Agentic AI in e-commerce is software that can carry out a shopping task from start to finish on a customer’s behalf. It searches, compares, decides and, with permission, pays. AI agents in e-commerce are now moving from suggesting products to buying them, and that raises a question every retailer should answer: who owns the customer relationship when software does the shopping?
This guide explains what agentic AI is, why adoption is slower than the headlines suggest, where the real benefits and risks sit, and how a business can prepare. It uses DoorDash’s latest earnings disclosure as evidence, and it ends with a practical test for deciding which purchases are worth handing to an agent.
What is Agentic AI in E-Commerce?
Agentic AI in e-commerce refers to artificial intelligence systems that can complete several steps of a shopping task independently to reach a goal the user has set. A conventional chatbot answers questions. Autonomous AI agents go further. They interpret an instruction, weigh the available options, connect to digital services and carry out the actions they have been permitted to take.
Consider a customer who asks an agent to find a laptop under $1,000 with a specific processor and screen size. The agent compares products, identifies the best value and prepares the order. If the customer has granted permission, it also completes the payment.
The difference from earlier tools is that search, decision-making and payment sit inside one conversation. This is conversational commerce taken to its logical end. As a result, AI agents in online shopping could become a second interface between consumers and retailers, sitting alongside websites and apps.
Why AI Agents in E-Commerce Are Not Yet Mainstream
Example:
DoorDash provides the clearest evidence of the gap between what the technology can do and what customers actually do.
TechCrunch reported on October 30, 2026 that DoorDash told investors during its third-quarter earnings call that it had received only “several hundred” orders from third-party AI agents, including Muse, Instinct and Fo. The company reported completing roughly three million deliveries a day on average in 2025, so the agent figure is small by any measure. CEO Tony Xu called agent-based ordering a dynamic area but warned against expecting quick adoption. He also said customers still place most orders through the DoorDash app.
DoorDash is not rejecting the technology. It has built its own AI-powered ordering tools, including a text-to-order agent, a command-line agent and Ask DoorDash, which accepts prompts and photos. It has also released a Model Context Protocol (MCP) integration so corporate customers can order through other chatbots.
DoorDash is building agents inside its own app so that the customer relationship stays with DoorDash, whichever interface the customer prefers. Other retailers should note the strategy.
Three reasons explain the slow start:
- Familiar apps are already fast: In online food delivery, opening an app, choosing a restaurant and paying can take less time than describing a preference to an agent and checking its answer.
- Customers want control: Food orders involve restaurant choice, delivery notes, substitutions and spending limits. Many people prefer to review these details before they pay.
- Accountability is unclear: If an agent misreads an instruction or makes an unauthorised purchase, it is not obvious who is responsible. One bad order can damage trust in the whole experience.
The UK government’s research on agentic AI and consumers points the same way. It describes current consumer-facing uses as mostly limited to controlled tasks, where human confirmation and escalation still matter.
AI shopping agents will succeed only where they solve a problem that existing apps and websites have not already solved.
Key Benefits of Agentic AI in E-Commerce
Adoption may be slow, but the benefits are real for retailers, marketplaces and service providers.
- Personalised product discovery: An agent can read a customer’s preferences, budget, purchase history and stated needs, then narrow hundreds of listings to a short, relevant list. For retailers, this connects the right product with the right buyer. It also supports replenishment reminders and sensible alternatives when an item is out of stock.
- Less friction in repeat purchases: Traditional online shopping asks the customer to search, compare, check stock, enter delivery details and pay. An agent can handle these steps in one exchange. A household that buys the same supplies every month could ask an agent to spot what is running low, compare prices and prepare an order within an approved budget. AI-powered shopping is most convincing in this kind of repeat task.
- Better service and lower operating cost: Retailers can use agents to answer product questions, check order status and start eligible returns or refunds. When the agent is connected to inventory, logistics and customer relationship management systems, it can coordinate work across departments. Google Cloud’s January 2026 discussion of agentic commerce describes this shift from passive digital interactions to AI that manages multi-step customer journeys.
- Stronger retention, if handled honestly: Agents can notice recurring needs and remind customers when to reorder. The risk is obvious. If recommendations exist only to raise purchase frequency, customers will notice, and trust will fall. Retention built on real usefulness lasts longer than retention built on nudges.
Challenges Businesses Must Address First
Giving software purchasing authority creates technical, financial and legal exposure:
| Challenge | Business implication |
| Data privacy | Agents may access payment details, preferences and purchase histories. |
| Payment security | Fraud, unauthorised transactions and unclear permissions create financial risk. |
| Integration | Agents need reliable links to catalogues, inventory, checkout and fulfilment. |
| Accountability | Responsibility must be clear when an agent buys or recommends the wrong item. |
| Consumer trust | Customers must know what the agent can see, decide and buy. |
| Commercial bias | Paid placements and brand deals can reduce the neutrality of recommendations. |
Payment security deserves the closest attention. An April 2026 report from the International Monetary Fund explains that agentic payments raise questions about transaction authorisation, traceability, cybersecurity and accountability. It stresses that clear authorisation and settlement controls must remain in place even when AI systems start the transaction.
In practice, businesses can set spending limits, require confirmation above a chosen amount, keep audit trails and define escalation steps. Customers should be able to review purchases, withdraw permissions and see how their data is used.
Interoperability is the second hurdle. An agent may recommend a product well and still fail because it cannot read live prices, stock levels, delivery options or payment systems. Infrastructure therefore matters as much as the choice of AI model.
Opportunities: Who Will Own the Customer Relationship?
For years, retailers have paid to attract customers through search, social media, apps, loyalty schemes and advertising. Those channels produce data about what people look at, what they ignore and what makes them return. If customers begin asking an outside agent to find, compare and buy, the retailer may no longer be the first point of contact.
The retailer may still receive the order, but it may lose the browsing behaviour that came before it. That includes the products a customer considered and rejected, the price points that caused hesitation and the questions asked along the way. That data is what powers personalisation and merchandising decisions. An order without the journey behind it is a thinner asset.
Three revenue models are likely to develop around AI commerce platforms:
- Advertising and sponsored recommendations: AI platforms could let retailers pay to appear within relevant recommendations, tied to a specific purchase need rather than a general audience. Transparency is essential. Consumers should be told when a result is sponsored and whether a commercial arrangement shaped it.
- Partnerships between retailers and AI platforms: Retailers can open their catalogues, inventory and checkout to external agents. Shopify’s January 2026 announcement of its Universal Commerce Protocol (UCP), developed with Google, points in this direction. The standard is designed to help agents work with merchants across different AI channels. For smaller retailers, such integrations could offer a route to AI-driven shoppers without building an agent of their own.
- Agent-enabled purchasing and subscriptions: Revenue can also come from automated replenishment, personalised shopping services and premium assistance. An online grocer could let customers set recurring preferences, then have an agent flag what needs restocking and prepare an order for approval. Premium help may suit categories that demand research, such as electronics, home appliances and specialist equipment.
The future of AI in retail will probably combine direct sales, platform partnerships, advertising and subscriptions. The difficult part is designing models that benefit both retailers and shoppers.
A Delegation Test for Deciding What Agents Should Buy
A more useful approach is to test each purchase type against three questions.
- Is the purchase repeatable? Items bought on a regular cycle, such as household supplies or pet food, give an agent clear patterns to learn from.
- Is a mistake cheap to fix? A wrong brand of washing powder is a small problem. A wrong flight is not. The lower the cost of error, the safer it is to delegate.
- Is the comparison heavy? Where a decision involves many specifications and prices, an agent saves real effort. Where the choice takes ten seconds, it does not.
A purchase that scores yes on all three is a strong candidate for an agent. A purchase that scores no on all three, such as a quick takeaway order from a favourite restaurant, is likely to stay in the app. This may explain DoorDash’s numbers. Many food orders are fast, personal and tied to a particular mood, which makes them poor candidates for delegation. I would expect early agent adoption to be strongest in replenishment and research-heavy categories, though the data to confirm this is still thin.
How E-Commerce Businesses Should Prepare
A business does not need to rebuild its systems at once. Gradual preparation is more sensible.
- Fix product data first: Agents depend on accurate descriptions, prices, specifications, stock levels, delivery details and return policies. Inconsistent data leads to wrong purchases.
- Review how systems connect to outside platforms: Emerging standards and integrations can help, and AI integration in business works best when it starts with the catalogue and checkout.
- Set clear transaction permissions: Customers should control spending limits, payment approval and purchase confirmation.
- Choose where automation helps: Prioritise complex product discovery, repeat orders and service tasks. Not every purchase needs an agent.
These steps allow experimentation without putting security, customer relationships or reliability at risk.
Conclusion
DoorDash’s figures show that technical capability and consumer adoption move at different speeds. They do not show that AI shopping agents will fail. Some customers will stay with familiar apps. Others will gradually hand over research, comparison and routine buying.
The businesses best placed to benefit will combine dependable commerce infrastructure with useful automation, honest recommendations and strong customer protection.
AI agents in e-commerce are more than another way to automate online shopping. They may change where purchasing decisions happen, which platforms shape those decisions and how retailers earn revenue. DoorDash’s experience suggests the change will depend on convenience, trust and a clear reason for customers to alter their habits.
Author’s Note:
I find the developments around agentic AI particularly interesting because they reveal something we often overlook when discussing new technology: people do not change their habits simply because something new becomes available. Take DoorDash, for example. Most customers still prefer opening an app, choosing their meal and placing an order themselves. It is familiar, straightforward and gives them a sense of control. An AI agent has to offer a genuinely better experience before people are willing to change that routine.
What interests me even more is what this shift could mean for businesses. If customers begin relying on AI agents to discover products and make purchasing decisions, retailers could gradually lose direct contact with the people buying from them. That is a concern worth taking seriously. I believe the real opportunity lies in using AI to make shopping easier while preserving the trust and connection customers have with brands. After all, people may welcome help with their decisions, but that does not mean they want to surrender those decisions entirely.
