Skip to main content
Agentic Commerce and the Maturity Problem
AI Insights

Agentic Commerce and the Maturity Problem

March 6, 2026

The recent decision by ChatGPT to withdraw its agentic commerce functionality has quickly produced a familiar conclusion: the model may not be viable after all.

Such interpretations, however, may reveal more about the current maturity of digital commerce infrastructure than about the capabilities of artificial intelligence itself.

Agentic commerce was never likely to emerge fully formed within an ecosystem that remains only partially prepared for machine-driven transactions.


The Promise and the Reality

Agentic commerce has become one of the most discussed ideas at the intersection of artificial intelligence and digital retail. The concept appears straightforward enough. Instead of merely assisting consumers with information, intelligent systems would actively perform parts of the purchasing process. They would search, compare, evaluate and eventually execute transactions on behalf of users. It is, one might say, the logical extension of every recommendation engine that has ever told someone they might also enjoy a different pair of shoes.

In theory, the technological foundation for such behaviour is already visible. Large language models can interpret complex instructions, evaluate alternatives and interact with structured digital systems. From a purely technical perspective, the emergence of purchasing agents therefore seems plausible. The question, as usual, lies in the distance between "plausible" and "actually happening."

And yet, the practical reality remains more restrained than the conference keynotes would suggest.

Recent commentary has interpreted ChatGPT's decision to effectively discontinue its agentic commerce features as evidence that the model itself may not work. These conclusions often arrive remarkably quickly, particularly within an industry that has grown accustomed to evaluating technological innovations within time horizons that would make a mayfly feel vindicated.

A closer look suggests a rather different interpretation.

Infrastructure, Not Intelligence

The difficulty lies less in the capabilities of artificial intelligence and more in the structural maturity of the commerce ecosystem into which these systems are expected to operate. This is not a glamorous explanation, which is presumably why it receives less attention than it deserves.

Agentic commerce requires an environment that is far more machine-readable than much of today's retail infrastructure currently provides. In order for an autonomous system to execute a transaction reliably, several conditions must be met. Product data must be structured and consistently formatted. Pricing logic must be accessible through stable interfaces. Inventory availability must be transparent. Checkout processes must be predictable and automated rather than dependent on the kind of human interaction that typically involves clicking through three consent banners and locating a reCAPTCHA.

Many digital storefronts were never designed for such conditions.

E-commerce platforms evolved primarily to serve human shoppers. Product pages are optimised for visual browsing rather than structured interpretation. Checkout flows often rely on interactive elements that are difficult for automated systems to navigate. Authentication requirements and fragmented data standards introduce additional friction. In fairness, most of these systems were built to delight humans, or at least to prevent them from abandoning their carts, which is not quite the same thing.

For human users these obstacles are manageable. For autonomous agents they can become prohibitive.

In this sense, the current hesitation around agentic commerce resembles a pattern observed repeatedly in the history of technological change. New capabilities often emerge before the surrounding infrastructure has matured sufficiently to support them. The automobile arrived before the motorway. The smartphone arrived before mobile-optimised websites. The pattern is familiar enough to be unremarkable, except to those who happen to be living through the current iteration.

The result is a temporary mismatch between possibility and environment.

Consumers themselves rarely articulate demand for systems that would fundamentally change the mechanics of purchasing. Surveys typically reflect existing habits rather than future behavioural shifts. Henry Ford's well-known observation about faster horses captures this dynamic with unusual clarity. People tend to imagine improvements within familiar frameworks, not the transformation of the frameworks themselves. One might argue that the same principle applies to most corporate strategy sessions, though that is a discussion for another day.

Artificial intelligence now introduces a comparable shift in the informational environment of commerce.

As explored in the essay When Signals Multiply, machine intelligence dramatically increases the number of detectable patterns within complex systems. What once appeared as noise can suddenly emerge as a signal. The challenge for organisations then shifts from discovering information to interpreting it. A shift that tends to be rather less comfortable than the preceding phase of simply not knowing.

Agentic commerce represents one possible extension of this shift.

If artificial intelligence can identify products, evaluate options and assess contextual preferences, it may eventually perform parts of the purchasing process itself. Yet such delegation of judgement requires a level of structural alignment that the current commerce landscape rarely provides. The technology, one might say, is ready to operate in an environment that has not yet fully agreed to receive it.

The Advertising Paradox

This also helps explain an apparent contradiction that observers have noted.

Some AI platforms appear willing to experiment with advertising integrations while simultaneously stepping back from agentic purchasing functionality. At first glance this may seem inconsistent. Advertising introduces commercial influence, while agentic commerce would simply automate transactions.

In practice the two models occupy very different positions within the decision architecture of digital markets.

Advertising leaves the final judgement with the user. A system may surface recommendations, but the act of selection remains human. Responsibility and agency therefore remain clearly located, which is a polite way of saying that if you buy something dreadful after seeing an advert, that is largely considered your own affair.

Agentic commerce, by contrast, moves the platform itself into the transactional chain. The system would not merely suggest products. It would evaluate merchants, prioritise options and potentially execute purchases. The distance between "you might like this" and "I have purchased this on your behalf" turns out to be rather more consequential than it first appears.

Such a shift transforms the platform from an informational layer into an infrastructural actor within the market itself.

This distinction has important implications.

When a system influences a decision, questions of neutrality and transparency arise. When a system performs the decision, those questions become far more consequential. Responsibility for outcomes becomes harder to attribute. Governance becomes more complex. And the legal profession, one imagines, becomes rather more interested.

Where Agentic Commerce Will Emerge First

Seen from this perspective, the slow emergence of agentic commerce appears less surprising than the expectation that it should have arrived more quickly.

What we may be observing is not the failure of an idea but the gradual maturation of the conditions required for it to function.

Artificial intelligence continues to expand the informational capacity of organisations and markets. Yet structural adaptation tends to occur more slowly. As discussed in The Incentive Paradox in AI Transformation, institutional systems often require time to adjust their incentives and decision processes to new technological capabilities. Patience, it seems, remains an undervalued strategic competence.

Commerce infrastructure is no exception.

Agentic commerce will likely appear first in environments where digital processes are already highly structured. Travel booking systems, software subscriptions and certain B2B procurement platforms already operate with standardised data models and automated transaction flows. In such contexts, intelligent agents may integrate relatively smoothly. The open retail web, with its glorious diversity of checkout experiences and cookie banners, presents a far more complex environment.

Here the transition will depend not only on advances in artificial intelligence but also on the continued evolution of digital commerce infrastructure itself. As explored in The Quiet Redistribution of Judgement, such transitions reshape not merely technology but the organisational conditions under which commercial decisions are formed and exercised.

Agentic commerce may therefore arrive more gradually than early narratives suggested. But gradual emergence should not be confused with failure. The two are related in much the same way that a slow train is related to a cancelled one, which is to say, not very much at all.

More often it simply reflects the pace at which ecosystems mature around the technologies that eventually reshape them.


Explore more essays on organisational maturity in the age of AI in the Decision Architecture series on ThePromptMonk.

Related Articles