The agentic AI vision where autonomous agents discover products, compare prices, place orders, manage logistics, and even returns on behalf of consumers, is certainly seductive. Instead of scrolling through hundreds of websites, shoppers simply tell an AI assistant what they want, the agent researches options, negotiates prices, completes the purchase and arranges preferred delivery automatically.
We read predictions that this will fundamentally reshape ecommerce while at the same time there are also views that consumers will never trust AI to make purchasing decisions involving money, personal preference or emotional judgement. The truth probably lies somewhere between those two extremes.
Consumer trust will undoubtedly take time to develop, particularly for high value or emotionally significant purchases. Regulation is still being worked over and many shoppers will continue to want final approval before an order is placed. However, all these things will work themselves through, a bigger question is, are retailers ready?
Most retail infrastructure is still designed for humans navigating websites rather than multiple AI agents collaborating across the customer journey. Agentic commerce is often presented as a single intelligent assistant managing an entire transaction whereas in reality, it is more likely to involve numerous specialised AI agents performing different roles – identify products, check availability, evaluate promotions, authorise transactions, determine best delivery option and curate post-purchase services to manage tracking, exceptions and returns.
Sounds good but consider, each agent depends on fast, accurate and secure access to data and if those systems cannot communicate consistently, autonomy quickly breaks down. A further complication is the need for a secure unified standard for agent-to-agent interaction such as an MCP universal plug.
Ecommerce platforms, warehouse systems, customer service platforms, payment providers, ERP systems and carrier networks have all become very sophisticated but they have all been developed on their own. Even where they are connected, many are reliant on fragmented integrations, manual workflows or isolated data.
This is the perfect description of business as usual where humans compensate for any gaps every day but AI agents can’t. Unlike people, AI has little tolerance for ambiguity. It requires structured data, consistent interfaces and reliable system connectivity where every disconnected API, inconsistent delivery status or isolated data source becomes another point where autonomous decision making can stall.
We can see this clearly on the delivery journey. Most discussions around AI in logistics focus on operational efficiency, specifically, intelligent carrier selection, predictive ETAs, automated exception management and smarter routing decisions. These are all delivering measurable improvements but they only represent the first phase of AI adoption. The bigger opportunity lies in recognising that delivery platforms are becoming orchestration platforms.
Once an order has been placed, multiple decisions continue throughout the customer lifecycle. Which carrier provides the best experience? How should disruption be communicated? Which customers require proactive updates? When should loyalty offers be presented? Which delivery outcomes indicate future purchasing intent?
These are all interconnected decisions driven by shared data which will make delivery information a very valuable strategic asset. Delivery captures customer preferences, communication behaviour, fulfilment performance, delivery reliability and satisfaction signals that extend well beyond logistics. And AI can connect these signals in ways that humans simply cannot process at scale.
Instead of merely reporting where parcels are, AI can begin recognising behavioural patterns. Which customers become anxious after delayed deliveries? Which delivery promises generate higher repeat purchase rates? Which carrier or service consistently creates better experiences for particular customer segments?
The result is that automation will enable recommendation so that better decisions can be made on behalf of customers. One important area is carrier integration but connecting hundreds of carriers makes no sense if retailers cannot intelligently decide which one should handle each individual order. AI must enable intelligent carrier orchestration.
However, this is only possible if the infrastructure behind them is capable of supporting autonomous collaboration. Retailers should therefore spend less time asking whether agentic commerce is real and more asking whether their systems are capable of supporting it. The goal is connected commerce platforms, where discovery, ordering, fulfilment, delivery and returns operate as a single intelligent ecosystem rather than a collection of disconnected systems. Then AI can really start to deliver
