The news that OpenAI has scaled back its Instant Checkout feature has attracted considerable commentary, much of it framing the move as a setback for agentic commerce. For some observers, it is being framed as evidence that agentic commerce is not yet ready for prime time, or worse, that it has been overhyped from the outset.

That conclusion overlooks the more important signal. What we are seeing is not a retreat from agentic commerce, but a necessary confrontation with the realities of deploying it at scale. The technology itself is advancing rapidly and continues to demonstrate impressive capabilities in controlled environments. The friction emerges when those capabilities are introduced into the complexity of real-world commerce.

Agentic commerce has always promised more than incremental change. It represents a fundamental shift in how purchase journeys start and how they are completed, moving from human-led browsing and decision-making to autonomous, AI-driven execution. That shift places entirely new demands on the infrastructure that underpins digital commerce. It is those demands, rather than any limitation in the AI, that are now coming into sharper focus.

The Hidden Complexity of Autonomous Transactions

At first glance, the concept of an AI agent completing a purchase on behalf of a consumer appears straightforward. In practice, it requires the orchestration of multiple systems that were never designed to operate in this way.

Every transaction sits at the intersection of several critical layers. Product information must be accurate, structured and interpretable by machines rather than humans. Identity must be persistent and verifiable, ensuring that the agent is acting on behalf of a legitimate user. There needs to be a record of what was requested by the user, offered by the merchant, and accepted by the agent. Payment processes must function seamlessly while incorporating fraud detection, authentication and compliance checks. Tax calculations must be applied correctly across jurisdictions. Liability must be clearly defined in the event of disputes, errors, or misuse.

Each of these components is complex in isolation. Bringing them together into a single, cohesive flow that can operate autonomously, in real time, and at scale is an entirely different challenge. It is this level of coordination that exposes the structural limitations of today’s commerce infrastructure.

Fragmentation as the Core Constraint

One of the most significant barriers to scaling agentic commerce is the lack of standardisation across the ecosystem. Communication between AI platforms and merchant systems depends on protocols that are often inconsistent, incomplete, or entirely absent.

Industry research reinforces this point. A recent report from Adyen, drawing on insights from more than 200 enterprise merchants and direct work with AI commerce platforms, highlights a set of structural constraints that continue to limit progress. These include fragmented protocols, product data that is not designed for machine consumption, the weight of legacy enterprise stacks, unclear trust and liability frameworks and the operational challenge of onboarding merchants at scale.

Without common standards, every integration becomes a bespoke exercise. This limits scalability and increases the likelihood of failure when systems encounter unexpected conditions. What works in a controlled demonstration can quickly break down when exposed to the variability of real-world environments.

Product data presents a parallel challenge. Most product catalogues today are designed for human consumption, not machine interpretation. Descriptions may be unstructured, attributes inconsistent, and availability data unreliable. For an AI agent attempting to make an autonomous purchasing decision, these inconsistencies introduce uncertainty and risk.

Legacy technology further compounds the problem. Many enterprise commerce platforms have evolved over decades, resulting in complex, layered architectures that are difficult to adapt. Integrating new capabilities into these environments is rarely straightforward, particularly when those capabilities require real-time coordination across multiple systems.

Taken together, these factors create a level of fragmentation that significantly constrains what agentic commerce can achieve today. The gap between theoretical capability and practical deployment is, to a large extent, a function of this fragmentation.

Where Payments Can Lead

Not all parts of the ecosystem face the same degree of difficulty. The payments layer, in particular, is relatively well positioned to make meaningful progress in the near term.

Payments is an industry that has long operated at scale, managing risk, complexity and coordination across multiple stakeholders. Challenges such as fraud detection, transaction routing and cross-border flows are not new. They have been addressed through a combination of shared standards, network effects and ongoing collaboration between participants.

This experience provides a foundation for tackling similar challenges within agentic commerce. Payment optimisation, dynamic risk assessment and automated tax handling are areas where progress is both necessary and achievable. The incentives are aligned, and the tools already exist in various forms.

That said, progress within payments alone is not sufficient to unlock the full potential of agentic commerce. It can reduce friction within the transaction layer, but it cannot resolve issues related to product data quality, identity management, or the structural complexity of merchant systems. These challenges sit beyond the traditional boundaries of payments and require broader coordination.

The Harder Problems Beneath the Surface

Some of the most critical barriers to agentic commerce are also the most deeply embedded. Product data, for example, is foundational to the entire experience. Without accurate, structured, and up-to-date information, AI agents cannot make reliable decisions. Improving data quality at scale requires changes in how merchants create, manage, and maintain their catalogues. It is not simply a technical issue, but an operational and human one.

Identity presents a similarly complex challenge. For an AI agent to act on behalf of a consumer, there must be a robust mechanism for verifying that authority. This involves not only authentication, but also ongoing trust management. Questions around consent, delegation, and revocation must be addressed in a way that is both secure and user-friendly.

Liability adds another layer of complexity. In a traditional transaction, responsibility is relatively clear. In an agent-driven model, it becomes less so. If an AI agent makes an incorrect purchase, who is accountable? The consumer, the merchant, the platform, or the provider of the AI? Resolving these questions requires legal, regulatory and commercial alignment, which takes time to develop.

Finally, there is the challenge of onboarding at scale. For agentic commerce to function effectively, large numbers of merchants must be integrated into the ecosystem. This involves not only technical integration, but also alignment on standards, processes and expectations. Achieving this level of coordination across a diverse and global merchant base is a significant undertaking.

Misinterpreting Merchant Caution

It is easy to misread the current state of the market. Merchant hesitation around enabling agent-driven transactions is sometimes interpreted as a lack of demand. In reality, the opposite is more likely to be true.

Consumers have consistently demonstrated a strong appetite for convenience. The idea of delegating routine purchasing decisions to intelligent systems aligns closely with broader trends in digital behaviour. From subscription services to automated reordering, the trajectory is clear.

Merchant caution reflects a different concern. It is rooted in the recognition that existing systems are not yet equipped to support agentic commerce at the level of reliability and security required. The risks associated with premature deployment are real, ranging from failed transactions to fraud exposure and reputational damage.

Understanding this distinction is critical. It shifts the focus away from trying to stimulate demand and towards addressing the underlying constraints that are holding the ecosystem back.

Building the Foundations for Scale

The path forward for agentic commerce is not defined by incremental feature development, but by the construction of foundational infrastructure. This includes the development of standardised communication protocols that enable interoperability between systems. It requires significant improvements in the quality and structure of product data, making it suitable for machine consumption.

It also involves the creation of robust identity frameworks that support secure and transparent delegation of authority to AI agents. Alongside this, clear and consistent approaches to liability must be established, ensuring that all participants understand their roles and responsibilities.

Equally important is the development of scalable onboarding mechanisms that allow merchants to participate in agentic ecosystems without prohibitive complexity or cost. This is essential for achieving the network effects that will ultimately drive adoption.

None of these elements can be delivered in isolation. They require coordinated effort across multiple stakeholders, including merchants, platforms, payment providers, and regulators. Progress will depend on collaboration, standardisation and a willingness to invest in long-term structural change.