Agentic shopping: the accelerating ascent of a-commerce

Executive summary: The agents are coming. With the emergence of software tools that can make decisions and transact purchases, shopping will be revolutionised. In the new world of agentic a-commerce, there will be a shift in power. Value will move from consumer attention and branding to data quality, pricing transparency and fulfilment reliability. Agents will optimise for value and not impulse. Revolutions do not happen overnight. Building trust and familiarity take time. Delegating purchasing authority to software is both psychologically and operationally hard. Nonetheless, by 2030, the US agentic commerce market could be worth up to $400bn in gross merchandise value, and China’s substantially more. Leading retailers such as Amazon and Walmart in the US as well as Alibaba and JD.com in China have already successfully launched agentic shopping assistants. More will follow. The ultimate winners, however, may be the enablers, or the providers of payments and financial rails as well as logistics and fulfilment leaders.

For many people – your author included – shopping is a chore. Agents will make it less so. Think of agents as representing the third wave of how commerce will be transformed. Beyond bricks and mortar, first we had desktop e-commerce, then mobile purchases and now the emergence of agents. Call it the agentic revolution, or a-commerce. The debate is moving from where we shop to who does the shopping.

The promise of a-commerce is that all consumers could have their own personalised digital interactive shopper. Give it specific instructions and the agent will carry them out. Agentic transactions herald “a new era of commerce”, according to the Chief Executive of American Express. Andy Jassy of Amazon believes that agentic commerce solutions will “expand the amount of shopping done online.” In the words of Mastercard’s CEO, “the [agentic] train is leaving the platform.” If so, then both consumers and businesses will need to adapt.

An agent, in its most mundane form, can be thought of as simply a piece of software that can understand what you want and take decisions. Agentic commerce, then, is a form of retail where AI agents do the buying on behalf of consumers or businesses. This could include discovering products, comparing options, negotiating terms, executing payment and managing reorders with minimal human input. A-commerce represents a practical application or use case for artificial intelligence, unlocking a new vector of growth for the industry.

Today’s non-agentic e-commerce shopping journey is fragmented and complex. Imagine instead of having to browse Amazon or book travel, an AI agent could be given an objective and constraints (such as price, brand, delivery time, sustainability, policy rules). The agent would then search for and evaluate options across merchants, make selections and complete the purchase autonomously. AI agents would have the ability to draw upon users’ preferences and their history combined with existing reviews and live merchant data feeds.

Agents can perform tasks as mundane as real time and continuous price comparisons, or making automated recurring purchases, such as grocery orders. There would be no longer any need to wait in long online queues to purchase sought-after tickets for a concert or sporting fixture – you could instruct an agent to buy them on your behalf; stipulating things such as seating preferences and maximum spend. Planning a holiday? Not only could the agent help arrange your travel plans, but it could also source hotel stays, book restaurants, buy tickets for major attractions and so on – and all from different merchants. In all the above cases, complexity is removed for the user.

Agentic commerce is very different to traditional e-commerce. In both, the customer is still human, but in an a-commerce world, the decision maker and transactor are software. Agents should not be thought of as super-concierges – which are often reactive and advisory. They do not (need to) recommend, but simply act, with delegated authority and at machine speed.

In the new world of commerce, four big changes look set to occur. There will be a shift in power. Value will move from consumer attention and branding to data quality, pricing transparency and fulfilment reliability. There will be lower friction too. Purchase decisions will become faster, more frequent and more price-rational. Since agents will optimise for value and not impulse, this will impact retailer margins. Marketing budgets would suffer in favour of scale and efficiency. New toll collectors will emerge. Platforms that host agents, control identity, payments or procurement rules can stand to capture meaningful economics.

None of this will happen overnight. Consumers may have become more comfortable with using AI assistants to streamline tasks and receive personalised recommendations. However, even by 2030, only 40% of consumers globally say they expect to use AI for comparison shopping according to a PWC survey conducted last year. But when it comes to making purchases, there are still some barriers. While 33% of respondents say they are comfortable with an AI agent planning their grocery shopping, only 25% would let AI place the order. Just 22% state that they would allow an agent to make a purchase. Many consumers still need reassurance on safety or that AI is prioritising their interests.

For agentic commerce to work practically a full tech stack needs to be put in place. This comprises not just an agentic commerce surface, or user interface, but also several mutually supporting protocols. Think of these as rails upon which agents operate and rules that govern communications and messaging between systems, asking whether the agent can be trusted, whether it is allowed to pay, and how it checks out.

The front-end will capture user intent, apply user-defined rules and orchestrate the shopping and checkout journey on the user’s behalf. A trusted agent protocol will then help merchants to identify who the agent is and whether it is legitimate. An agent payments protocol will help prove what the agent is authorised to do financially, within defined user mandates. Finally, an agentic commerce protocol will execute the payment over existing rails, handling authorisation, clearing, settlement and operations.

China is arguably ahead of the US in practical adoption of agentic commerce, albeit framed less as a standalone concept and more as an extension of super‑apps. Agent-like functionality exists in the form of automated re‑ordering, smart recommendations, one‑click execution. Adoption has been fastest in low‑friction, high‑frequency categories such as food delivery, local services, travel, in‑app retail and financial services. Compared with the US, China benefits from tighter platform integration, ubiquitous digital payments, weaker consumer resistance to automation and fewer legacy intermediaries.

In the western world, younger digital natives are likely to be the earliest adopters. First purchases may be utility-like products and then more emotional transactions. The gross merchandise value (GMV) of American a-commerce purchases is set to grow from 1% of digital purchases in 2026 to 10% by 2030, according to Morgan Stanley. GMV grows from $8bn to $192bn over this period, equivalent to a compound annual growth rate of over 100%. Other forecasters believe the market could be worth up to $400bn by the end of this decade. For context, China’s a-commerce GMV is already estimated to be over $1tr.

Consumers are experimenting already. Amazon and Walmart, America’s two largest retailers, are in the vanguard. Around 300m customers used Amazon’s Rufus agent last year. Rufus is an AI shopping assistant that helps customers research products, compare options and decide what to buy directly within the Amazon app. Amazon says that shoppers who engage with Rufus are 60% more likely to complete a purchase than those who do not use it. The retailer estimates that Rufus generated $10-12bn of incremental annualised sales in 2025. Last month, Amazon took the decision to fold Rufus into Alexa to create a single, unified AI assistant across its platform. All the AI models underpinning Rufus have now been embedded into Alexa.

Walmart launched a similar agentic assistant, called Sparky, in June last year. Around half of its customers have tried it, per the company and those who do generate c35% higher average order value than non-Sparky users. Elsewhere, Macy’s has said that users of its new AI-powered ‘Ask Macy’s’ chatbot are spending 400% more relative to non-users, early testing suggests.

In Europe, Zalando Assistant has led to at least a 20% increase in product clicks and more than a 40% increase in items added to customer wishlists, according to the retailer. Other businesses including CarMax, Carrefour, DoorDash, Instacart, Kroger and Lowe’s are also trialling agentic services. Alibaba’s Wenwen service and JD.com’s Jingyan smart shopping assistant are among the most successful examples of agentic services in China, although limited public disclosure is available.

All the above are critically being supported by robust protocols such as UCP and ACP Mechanics. The Universal Commerce Protocol is an open standard developed by Google that it describes as a “common language” that allows AI agents and merchant commerce systems to work together across the full consumer shopping journey. Multiple payment processors (including Adyen, American Express, Mastercard and Stripe) and retailers (Best Buy, Shopify, Target, Walmart) are adopting UCP. Individuals can use the protocol to check out on eligible Google listings in Search AI Mode and Gemini using Google Pay. ACP Mechanics, developed by OpenAI and Stripe, follows a similar approach that powers ‘instant checkout’ by letting AI agents complete purchases inside ChatGPT. Microsoft has also launched Copilot Checkout feature.

At the same time, the major payment processors have launched B2B enterprise offerings aimed at building, testing and deploying fit-for-purpose agents. Mastercard’s Agent Suite combines technical support with customisable AI agents, leveraging Mastercard’s extensive payments expertise, data-fuelled insights, proprietary technology platforms, and 4,000 global advisors. Visa has a similar offering called Intelligent Commerce, while American Express has released its ACE (Agentic Commerce Experiences) developer kit.

Everything, however, is nascent. “We’re not even in the first innings” said the Chief Executive of American Express in April, while “it’s such early days” with a “big change processes” required, noted Rene Haas, the Chief Executive of ARM Holdings, the chip designer, last month. Think of agentic commerce (or agents more generally) as being arguably where the web was 30 years ago. Lots of bold ideas and acronyms are being thrown around, but building trust and familiarity take time.

The delegation of purchasing authority to software is both psychologically and organisationally hard. For many the perceived loss of control – allowing an agent to commit spend without human review – may be uncomfortable. Consumers may legitimately wonder what happens were an agent to make a poor decision. With whom does the accountability lie: the vendor, the platform, the model or the user? Equally, consider the issue of auditability. If agent decisions are probabilistic and non-deterministic, then they may be hard to explain.

Real world complexity will continue to exist, even in a world of agents. Factors such as product attributes, delivery reliability and after-sales services are often inaccurate and non-standardised. Agents may therefore struggle. Further, certain categories such as luxury, travel and lifestyle rely on discovery and serendipity. This looks unlikely to change overnight. Agents operating in any retail category are also high value targets for attackers. Credential abuse constitutes one consideration. Equally, malicious actors could adversely influence agent behaviour via data poisoning or adversarial inputs.

Ultimately, consumer behaviour may prove stickier than expected. Practically for a-commerce to be a success, there needs to be multi-sided adoption by consumers, merchants, platforms and regulators simultaneously. Against this background, progress may be incremental rather than disruptive, and only in narrow categories. It is an incorrect assumption to believe that just because agents can buy, it does not mean that they will be allowed to buy.

The rise of agentic commerce will create inevitable winners and losers. However, the story will be less about agents themselves and more about a reweighting of value towards infrastructure, execution and trust. Do not forget that agents will optimise for reliability over brand, predictability over optionality and certainty over choice. Agents will need confidence in merchant legitimacy, inventory accuracy, fulfilment performance and recourse if something goes wrong.

Against this background, likely beneficiaries could emerge from one of four major areas: payments and financial rails, logistics and fulfilment leaders, cloud and AI infrastructure, and marketplaces that choose quickly to pivot towards being agent native. Think of the former – companies such as Mastercard and Visa – as being de facto toll collectors. Agents should increase transaction velocity and frequency rather than disintermediate payments processers. Card payment networks benefit from volume growth regardless of who initiates the transaction. Embedded controls such as identity and fraud management should only become more valuable in an agent‑driven world. Both major payment networks currently enjoy 99.9999% uptime on their core platforms with less than 0.1% fraud rates.

If agents optimise for reliability, speed and cost, then scaled, high‑quality fulfilment networks such as Amazon, DHL and UPS as well as the owners of critical adjacent infrastructure (Prologis, for example), should benefit. Last‑mile execution may become more important than brand storytelling in an agentic world. At the same time, since agents are compute‑intensive and require persistent inference, orchestration and data access, then the major cloud and AI infrastructure players (AWS, Google, Microsoft, and NVIDIA by default) also look well-placed.

As the world becomes increasingly agentic, some businesses will clearly face challenges. If we accept that agents can reduce the value of brand awareness and impulse-driven discovery, then advertising-led consumer brands – whether Nike and Adidas or Unilever and Procter & Gamble – may be at risk. Marketing spend becomes less effective, since agents ignore persuasion. Similarly, retailers that are reliant on browsing and footfall – department store owners or general retailers – may suffer. Inventory and fulfilment trump store experience in the agentic age. Price opaque intermediaries such as traditional travel agents or comparison sites could be at risk too.

Even if agentic adoption does take time, the agents are coming. It is up to both retailers and consumers to embrace them. Watch this space. In the future, you may never have to visit a shop or waste time on a website again.

—

Originally Posted on June 23, 2026 – Agentic shopping: the accelerating ascent of a-commerce

The above does not constitute investment advice and is the sole opinion of the author at the time of publication. Heptagon Capital is an investor in American Express, Mastercard and Prologis. The author of this piece has no personal direct investment in the business. Past performance does not predict future returns, the value of investments and income from them can fall as well as rise.

Disclaimers

The document is provided for information purposes only and does not constitute investment advice or any recommendation to buy, or sell or otherwise transact in any investments. The document is not intended to be construed as investment research. The contents of this document are based upon sources of information which Heptagon Capital LLP believes to be reliable. However, except to the extent required by applicable law or regulations, no guarantee, warranty or representation (express or implied) is given as to the accuracy or completeness of this document or its contents and, Heptagon Capital LLP, its affiliate companies and its members, officers, employees, agents and advisors do not accept any liability or responsibility in respect of the information or any views expressed herein. Opinions expressed whether in general or in both on the performance of individual investments and in a wider economic context represent the views of the contributor at the time of preparation. Where this document provides forward-looking statements which are based on relevant reports, current opinions, expectations and projections, actual results could differ materially from those anticipated in such statements. All opinions and estimates included in the document are subject to change without notice and Heptagon Capital LLP is under no obligation to update or revise information contained in the document. Furthermore, Heptagon Capital LLP disclaims any liability for any loss, damage, costs or expenses (including direct, indirect, special and consequential) howsoever arising which any person may suffer or incur as a result of viewing or utilising any information included in this document. 

The document is protected by copyright. The use of any trademarks and logos displayed in the document without Heptagon Capital LLP’s prior written consent is strictly prohibited. Information in the document must not be published or redistributed without Heptagon Capital LLP’s prior written consent.