trending
For three decades, e-commerce has relied on a familiar grammar: a search bar, a product page, and the decisive click on a “buy now” button.
Agentic AI: Prompts replace buy buttons
But the rise of agentic AI suggests this model may soon look outdated. In its place, natural language prompts — simple requests such as “Find me the best running shoes under £100 and deliver them tomorrow” — could trigger entire transactions, executed seamlessly in the background.
This evolution, explored in the latest Prompt Economy Tracker in collaboration with Visa, suggests commerce is moving away from clicks and taps toward a conversational mode of engagement.
The consequences for retailers, payment providers, and consumers could be profound.
From Fetching Answers to Taking Action
The defining shift between traditional AI and its agentic successor lies in autonomy. Generative AI has already demonstrated how machines can produce text, images, or recommendations.
Agentic AI goes further: it takes initiative, makes decisions within pre-set rules, and executes tasks. Where an older system might provide a list of hotels, an agentic one can reserve the room, pay for it, and confirm the booking — all triggered by a single prompt.
This distinction is critical.
In commercial terms, it means that the buy button is no longer the ultimate conversion point. Instead, intent is collapsed into action.
Amazon’s Rufus tool already guides shoppers to purchases using open-ended prompts, while Klarna reports that its AI assistant resolves the majority of customer queries within minutes without escalation.
Living With an Agent
The Tracker illustrates this new dynamic through the fictional day of “Taylor,” a payments professional who relies on agentic AI not as an assistant, but as an autonomous partner.
Her AI reconciles invoices overnight, prepares regulatory analysis before meetings, and drafts communications for her team.
In essence, the agent anticipates what needs to be done and executes accordingly, leaving Taylor to focus on higher-value strategic tasks.
The lesson for commerce is clear: speed, compliance, and efficiency can no longer depend solely on human bandwidth.
Instead, systems will increasingly “do” rather than merely “suggest.”
The Technical Foundations
Such a leap is only possible because of a rapidly maturing technical stack. Agentic AI builds on large language models, but adds orchestration tools, vector databases, and compliance guardrails.
The crucial innovation is interoperability. Just as USB-C standardised how devices connect, the Model Context Protocol (MCP) is standardising how agents plug into data, applications, and payment rails.
Backed by Visa, Microsoft, OpenAI, and Anthropic, MCP could allow agents to transact across ecosystems without friction.
This matters because commerce is not a single silo. A booking might involve airlines, hotels, payments networks, and loyalty schemes.
The promise of interoperability is that agents can stitch these together invisibly, delivering a finished outcome to the user in seconds.
Early Examples in the Wild
The seeds of this shift are already visible.
Klarna’s service agent now handles two-thirds of queries across its 23 markets. Instacart’s “Ask” tool can turn a recipe into a ready-to-order shopping basket.
Walmart is trialling “Super Agents” that can reorder household essentials, summarise reviews, and schedule deliveries.
These are not pilots of marginal interest; they are real-world signals of an accelerating trend.
For consumers, this reduces the need to browse, click, and compare. For retailers, it raises new questions: how do you surface products to AI agents, rather than to human shoppers?
Visibility and trust may matter as much to machines as they do to people.
The Strategic Implications
Replacing the buy button with prompts is not merely a technical novelty. It could reshape the competitive landscape.
Retailers may need to optimise for “agent discoverability” in the same way they once optimised for Google search. Payment providers will have to embed security and compliance into automated flows where card numbers may never be seen by the end user.
Crucially, adoption will hinge on trust.
Consumers may happily let an agent reorder groceries, but will they allow it to manage investments, negotiate supplier contracts, or approve major purchases?
Building transparency, control, and clear override mechanisms will determine whether agentic commerce scales from niche tasks to mainstream behaviour.
Commerce Without Waiting
The direction of travel, however, appears irreversible. The Tracker argues that autonomy in commerce is not optional but inevitable.
The firms that succeed will be those that embed agentic AI into their operations early — not as a gimmick, but as an architectural shift.
In such a world, the buy button becomes a relic of an earlier age. Commerce will not wait for the shopper’s click. It will happen at the speed of a command.
Comments