From Interfaces to Infrastructure: What I Learned About Canada’s Path to Agentic Commerce

At this year’s Payments Canada SUMMIT, I attended several discussions focused on the future of AI-driven commerce, digital payments, and financial infrastructure. The conversations connected emerging AI capabilities with the realities of payment systems, trust, and interoperability in a way that made agentic commerce feel less like a future concept and more like the natural next step in the evolution of digital commerce.

A major highlight was the discussion with Dan Iwachiw around Visa’s “Agentic Ready” initiative, designed to help issuers and ecosystem partners safely explore agent-initiated payments as AI commerce continues to evolve. What I found especially interesting was the emphasis on enabling innovation responsibly, with security, reliability, and governance built directly into the infrastructure from the start.

The broader conversation centered on a key shift already beginning to take shape: AI systems are moving beyond recommendation and assistance toward autonomous execution. Agents are increasingly being designed not only to interpret requests, but to negotiate, authorize, and complete transactions on behalf of users.

And that changes the role of payments infrastructure entirely.

Because ultimately, agentic commerce depends on much more than intelligence. It depends on trust.

The Shift from Human-Centric to Agent-Centric Banking

One of the clearest themes throughout the session was that banks and payment providers will require fundamental architectural redesigns to support this future.

We are moving from:

“Serving humans through interfaces”

to:

“Serving autonomous agents through programmable infrastructure.”

Traditional banking systems were designed around human interaction.

But the emerging model looks very different:

This means financial systems must support:

• real-time decision making
• policy-driven execution
• machine-to-machine authorization
• continuous operation
• programmable trust frameworks

Agentic AI: Systems That Don’t Just Recommend, They Act

A major focus of the discussion was the evolution from traditional AI assistants to truly agentic systems.

Agentic AI systems don’t simply generate suggestions. They:

• interpret goals
• break tasks into workflows
• coordinate with other systems
• negotiate outcomes
• execute transactions autonomously

In commerce, this creates what many are calling agentic payments.

Instead of:

“Here are three hotels you may like.”

The next generation of AI systems may:

“I booked the optimal hotel within your budget, policy constraints, and loyalty preferences.”

That’s the key shift:

Human clicks “Buy” → Policy-driven machine execution.

Understanding the Emerging Agentic Stack

One of the most useful frameworks discussed during the session was the layered architecture emerging around agentic commerce.

I)  MCP (Model Context Protocol)

Developed by Anthropic, MCP standardizes how agents access tools, APIs, data, and services.

The simplest way to think about it is:

“USB-C for AI systems.”

MCP gives agents capability.

II) A2A (Agent-to-Agent Protocol)

At the same time, initiatives associated with Google and the broader ecosystem are working on A2A protocols that standardize how agents communicate and coordinate with one another.

A2A gives agents collaboration.

III) Payment Protocols

AP2, x402, and other emerging frameworks. They allow value exchange between agents, merchants, and financial systems.

At that point, agents stop functioning as assistants and start functioning as independent economic participants.

The Core Problem: Identity and Trust

As soon as agents can transact autonomously, one question becomes absolutely critical:

Who is this agent, and should I trust it?

Traditional e-commerce revolves around human identity. Agentic commerce introduces machine identity.

The session explored two major approaches to solving this problem.

Visa TAP vs ERC-8004

What I found particularly interesting is that these approaches aren’t necessarily competing. They’re complementary.

A future trust stack may look something like this:

The distinction that stayed with me was:

• Agent identity = who the agent is
• Agent trust = whether its actions should be accepted in context

TAP creates trust in the moment.

Reputation systems create trust over time.

Why Canada’s Real-Time Rail Matters

A major topic throughout the summit was the importance of Canada’s Real-Time Rail (RTR) infrastructure in enabling agentic commerce.

Agentic systems operate continuously. That means payments infrastructure must also operate continuously.

RTR becomes incredibly important because it supports:

• instant settlement
• 24/7 operation
• rich ISO 20022 messaging
• real-time processing

Unlike traditional batch systems, agentic payments happen dynamically, mid-process, and autonomously.

That makes real-time infrastructure essential.

Discovery, Checkout, and Autonomous Payments

Another area that caught my attention was how discovery, checkout, and payment are converging into fully AI-native experiences.

Several protocols are emerging to support different parts of that journey:

• UCP → discovery and shopping
• ACP → checkout orchestration
• x402/AP2 → payment execution

This enables a very different kind of commercial experience.

Each protocol fills a different layer of the transaction lifecycle.

Key characteristics include:

• Delegated authority
• Autonomous decision-making
• Real-time execution
• Negotiation before payment

The important distinction is that this is not simply automated payments.

This is decision-making plus execution.

Example: Autonomous Travel Booking

Consider the simple user request:

“Book me a three-day trip under $1,200.”

What will happen:

All of this can happen directly inside the conversational interface, without redirects, forms, or manual checkout flows.

That’s a dramatic change from today’s commerce experience.

The Rise of the B2AI Economy

One of the strongest ideas discussed was that AI agents are becoming an entirely new buyer category.

This creates a new commercial model:

B2AI: Business to AI Agent

Businesses won’t just optimize for humans anymore.

They’ll increasingly need to optimize for:

• machine-readable APIs
• structured pricing
• interoperable systems
• programmable commerce flows

In many ways, this feels like the next evolution of digital transformation.

From SEO to AIO

We once optimized for search engines.

Soon, businesses may optimize for AI agents.

That means moving from:

Some are already calling this:

“SEO → AIO (Agent Interface Optimization).”

The Economic Implications

The implications extend far beyond payments.

1. Relentless Price Optimization

Agents optimize for:

• price
• speed
• availability
• policy alignment

That may significantly reduce brand-driven pricing power in some industries.

2. Explosion of Microtransactions

Agents can transact:

• per API call
• per second
• per task

That unlocks entirely new monetization models.

3. Platform Disruption

Large intermediaries may lose influence as agents transact directly across ecosystems.

4. New Business Models

We’re likely to see:

• agent-native marketplaces
• AI subscription services
• autonomous procurement systems
• programmable financial products

My Biggest Takeaway

The biggest realization I left the summit with is that agentic commerce is not just about smarter AI.

It’s about creating a machine-to-machine economic layer where autonomous systems can:

• collaborate
• negotiate
• authorize
• purchase
• settle value independently

The technology stack is already emerging:

• AI reasoning systems
• interoperability protocols
• identity frameworks
• real-time payment rails
• programmable authorization systems

The next challenge is operationalizing all of it securely.

And in my view, the most important point is this:

Agentic commerce only works if trust is embedded directly into the infrastructure itself.

Written by: Lesya Berbeka, AI/Quantitative Research and Analytics Lead at Kodershop