When purchasing is delegated, persuasion gives way to proof.
Commerce has long been organised around a relatively simple assumption: firms market to people. They compete for attention, shape preferences, reduce uncertainty, establish a reputation and make the path to purchase as easy as possible. Product packaging, advertising, brand identity, retail placement, search ranking and loyalty programmes all respond to this basic fact.
That assumption is beginning to change.
AI agents are not yet replacing consumers at scale, and many claims about autonomous purchasing are ahead of current practice. But the direction of travel is clear. Individuals and organisations are increasingly using software to compare products, renew subscriptions, book travel, source suppliers, manage inventory, analyse contracts and recommend purchases. As those systems gain authority to act within defined limits, parts of commercial decision-making will shift from human attention to machine-mediated selection.
The distinction matters. A human buyer may be influenced by a slogan, a familiar logo, an aspirational image or the convenience of a well-designed checkout page. An agent acting for that buyer will work differently. It may compare specifications, prices, delivery windows, contractual terms, return conditions, service history, compatibility, accessibility, emissions information or verified customer outcomes. It may be instructed to avoid certain firms, favour local suppliers, observe a spending limit, exclude products with particular ingredients or give greater weight to durability than initial price.
Marketing will not disappear. But it will need to become more evidentiary.
The future question for many firms will not be only, “How do we persuade the customer?” It will be, “How do we make our offer legible, comparable and credible to the system acting for the customer?”
A new principal-agent problem
The phrase “agent commerce” describes transactions in which an AI system takes some role in searching, comparing, recommending, negotiating or completing a purchase on behalf of a person or organisation. The degree of delegation may vary considerably.
At one end, an agent may simply summarise options. A consumer asks for a comparison of mobile plans or accommodation. The system identifies relevant alternatives, explains trade-offs and leaves the final decision to the user. At the other end, an agent may have standing authority to act. It might renew a software subscription if the price remains within an agreed range, reorder office supplies from approved vendors or purchase a flight that meets fixed timing and budget constraints.
This creates a new version of an old economic problem. The principal, whether a person, household or firm, wants the agent to act in their interests. But the agent may be influenced by the information it can access, the commercial arrangements of the platform on which it operates, the objectives set by its provider or the narrowness of the instructions it receives.
Human consumers face familiar forms of manipulation and information asymmetry. AI agents face a different set of risks. A system may favour suppliers whose data is easiest to read. It may overvalue a low price while underweighting quality or service. It may give prominence to vendors that have paid for placement. It may fail to distinguish a verified claim from an unverified one. It may make a technically coherent choice that conflicts with the user’s preferences because those preferences were poorly specified.
Agent commerce will therefore depend on delegation with constraints. The central design challenge is not simply creating a capable agent. It is creating one that can act within a clear mandate, explain the basis of its decision and be reviewed when its choice is contested.
The market moves from attention to structured information
In a human-centred market, firms spend heavily to gain attention. The scarcity is cognitive. People have limited time, limited capacity to compare options and limited appetite for reading terms and conditions. Branding and advertising help firms simplify a choice, create familiarity and establish an emotional association with a product.
An agent does not have the same limits. It can review a far larger number of offers, compare product attributes systematically and check conditions that most people would not have time to read. This does not make the market perfectly rational. It changes the terms on which competition occurs.
For many products and services, firms will need to provide structured, current and verifiable information. Prices will need to be clear. Product data will need to be consistent. Availability, delivery, compatibility, warranties, service levels and return conditions will need to be machine-readable. Claims about sustainability, quality, safety or performance will need supporting evidence if they are to be weighted meaningfully in a comparison.
This is likely to make some forms of marketing less effective. Vague superlatives, ambiguous discounts and hidden conditions may perform poorly when agents are instructed to compare total cost, product quality and contractual terms. A system can identify that a lower advertised price is offset by delivery charges, restrictive cancellation conditions or a weaker warranty.
At the same time, agent commerce may produce new ways to manipulate choice. Firms may learn to present information in forms that systems rank favourably while providing limited substantive value to the buyer. They may optimise their data feeds for particular recommendation systems, much as firms have optimised web pages for search engines. Product descriptions may be written as much for a model’s interpretation as for a customer’s understanding.
The issue will not be whether marketing becomes less important. It will be whether marketing becomes more accountable.
Search is no longer the same market
The growth of search engines created a large industry around search visibility. Firms sought prominence through search-engine optimisation, paid placement and content designed to appear relevant to a query. Agent-mediated purchasing may alter this market further.
A conventional search result gives a person a list of options. The person can compare sources, recognise familiar brands and make a judgement about credibility. An agent may instead receive product feeds, application-programming interfaces, supplier databases and commercial listings, then return a short recommendation or act directly.
This creates significant power for the systems that sit between buyer and seller. The interface may decide which sellers are visible, what information is presented, how trade-offs are ranked and whether sponsored results are distinguishable from independent recommendations. A small number of agent platforms could become important commercial gatekeepers.
The economic question is therefore not only whether agents reduce search costs. They may also increase dependence on a new class of intermediary.
Intermediaries are not inherently harmful. Good intermediaries can lower transaction costs, improve matching and reduce the burden on consumers and firms. But their incentives matter. If a purchasing agent is rewarded by commissions, preferred-provider arrangements or access fees, users should know that. If a platform gives its own products or partners preferential treatment, that preference should be visible. If an agent is unable to assess a class of suppliers because their information is inaccessible, that limitation should be disclosed.
Transparency will be central to trust.
Brands will still matter, but for different reasons
It would be premature to declare the end of brands. People do not purchase only through calculation. They seek reassurance, identity, familiarity and an expectation of quality. Brand reputation can also compress information. A trusted provider may be preferable to an unknown supplier even where the latter offers a slightly lower price.
Yet branding will face a stronger evidentiary test in agent-mediated markets.
A brand that is trusted because it has delivered reliable service, honoured its commitments and treated customers fairly will have a form of reputation that agents can use. A brand that depends primarily on visibility, ambiguity or the difficulty of comparing alternatives may be more exposed.
This may be particularly relevant in business-to-business markets. Procurement teams already value documented performance, service levels, compliance, integration capability and continuity of supply. Agents may make these requirements easier to assess across a wider set of suppliers. Firms with reliable records and clear product information may become more competitive, even if they are not the most visible in conventional advertising.
The strategic task is to make trust operational. This includes accurate documentation, meaningful service commitments, transparent pricing, clear escalation processes and evidence that can be checked. Brand will continue to matter, but it will need a stronger foundation than recognition alone.
Product design becomes market design
Agent commerce also raises questions about the design of products and markets. A product that is difficult to compare may have benefited from consumer inattention. An agent can weaken that advantage by comparing full cost, interoperability, quality and ongoing obligations.
Firms may respond by making products more modular, more transparent and easier to evaluate. Consumers may benefit where this leads to clearer terms and stronger competition on quality. However, firms may also attempt to make comparison more difficult, through proprietary formats, bundled services, restrictive data access or complex pricing structures.
This is where standards and regulation may become important. Markets work better when buyers can compare meaningful information. Common product identifiers, transparent fees, interoperable data formats and clear disclosure of commercial relationships can reduce information asymmetry. The aim should not be to prescribe every feature of agent commerce, but to support fair comparison and informed delegation.
There will also be questions about liability. If an agent purchases the wrong product, accepts an unfavourable term or fails to identify a material risk, who is responsible? The user who delegated authority? The provider of the agent? The platform that ranked options? The supplier whose information was incomplete or misleading?
These questions will need more than technical answers. They concern consumer protection, contract law, competition policy and the distribution of risk between large platforms, sellers and buyers.
Marketing to an agent requires knowing the human behind it
The most important point is that agent commerce should not become a way of removing people from markets. The agent acts for someone. It should serve their interests, values and constraints, rather than reduce every choice to the cheapest available option.
A household may value price, but also reliability, labour standards, local employment, privacy, accessibility or environmental performance. A business may value short-term cost, but also supply resilience, security, integration and long-term partnership. A public agency may have obligations around equity, transparency and public value that cannot be reduced to a procurement score.
Good agents will make such preferences more actionable. They will allow buyers to state priorities with greater precision and apply them consistently across many choices. This may improve markets by making previously overlooked values visible in purchasing decisions.
But only if the preferences belong to the buyer.
The risk is that commercial platforms will shape the mandate before the buyer has meaningfully expressed it. Defaults matter. Ranking criteria matter. Information access matters. A system designed around the commercial priorities of an intermediary may appear to serve the user while quietly directing demand elsewhere.
Agent commerce will be most valuable when it expands the capacity of people and organisations to make considered choices, rather than simply creating a more efficient channel through which attention and purchasing power can be sold.
The firms that do well in this market will not only learn how to market to machines. They will learn how to provide the evidence, transparency and service quality that make a machine acting for a person willing to recommend them.
