AXD Observatory: Essays on Agentic Commerce and AI Design

Trust in agentic systems is not a feeling. It is an architecture - with layers, load-bearing elements, and failure modes that can be designed.

Banking services negotiated while you sleep. Insurance that adjusts without a form. The stratum of agentic capability that creates value without any visible interface.

Every agentic system begins with an act of delegation. How that delegation is designed determines whether the system serves or subverts its principal.

In the Japanese art of kintsugi, broken pottery is repaired with gold. The fractures are not hidden - they are illuminated.

By 2028, autonomous agents will be making purchasing decisions, negotiating contracts, and managing financial relationships without human intervention.

When you delegate authority to an autonomous agent, you give consent that extends across time in ways fundamentally different from any previous form of digital permission.

In agentic systems, the interface is not the experience - it is the interruption. The A2UI is the narrowest possible opening between human attention and autonomous action.

In the agentic age, interfaces are not designed once and deployed - they are assembled dynamically by agents from modular components, tailored to context, task, and user.

Every system will fail. The measure of an agentic system is not how rarely it fails, but how gracefully it recovers. Trust is a muscle that strengthens through repair.

An operational envelope defines the precise boundaries within which an agent is authorised to act. Too narrow, and the agent is useless. Too wide, and it becomes dangerous.

The relationship between a human and an autonomous agent is not static. It follows a developmental arc - from cautious introduction through deepening trust to mature partnership.

The more you observe an autonomous agent, the less autonomous it becomes. The less you observe it, the less you trust it. Resolving this paradox is the central design challenge.

Beneath every agentic experience lies an intelligence layer - the substrate of reasoning, learning, and adaptation that makes autonomous action possible.

Like technical debt, trust debt accumulates silently. Each small violation - a missed expectation, an unexplained decision - compounds until the relationship collapses.

What happens when an agent's ethical constraints conflict with its user's instructions? Autonomous integrity is the design principle that governs these moments.

Trust is not a constant. It fluctuates with time, context, and experience. Designing for temporal trust means designing for a relationship that is always in motion.

Delegation scope defines what an agent may do, under what conditions, for how long, and with what resources. It is the grammar through which humans express authority.

An interrupt that comes too often trains humans to dismiss it. An interrupt that comes too rarely allows consequential errors to accumulate. Getting this right is a design problem of the first order.

The shift from instructing to specifying outcomes is the fundamental interaction paradigm of agentic systems. Users must learn to express intent, not procedure.

Autonomy is not the absence of design - it is the highest expression of it. Every boundary, every operational envelope, every trust calibration mechanism is a design decision that shapes what an autonomous agent can become.

Three protocols are emerging as the connective tissue of the agentic age. MCP standardises tool access, A2A standardises agent delegation, and ACP standardises content exchange. Together, they form the structural grammar through which trust, delegation, and observability are implemented at scale.

When an autonomous agent reaches for the wallet, every assumption in the payments stack collapses. AP2 solves this with cryptographically signed mandates - Verifiable Digital Credentials - that anchor every transaction in provable human intent.

A mandate is not a command and not a wish. It is a contract between human intention and machine autonomy. This essay catalogues the five design patterns - intent capture, boundary specification, authority gradients, lifecycle management, and failure escalation - that define how humans delegate authority to autonomous agents.

Don Tapscott's identic AI - personal agents as cognitive extensions of the individual - is a superpower for consumers. For banking, it is a structural reckoning. What happens when your customers deploy AI on behalf of themselves, armed with infinite patience for optimisation and zero brand loyalty?

When machines become customers, identity verification must evolve beyond human biometrics. Know Your Agent introduces four pillars - authentication, mandate verification, behavioural fingerprinting, and principal traceability - as the regulatory and design foundation for admitting autonomous agents into commercial relationships.

Commerce has always been bilateral. A buyer and a seller. Agentic commerce shatters this assumption. When a human delegates purchasing authority to an autonomous agent, and that agent transacts with a service provider, the straight line becomes a triangle - three parties, three relationships, three distinct channels of trust.

For a century, commerce has been a visual medium. When a machine customer evaluates your product, it does not see your photography. It reads structured data, queries APIs, and evaluates freshness timestamps. Signal clarity is the first pillar of AXD readiness.

Trust is calculated. This is the sentence that separates the human era of commerce from the agentic era. A machine trusts a vendor with a 99.9% verifiable uptime score over a vendor with a Satisfaction Guarantee badge every single time.

A human walks into a shop and says something vague. An agent receives a structured mandate. Intent translation is the third pillar of AXD readiness - the discipline of ensuring your products can be found when agents search with structured criteria rather than emotional impulses.

The checkout button is dying. An agent does not navigate a website. It calls an API endpoint, passes a structured payload, and receives a confirmation in milliseconds. Engagement architecture is the fourth pillar of AXD readiness.

The agentic transition is not a single event. It is a structural shift. The AXD Institute's answer is the Four Pillars of AXD Readiness - Signal Clarity, Reputation via Reliability, Intent Translation, and Engagement Architecture - a unified strategic framework for businesses preparing for the agentic economy.

Financial services faces the most consequential AXD readiness challenge of any sector. This essay applies the Four Pillars framework to banking, insurance, and wealth management, mapping the structural gaps that will determine which institutions thrive in the agentic economy.

Retail faces the most immediate AXD readiness challenge. With AI agent traffic up 4,700% year-over-year, this essay applies the Four Pillars framework to e-commerce, grocery, and luxury retail - mapping the structural transformations that determine which retailers thrive when machine customers do the shopping.

Before an agent can trust, delegate, or transact, it must answer a more fundamental question: who am I looking at? Entity resolution - determining when different records refer to the same real-world entity - is the identity infrastructure upon which the entire edifice of agentic commerce depends.

On January 11, 2026, Google announced the first open standard that gives autonomous agents a common language for commerce. UCP is not a product - it is infrastructure. And infrastructure determines what can be built on top of it.

When autonomous agents optimise metrics without understanding purpose, they can hit every target while violating every value. Intent Engineering is the discipline of encoding what an organisation actually wants — not just what it can measure.

Markdown has evolved from a developer documentation format into the primary communication layer between humans and autonomous AI agents. The three layers — discovery, content, and instruction — form the complete architecture of the agentic web.

Stripe's 2025 annual letter introduced a five-level maturity model for agentic commerce. The levels describe what agents can do - but not what humans experience. An AXD analysis of what the model reveals, what it omits, and what trust architecture demands.

Stripe's x402 protocol and Shared Payment Tokens are building the payment infrastructure for Level 4-5 agentic commerce. But payment rails carry money - not trust. An AXD analysis of what machine payments solve and what they leave dangerously undesigned.

Mastercard treats the agent as a new entity with its own identity. Visa treats the agent as an extension of the human. Both build payment infrastructure. Neither builds trust architecture. The deepest architectural fault line in agentic commerce.

Stripe launched x402. Mastercard completed the first authenticated agentic transaction. Visa piloted Intelligent Commerce. Alipay processed 120 million transactions in a week. Everyone built payment rails. Nobody built trust rails. The first quarterly briefing on where agentic payments stand and what they still lack.

NIST published its concept paper. The EU AI Act reaches full applicability in five months. Six private-sector organisations launched KYA frameworks in sixty days. The regulatory reckoning for agentic commerce is not a future event. It is a present one. Five jurisdictions, the SCA paradox, and the four gaps no framework addresses.

Mastercard's Verifiable Intent is not a product announcement. It is an architectural declaration - the first payment network protocol that treats agent intent as a verifiable, auditable, constrainable object. Five shifts that change everything: from identity to intent, from static to dynamic, from bilateral to trilateral, from authentication to attestation, from compliance to governance.

A customer asks an AI assistant about their finances. The agent answers, routes, and executes - and the bank that holds the mortgage, the current account, and fourteen years of relationship data was never consulted. The Principal Gap measures the distance between where intent forms and where institutions receive it. As that distance hardens, the window to become an Instruction Receiver is closing.

On a Tuesday in March 2026, a brand's autonomous media buying agent submits a bid for a premium video placement. The publisher's selling agent evaluates it against fourteen other bids and accepts. The transaction completes in eleven milliseconds. No human on either side was consulted. The IAB Tech Lab's AAMP initiative - with its three pillars of ARTF, Agentic Protocols, and the Agent Registry - is the most advanced real-world implementation of agent-to-agent commerce infrastructure in any industry.

Pull up your calendar from last week. Count the hours transferring information, translating knowledge into documents, and synchronising humans who cannot share context directly. Microsoft's 2025 Work Trend Index found 57% of work is communication, 43% creation. Sana puts it at 60% 'work about work.' This essay examines what happens when the coordination layer that defines modern organisations begins to dissolve - and what genuinely human work survives.

In May 2025, Shopware's CEO published the first comprehensive platform vendor playbook for agentic commerce readiness. It proposes four KPIs, emerging trust metrics, and a phased implementation roadmap. This essay analyses the playbook through the AXD lens - mapping its operational metrics to the Five Pillars of AXD Readiness and identifying the structural gap where trust architecture begins.

Zero-click commerce is not a UX optimisation - it is the structural consequence of delegation. When an agent purchases on behalf of an absent human, the checkout disappears entirely. This essay examines the design architecture of commerce that happens without a click, without a screen, and without a customer present - and why absent-state design is the defining competence of the agentic age.

Agent legibility is the property that makes an autonomous system's state, capabilities, constraints, and reasoning accessible to human understanding. Without legibility, there is no basis for trust calibration. This essay examines why legibility - not transparency - is the design requirement for agentic AI, and how Signal Clarity provides the framework for making agents readable.

B2B agentic commerce is the domain where autonomous AI agents represent organisations in procurement, negotiation, and fulfilment. Unlike B2C agentic shopping, B2B involves multiple principals, complex delegation chains, and agent-to-agent negotiation. This essay examines the multi-principal problem, inter-organisational trust architecture, and the design of the agent-to-agent marketplace.

A non-human economic actor (NHEA) is an autonomous AI system that participates in economic activity as an active participant rather than a passive tool. This essay examines the five dimensions of NHEA design: identity and authentication, delegation and authority, legal and regulatory status, market participation roles, and trust governance — arguing that economic infrastructure must be redesigned for participants that are not human.

A customer configures their AI agent on a Friday evening. Six weeks later, they lose their job. The agent does not know. It was not designed to know. KYC verified identity at onboarding. KYA verified the agent's legitimacy. But nothing verifies the human behind the mandate in the present moment. The AXD Institute introduces Know Your Human (KYH) — the design framework for continuous human validation, authority drift, and graceful suspension in the agentic economy.

On March 18, 2026, Stripe and Tempo published the Machine Payments Protocol — an open standard for machine-to-machine payments over HTTP. The same day, Visa released a card-native specification. MPP is the third major payment protocol for the agentic economy, alongside ACP and x402. The AXD Institute analyses what MPP solves, what it leaves undesigned, and why the protocol proliferation problem is the defining infrastructure challenge of agentic commerce.

Every commercial metric in agentic commerce reduces to a trust problem. Conversion is delegation willingness. Retention is trust maintenance. Cost-to-serve is trust automation. The AXD Institute examines why trust architecture is not a design concern but the primary economic lever of the agentic economy - and why organisations that treat trust as infrastructure will outperform those that treat it as branding.

The purchase is the easy part. Returns, subscriptions, loyalty programmes, and ongoing service were designed for humans who remember, feel, and can be persuaded. When the customer is an agent, every post-purchase interaction becomes a trust maintenance problem. The AXD Institute examines why post-purchase operations are the most underdesigned layer of agentic commerce.

Agentic commerce does not simply add agents to existing value chains. It restructures them. Platform control points shift. The merchant-of-record question becomes a liability question. Ecosystem lock-in operates through trust infrastructure rather than network effects. The AXD Institute maps how the value chain redesigns itself when the customer is a machine.

Not all agents are the same. A shopping agent that compares prices operates under different delegation constraints than a negotiation agent that makes binding commitments. A budget agent that manages recurring spend requires different trust architecture than a discovery agent that explores new categories. The AXD Institute proposes a taxonomy of commercial agents defined not by capability but by delegation pattern.

Every system in the merchant's technology stack was built for human customers. Order management assumes a person tracks their delivery. CRM assumes someone reads the email. Support assumes a human describes the problem. When the customer is an agent, every integration point becomes a trust interface. The AXD Institute examines the merchant's stack through the lens of delegation design.

Checkout is where every design decision in agentic commerce converges into a single moment. Delegation authority must be verified. Trust thresholds must be evaluated. Mandate constraints must be enforced. Payment protocols must execute. The agent checkout is not a simplified version of human checkout. It is the compression point of the entire AXD framework.

Brand was built for human perception. Colour, typography, tone of voice, emotional resonance - these are instruments of attention. When the customer is an agent, attention is irrelevant. The agent cannot see the logo. It cannot feel the brand story. Brand in the agentic economy is not visual identity. It is trust made legible to machines.

When an agent makes a purchase the human did not intend, who is liable? When an agent negotiates a price the merchant did not authorise, who bears the loss? Liability in the agentic economy is not a legal afterthought. It is a design problem. The AXD Institute examines why delegation design determines where liability falls - and why every design decision is a liability decision.

Consumer surveys consistently show that willingness to delegate to AI agents plateaus well below full autonomy. This is not a technology problem. It is a trust architecture problem. The AXD Institute examines the structural ceiling on consumer delegation and what it means for agentic commerce design.

Mastercard, Visa, Stripe, and a dozen startups are building agent payment protocols simultaneously. Identity verification layers are fragmenting. Agent-to-agent communication standards are proliferating. The AXD Institute maps the convergence - and the collisions - that will define agentic commerce infrastructure.

In a single week in Las Vegas, Google, OpenAI, Shopify, Mastercard, Stripe, and Walmart each made structural commitments that moved agentic commerce from keynote speculation to operational infrastructure. The AXD Institute analyses what happened, what it means, and what the emerging architecture reveals about the future of autonomous shopping.

The agentic age requires eight new design capabilities that have no precedent in traditional UX practice. From intent architecture to governance-aware experience design, this essay names the competencies every Head of Design must build now and maps each to the AXD frameworks that make them operational.

When every party in the B2B2C chain deploys its own agent, the intermediation problem becomes a design problem. Trust must propagate across three or more parties. Delegation must cascade through multiple agent boundaries. This essay examines how B2B2C agentic commerce restructures the value chain and what it demands of AXD.