Latest news with #shoppingassistant
Forbes
3 days ago
- Business
- Forbes
Walmart Bets Big On Generative And Agentic AI With ‘Sparky'
Walmart last week unveiled Sparky, a genAI-powered shopping assistant embedded into the Walmart app. The new AI assistant, Sparky, isn't just another chatbot bolted onto an app. It's part of a much bigger plan to use autonomous agents to transform how people shop. Sparky isn't designed to just answer product questions. It can act. If you're planning a cookout, Sparky won't just list grill options. It'll check the weather, suggest menus, and help schedule delivery. If you're reordering household supplies, it remembers preferences, checks stock, and confirms shipping options. The idea is to reduce friction and turn shopping from a search problem into a service experience. Right now, Sparky can summarize reviews, compare products, suggest items for occasions such as beach trips or birthdays, and answer real-world questions such as what sports teams are playing. In the coming months, additional features will include reordering and scheduling services, visual understanding that can take image and video inputs, and personalized 'how-to' guides that link products with tasks such as fixing a faucet or preparing a meal. According to Walmart's own research, consumers may be more ready for the shift to agentic and generative AI-powered shopping than anyone expected. In the company's latest 'Retail Rewired 2025' report, 27% of consumers said they now trust AI for shopping advice, more than the number who trust social media influencers (24%). That marks a clear break from traditional retail playbooks. Influence is shifting from people to systems. AI already outranks influencers. AI's rapid emergence at the core of ecommerce transactions from LLM chats to embedded applications is clear. A core reason for the adoption of AI is that speed dominates. 69% of customers say quick solutions are the top reason they'd use AI in retail. Some of Walmart's internal research results are genuinely surprising. Nearly half of shoppers (47%) would let AI reorder household staples, but just 8% would trust an AI to do their full shopping without oversight. 46% say they're unlikely to ever fully hand over control. Likewise, data transparency matters. Over a quarter of shoppers want full control over how their data is used. So, Sparky isn't meant to replace human decisions. It's meant to take care of the repetitive stuff, and ask for help when the stakes are higher. Competitors like Amazon, IKEA, and Lowe's are also racing to launch AI assistants. But Walmart is going further. It's building a full agent framework, not just customer-facing bots. Sparky handles shopping. Wally helps merchants. Internal systems assist associates. It's a top-to-bottom AI rollout. Sparky's promise goes beyond convenience. Where recommendation engines once matched products to past clicks, Sparky looks to understand intent in context. If you say, 'I need help packing for a ski trip,' Sparky should infer altitude, weather, travel dates, previous purchases, and even airline baggage limits to propose a bundle, jacket, gloves, boots, and all. This leap requires multimodal AI capabilities including text, image, audio, and video understanding. Imagine snapping a photo of a broken cabinet hinge and getting the right part, DIY video, and same-day delivery. That's the Sparky roadmap. Walmart is also developing its own AI models, rather than relying solely on third-party APIs like OpenAI or Google Gemini. According to CTO Hari Vasudev, internal models ensure accuracy, alignment with retail-specific data, and stricter control over hallucination risks. The retail industry is saturated with automation at the warehouse and logistics layer, but AI agents at the consumer-facing layer are still new territory. Sparky might be the first mainstream proof of concept. But the real story is the architecture: a system of purpose-built, task-specific agents that talk to each other across user journeys, all tuned for high-volume retail complexity. That's a blueprint other enterprises will want to study, and possibly copy. With greater autonomy comes greater risk. Will Sparky recommend the wrong allergy product? Will it misread an image and send the wrong replacement part? Walmart is trying to stay ahead with built-in guardrails: human-in-the-loop confirmations, user approval on sensitive actions, and transparency around how data is used. But the challenge will scale. Sparky's real-world performance, not its launch sizzle, will determine if customers trust it to become a permanent fixture in their shopping lives. Walmart's Sparky is the company's most aggressive bet yet on autonomous digital agents. The trust delta between AI and influencers may seem small now, but it will only widen. The underlying implication? E-commerce interfaces are about to go away. Instead of clicks and filters, shopping becomes a dialogue.
Forbes
4 days ago
- Business
- Forbes
Agentic Commerce And EU VAT Compliance: When AI Buys For You
Digital background depicting AI technologies. Agentic commerce is a new way of shopping where AI agents (smart computer programs) do the shopping for you. Instead of having to search, compare, and buy things yourself, you can tell an AI agent what you want, and it will take care of everything. These agents don't just suggest products — they can actually make decisions and complete the purchase for you, using your saved payment methods. You still stay in control by setting the rules and determining what the agent is allowed to buy. Using smart purchasing agents may save you time and effort. Rather than having to remember to place orders or spend time searching for the best deals, your agent does all of that work for you. For busy people, or for routine purchases like groceries or pet supplies, it's like having a tireless, always-on shopping assistant. How does agentic commerce actually work? You simply give the AI agent a task, just as you would give instructions to a personal assistant. Here's a simple, realistic example of how this could work in everyday life. Imagine you have a dog named Max, and every month you need to buy a specific type of dog food. With an AI agent, you can set it up once by saying something like, 'Every month, order Max's dog food. Look for discounts, and keep the cost under $50. Only buy from stores that have 90% positive reviews.' From then on, the agent automatically handles it for you. It goes to work behind the scenes by browsing different online stores, comparing prices, and reading product reviews. Once it finds a product that meets your criteria, it selects it, checks out, and pays for it using your saved and securely protected payment information. You receive a notification or confirmation once the purchase is complete, so you always know what's happening. Max's food arrives at your door, and you don't have to do a thing. As AI-powered shopping agents start making purchases on behalf of consumers, one question looms large for tax authorities and businesses alike: who actually participates in the transaction? In the EU VAT system, it is essential to identify who is making the purchase and who is selling the product. Under traditional VAT rules, this is straightforward: the buyer and seller are natural or legal persons, and the transaction occurs directly between them. However, the rise of autonomous AI agents introduces new questions. Does the presence of an AI agent in the transaction flow change the identity of the purchaser? Since these agents act much like intermediaries—searching, selecting, and even placing orders—do they alter the transaction flow? Under EU VAT law, if an intermediary (also called an 'undisclosed agent') makes a purchase in their own name but on behalf of a third party, a legal fiction of two transactions is created: first, from the supplier to the intermediary, and then from the intermediary to the final customer. However, this rule only applies when the intermediary is a natural or legal person—someone who can hold legal rights and obligations. AI agents do not have legal personalities. They cannot conclude contracts or assume legal responsibility in their own name. Even though an AI agent may sit 'between' the buyer and seller in the technical transaction flow, it cannot legally act as either the buyer or the seller. As a result, no VAT fiction of a two-stage sale is created when an AI agent is involved. For VAT purposes, the transaction is treated as a direct sale from the seller to the buyer. All key VAT elements—such as the place of supply and applicable VAT rates—are determined based on the buyer's legal status and location, exactly as if the buyer had completed the purchase without the help of an AI agent. Online sellers are required to determine where their customers are located in order to apply the correct VAT rate. EU VAT rules stipulate that sellers must collect two pieces of non-contradictory evidence—for example, an IP address plus a billing or bank address—to identify the buyer's location. However, some online sellers do not strictly follow these rules and rely on only one method of customer location verification. Some sellers use IP geolocation to estimate the buyer's location. When a smart agent logs in via its own servers, it might appear that the 'buyer' is located elsewhere. An AI agent may run on servers located anywhere in the world—Europe, Asia, North America—regardless of the buyer's actual residence. This could cause sellers to apply the wrong VAT rate if VAT is calculated automatically based on the IP address rather than verified billing data. For instance, if an AI agent uses a UK-based IP while the buyer is actually in Germany, the seller might mistakenly apply UK VAT instead of German VAT. The emergence of agentic commerce underscores the necessity of proper customer location verification. In agentic commerce, the AI agent's IP address may differ from the buyer's actual location—and that is acceptable from a VAT standpoint. What matters is the location of the actual buyer, which must be verified using reliable data such as a billing address or payment details. Even if an agent appears (via IP) to be in one country, the sale is taxed according to where the buyer resides or where the product is used. Sellers may use IP as a supplementary cross-check, but it cannot replace the legally required customer information. As agentic commerce gains momentum, one emerging risk stands out: what happens when consumers instruct AI agents to seek VAT-free purchases or use 'strategies' to avoid VAT? At first glance, telling an AI agent to "shop tax-free" may seem harmless as everyone likes a good deal. But within the EU VAT system, this can become a serious compliance issue. In B2C transactions within the EU, VAT is generally unavoidable. Instructing an agent to seek out "tax-free" products can easily steer it toward non-compliant or even fraudulent behavior. The agent might start prioritizing purchases from sellers who falsely advertise 'VAT-free' sales when VAT should, in fact, be charged. Cross-border e-commerce introduces additional risks. AI agents can easily split large orders into multiple smaller transactions to stay below VAT and customs exemption thresholds. While the EU removed the VAT import exemption of €22 in 2021, the broader customs duty exemption threshold of €150 remains. Parcels under €150 also benefit from fewer border inspections and simplified processes, such as the Import One Stop Shop scheme. The EU already considers VAT fraud involving low-value, cross-border e-commerce a major concern. Agents programmed to optimize tax outcomes could significantly increase the number of low-value shipments. For example, a €200 order could be broken into several €50 transactions to slip below the €150 threshold. Intra-EU B2B trade presents another potential abuse case. Under EU VAT rules, cross-border sales to business customers are generally VAT-exempt. The seller typically requires the customer to provide a valid VAT ID number to confirm their business status. But a private individual could instruct an AI agent to falsely claim business status to avoid paying VAT. Agents can easily automate this by harvesting or generating fake VAT IDs and submitting them across multiple sellers. Many small EU sellers do not automatically validate VAT IDs in real time (though they should), making them vulnerable to unknowingly participating in this type of fraud. While this kind of deception is already possible manually, AI agents can scale it rapidly, increasing both the risk and potential volume of fraud. As agentic commerce grows, it will become essential to establish clear guidelines—a kind of 'responsible AI shopping agent policy.' Agents must be designed to handle identity and VAT status in a compliant manner. At the same time, online sellers should implement automated VAT ID validation checks to defend against abuse. Without these controls, the next wave of digital commerce could create new tax compliance challenges across the EU. Agentic commerce is expected to become both smarter and more widespread in the coming years. According to research by Gartner, 33% of businesses will incorporate agentic AI by 2028, up from less than 1% today. In just a few years, it's likely that AI-driven shopping will be a common part of consumer behavior—helping people save time, discover better deals, and automate routine purchases with minimal effort. Agentic commerce doesn't 'break' the EU VAT system, but it certainly amplifies many of its long-standing complexities—from questions about buyer location and identity verification to the potential misuse of exemption thresholds. Agentic commerce will only thrive in the long term if it remains trusted by consumers, businesses, and tax authorities alike. That trust depends on building a system where AI-enabled convenience does not come at the expense of VAT compliance. Striking that balance will be one of the challenges in the evolving landscape of digital commerce. The opinions expressed in this article are those of the author and do not necessarily reflect the views of any organizations with which the author is affiliated.
