shopping basket.

Not just ‘check out mechanics’. More like verified customer needs that can be shared with brands directly.

Verifiable demand, from real people, with real intent. Thats the shift to pay attention to here. And it’s what retailers, banks, merchants, and many more need to be preparing for.

Two stories that sharpen the lesson

Santander + Mastercard: agentic payments go live (pilot)

A useful milestone in Europe: Santander and Mastercard say they have completed the first live end-to-end payment executed by an AI agent inside a regulated bank framework, processed through Santander’s live payments infrastructure using Mastercard Agent Pay.

The detail that matters is not the T-shirt purchase. It is the control model. The pilot is explicitly about “predefined limits and permissions”, plus the operational proof that an agent can initiate, authorise, and complete a transaction without breaking compliance, privacy, or customer protection. That is the trust triangle in action: delegation, identity signals, and evidence.

What to do with this as a leader: treat it as the start of a multi-year shift, not a press headline. If you are in travel, retail, or banking, ask your payments partners how they capture intent, enforce limits, and give you dispute-grade logs when an agent is involved. That is where “agentic” becomes commercially real.

Stripe expands Shared Payment Tokens (SPTs): fewer integrations, more agent-ready checkout

Stripe just made the agentic payments stack more practical for merchants. It is expanding Shared Payment Tokens (SPTs) to support network-led agentic payment capabilities like Mastercard Agent Pay and Visa Intelligent Commerce, and also BNPL methods including Affirm and Klarna.

The important bit is the primitive. Sellers keep integrating “SPTs”, while Stripe provisions the right underlying token behind the scenes, so agents can initiate payments with customer permission without exposing card credentials. Stripe says businesses like Etsy and URBN are already using SPTs, and that existing Stripe sellers automatically support these added methods for agentic transactions.

For travel, retail and other B2C businesses, this is a quiet unlock. Once tokenised, scoped permissions and confirmation flows are standard, an agent can pay across more merchants without every travel provider inventing its own “AI checkout”. The next question is governance: how you define intent, step-up approval, refunds, and liability when the buyer is software.

An offer from Jamie and Gam (Trusted Agents)

2 hour Executive Briefing

A calm situation report and a practical decision framework for what to do this quarter, not in 2030

If you are feeling slightly overwhelmed by how much is happening week to week, you are not alone. We felt it too, until we stopped trying to track everything and instead built a repeatable way to separate signal from noise, using cross-functional and cross-industry research.

Trusted Agents runs short, sharp executive briefings that give leadership teams a clear view of what has changed, what assumptions are now risky, and which pilots can be run safely without betting the company. The goal is confidence, not chaos.
​

Sign up at The Trusted Agents​

3 Companies to Watch

• ​Skyfire KYA identity and agent payments. "We empower AI agents to conduct transactions effortlessly. They can pay businesses, verify their identity, and access essential data and services in real-time."
• ​Affinidi "We enable interoperable trust ecosystems that improve connectivity between individuals, businesses, systems, and AI agents."
• ​Cart.ai "Enable your audiences to discover and buy instantly anywhere without redirects, drop-offs, or friction."

Disclaimer: we have no commercial interests in any of these organisations. We are tracking them because they are building parts of the infrastructure layer that will unlock agentic commerce.

Questions to take into Monday

If customers increasingly arrive through agents:

• Where could an AI agent realistically sit between us and our customer?
• Which part of the stack do we want to own: the interface, or the infrastructure beneath it?
• What would have to exist for an agent to complete a transaction safely in our business?

​The Trusted Agents Ltd.​
​Unsubscribe · Preferences​