In Part 1 of this series, we established what agentic commerce is and why business leaders need to act now. This instalment moves beyond strategy into the technical infrastructure required to participate in — and benefit from — an economy where AI agents research, negotiate, and transact on behalf of consumers.
There are four critical areas every merchant must master: data architecture, payment protocols, brand-owned agents, and organisational readiness. Each is interconnected. Miss one, and the others become significantly less effective.
1. Data: The Foundation of Agentic Commerce
From SEO to GXO
Search engine optimisation was built for human eyes. The emerging discipline of GXO — Generative Experience Optimisation — is built for AI agents. Where SEO focused on keywords, backlinks, and crawlability, GXO requires semantically rich, machine-readable digital assets that AI systems can interpret, trust, and recommend.
Your product catalogue is no longer just a sales tool. It is a primary interface through which AI agents will evaluate whether your products are worth recommending to their users.
Building an AI-Ready Data Architecture
Three components form the foundation:
- Structured Data: Implementing Schema.org and well-defined schemas so AI agents can interpret product information without ambiguity. If an agent cannot confidently understand what you sell, it will not recommend it.
- Enriched Content: Detailed product attributes, high-quality imagery, and rich metadata. Comprehensive catalogues with extensive attributes and semantic enrichment connecting products to customer needs — usage context, benefits, reviews — give agents the information they need to match your products to user intent.
- Optimised Taxonomies: Consistent category structures that support natural language processing and agentic discovery. Inconsistent naming conventions and fragmented categories create gaps that agents fall through.
The Model Context Protocol (MCP)
MCP is an open standard facilitating secure, standardised connections between data sources and AI tools. Think of it as the AI equivalent of the API standards that enabled the web economy — it allows agents to access your product data in a consistent, trusted format. An API-first approach, with standardised data formats and interconnected context availability, makes your catalogue accessible to the agents your customers will increasingly rely on.
Trustworthy data — clean, structured, connected, and permissioned for governance and privacy compliance — is not just a technical requirement. It is a competitive edge. Agents will preferentially surface products from merchants whose data they can trust.
2. Agentic Payments: The Engine of Autonomy
For agentic commerce to function, AI agents need the ability to transact securely on behalf of users. A new layer of payment infrastructure is emerging specifically to support this. The major protocols currently being developed:
- Agentic Commerce Protocol (ACP) — Stripe & OpenAI: An open standard enabling AI agents to interact with merchant checkout flows via Shared Payment Tokens.
- Agent Payments Protocol (AP2) — Google: Focused on trust, authorisation, and verifiable intent through Verifiable Digital Credentials that create cryptographic audit trails.
- x402 Protocol — Coinbase:Internet-native micropayments using the HTTP 402 "Payment Required" status for stablecoin-based transactions.
- Mastercard Agent Pay: A framework for trusted AI agent recognition and secure transaction tokenisation.
- Visa Intelligent Commerce: Empowers agents to pay according to consumer preferences and spending limits, with personalised security delegation.
How Agentic Payments Work: The Transaction Lifecycle
A fully autonomous purchase flows through five stages:
- Delegation & Constraints:The user grants the agent payment authority with defined limits — an "Intent Mandate" that specifies what the agent can and cannot do.
- Discovery & Negotiation: The agent autonomously identifies products, compares options, and negotiates terms within its authorised parameters.
- Secure Authorisation: The transaction is presented with non-repudiable consent proof — the agent must demonstrate it acted within its mandate.
- Execution & Settlement: Payment executes via secure, tokenised credentials, completing the transaction without human intervention.
- Auditing: Every step is logged to create a complete record for dispute resolution and fraud analysis.
The payment providers you work with today need to be evaluated against this roadmap. Those that cannot support agentic transaction flows will become bottlenecks as the market matures.
3. Launching Your Own Agents: The Offensive Strategy
Waiting for third-party platforms to mediate your customer relationships is a defensive position. The merchants who will win in the agentic era are those who build proprietary agents that represent their brand, serve their customers, and participate as peers in agent-to-agent commerce.
Why Build Your Own Agent?
- Personalised AI Experiences:A brand-owned agent can deliver product recommendations, customer service, and purchasing assistance that reflects your brand voice and leverages your proprietary customer data. Lowe's "Mylow" is an early example of a retailer building this kind of owned AI presence.
- Interoperability: As merchant-to-merchant (M2M) agent interactions become standard, having a capable agent means participating directly in that ecosystem rather than being represented — imperfectly — by a third party.
- Brand Preservation: An agent you build carries your voice, your expertise, and your values. A third-party agent optimises for its own objectives.
The A2A Protocol
The Agent-to-Agent (A2A) Protocol, developed by Google and now managed by the Linux Foundation, provides the common language through which agents communicate. Key concepts include the Agent Card — a JSON metadata file detailing your agent's identity and capabilities — and standardised operations (SendMessage, GetTask) that allow agents to collaborate regardless of the underlying technology stack.
Who is Already Doing This
- Walmart:"Sparky" handles customer-facing interactions while "Wally" manages merchant operations. Their OpenAI partnership enables ChatGPT integration, making Walmart's inventory accessible to one of the world's largest AI platforms.
- Shopify: Building cross-merchant cart infrastructure that allows a single agent to purchase across multiple Shopify stores in a single session.
- Perplexity (Buy with Pro): Evolved from a search engine into a full commerce agent handling product selection, shipping, and execution.
4. Building an AI-Ready Enterprise
Technology alone does not make a business agentic-ready. The organisational structures, skills, and culture that succeed in the agentic era look fundamentally different from those built for traditional eCommerce.
A New Operating Model
The shift from human-centric workflows to AI-first approaches with human oversight is not a technology project — it is a business transformation. The agentic organisation is a network of empowered, outcome-aligned teams working with AI agents as collaborative partners, not tools to be operated.
Cultivating AI Competencies
Four capability areas require deliberate investment:
- AI Literacy for Non-Technical Teams: Everyone in the organisation needs to understand how AI creates and destroys value in their domain. This is not optional upskilling — it is foundational.
- Data Engineering & Knowledge Architecture: The teams that build and maintain the data foundations that agents depend on.
- Prompt and System Design: The skill of crafting effective agentic instructions — defining what agents should do, within what constraints, and to what end.
- Governance & Ethics: Ensuring AI deployment is responsible, compliant, and aligned with customer trust. Agents that behave badly damage brands at scale and at speed.
Fostering an Experimentation Culture
Agentic commerce is too new for anyone to have the definitive playbook. The organisations that will lead are those that build safe sandboxes for experimentation, iterate rapidly, and make decisions based on what they learn from pilots rather than predictions. An "Experiment, Evaluate, Deploy, Repeat" mindset is not a nice-to-have — it is the only way to move forward responsibly in a landscape that is changing this quickly.
The Risks of Inaction
Businesses that do not build these capabilities face three compounding risks:
- Widening Skills Gaps: As autonomous systems become standard, the distance between those who can manage them and those who cannot grows rapidly.
- Competitive Disadvantage: Early adopters who build agentic capabilities now will accumulate data, experience, and customer trust that latecomers cannot easily replicate.
- Operational Risk: Compliance and security vulnerabilities in agentic systems — left unaddressed until they become critical — are far more expensive to fix than to prevent.
A Five-Point Roadmap
Based on where the market is heading and the technical building blocks available today, here is a practical starting point:
- Prioritise AI literacy across all organisational levels — not just technical teams.
- Redefine structures toward flat, cross-functional teams aligned around agentic outcomes.
- Invest in data foundations — structured catalogues, MCP implementation, clean taxonomies — as a primary competitive differentiator.
- Balance innovation with ethics and risk management. Speed without governance is expensive.
- Build internal capabilities to reduce dependency on external platforms whose interests will not always align with yours.
The Bottom Line
Agentic commerce is no longer a concept. It is happening now. The businesses that succeed will be those that accept this revolution and actively shape its direction — not those that wait for the wave to arrive before deciding whether to swim.
The technical foundations are available. The protocols are being standardised. The early movers are pulling ahead. The question now is execution.