Why Walmart and OpenAI Are Shaking Up Their Agentic Shopping Deal

When Walmart and OpenAI announced their collaboration on the Instant Checkout feature, it seemed like a breakthrough moment in e-commerce history. The chatbot was supposed to enable users to purchase products directly within ChatGPT without leaving the conversation — elegantly, quickly, almost magically. However, reality turned out to be more complicated. After several months of testing, both companies decided to completely change their strategy. Instead of an agentic assistant that would make purchases independently, Walmart decided to embed its Sparky chatbot directly into the interfaces of ChatGPT and Google Gemini — an approach that is significantly more conservative, but potentially more practical.

This change signals something important for the entire industry: agentic AI systems that operate completely autonomously are still far from being ready for mass deployment. Instead of waiting for future technology, large corporations are pragmatically returning to present reality — where humans remain in the decision loop, and AI supports rather than replaces. For Polish users and technology creators, this is a lesson about how the hype around autonomous agents must face practical limitations and consumer expectations.

Instant Checkout — an ambitious plan that didn't survive contact with reality

When OpenAI and Walmart presented Instant Checkout, it was a vision of the future of digital commerce. A user could simply write in ChatGPT: "Buy me milk, bread, and butter," and the system would automatically search Walmart's catalog, select products, add them to the cart, and — if it had access to payment data — complete the transaction. All without clicking links, without visiting the store's website, without the traditional e-commerce interface. This was an example of what the industry calls agentic AI — artificial intelligence that operated autonomously on behalf of the user, making real decisions and taking actions.

The problem appeared quickly. First, users were reluctant to give such control to an AI system. Shopping is not just a transaction — it's a decision where we want full control: exactly which product, which brand, what price, whether it's really what we need. Instant Checkout turned out to be too autonomous, too distant from real consumer preferences. Second, legal and security issues — giving AI access to payment data and order history is a sensitive matter, especially in the United States, where consumer privacy is hotly debated. Third, practical problems: the system would need to handle returns, complaints, delivery issues — the entire logistics that traditionally handled by humans.

The result? Instant Checkout never moved beyond the pilot phase. It was interesting as an experiment, but as a commercial product, it turned out to be premature. Walmart and OpenAI had to admit that their initial vision was too ambitious given the current state of technology and market readiness.

Sparky — a return to fundamentals, or an assistant instead of an agent

The new strategy is much more pragmatic. Instead of a completely autonomous system, Walmart is embedding Sparky — its chatbot — directly into ChatGPT and Google Gemini. This means that the user talks to Sparky, but still remains in control of the purchasing process. Sparky can show available products, suggest options, explain differences between brands, but the final decision — whether to actually buy and what to buy — is made by the human.

This approach has several advantages. First, users feel safer. They see what's happening and can change their mind at any stage. Second, Sparky can be much more intelligent in terms of recommendations and personalization. Instead of blindly following orders, it can ask questions: "Are you looking for regular milk or lactose-free? Do you prefer a premium brand or an economical one?" Third, this solution is easier to implement legally and in terms of security — there's no transfer of full account access, just integration with the product catalog.

Sparky will be available as a plugin or integration on both platforms. A ChatGPT or Gemini user can invoke Sparky at any time, talk about what they want to buy, and the chatbot will show them options available at Walmart. If they like something, they can proceed to checkout — but fully consciously, with full control over their data and decisions.

Why agentic AI is not yet ready for mass deployment

The story of Walmart and OpenAI is a symptom of a larger problem that the entire AI industry is struggling with. Agentic systems — AI capable of autonomous decision-making and action — are fascinating in theory but prove problematic in practice. There are several reasons why this technology is not yet ready.

First and foremost, it's a matter of trust. Humans naturally don't trust systems that operate outside their view. If AI buys me a product I didn't want, or makes a transaction I could have blocked — that's not convenience, that's a nightmare. Trust in AI is built slowly, through small, safe steps, not through immediately handing over full control of finances and personal data.

Second, there's the issue of accountability. If agentic AI makes a mistake — for example, buying a product with an expired expiration date or sending an order to the wrong address — who is responsible? Walmart? OpenAI? The user? The law has no clear answers to these questions. Until this issue is resolved, large corporations will be cautious about deploying fully autonomous systems.

Third, it's a matter of complexity. The real world of online commerce is full of exceptions, promotions, price changes, and availability issues. An agentic system would need to handle all these scenarios, and that requires an incredible level of sophistication. Simple chatbots can't handle this — and more advanced systems are too expensive and complicated to maintain.

Integration with ChatGPT and Gemini — new ecosystems of digital commerce

Walmart's decision to embed Sparky in ChatGPT and Google Gemini has deep implications for the future of e-commerce. This means that large AI platforms are becoming a new distribution channel for commerce. It's no longer just a website or mobile app — it's a conversation in a chatbot that naturally transitions into a commercial transaction.

This changes market dynamics. Traditionally, merchants invested in their website, optimized it for search engines, built a brand. Now they need to think about how to be visible in AI interfaces. Will their products be recommended by Sparky? Will they be visible in results when someone asks ChatGPT about "the best milk on the market"? This is a new battleground for consumer attention.

For OpenAI and Google, this is also a strategic shift. Both companies realize that the future doesn't belong to pure search engines or informational chatbots — it belongs to platforms that can directly support transactions. By integrating Sparky, OpenAI and Google are making their platforms more useful to users and more profitable for themselves (through transaction commissions).

The Polish e-commerce market is much smaller than the American one, but this trend will reach here. Polish e-commerce platforms — Allegro, Ceneo, OLX — must prepare for a scenario where the main entry point for consumers is not their own website, but a conversation with a Polish-language AI chatbot. This requires a completely new approach to SEO, marketing, and customer service.

Security and privacy — why Instant Checkout couldn't be fully autonomous

One of the main reasons Instant Checkout failed is security and privacy issues. For the system to work fully autonomously, it would need access to payment data, order history, user preferences, and even location (to calculate shipping costs). This was too big a leap for consumers.

In Europe, and especially in Poland, data protection regulations (GDPR) would be an additional obstacle. The system couldn't automatically process personal data without explicit, informed user consent. And this isn't about clicking "I agree" — it would have to be conscious, real consent that AI will make decisions on behalf of the user. Most people would never agree to that.

Sparky solves this problem through transparency. The user sees what's happening at every stage. The chatbot shows products, the user reviews them, and if they want to buy, they proceed to checkout themselves. This is compliant with regulations, safe for consumers, and comfortable for all parties.

Implications for the AI industry and the future of agentic systems

The story of Walmart and OpenAI is a lesson for the entire AI industry. The hype around agentic AI systems is enormous — everyone talks about how AI will autonomously manage our finances, book flights, do our shopping. But reality is more complicated. The technology is ready, but the market is not.

The most advanced AI models, such as GPT-4 or Claude 3, are actually capable of autonomous action in certain contexts. They can write code, analyze documents, make decisions based on data. But giving them full control over financial transactions or other sensitive activities is a step that requires not just advanced technology, but also cultural, legal, and organizational changes.

The future of agentic AI systems will probably look more like Sparky than like Instant Checkout. Systems that support humans, recommend, explain, but leave final decisions to the user. This may seem less futuristic than full autonomy, but it's more realistic and safer. And in the long run — more profitable for business, because it will build consumer trust.

What this means for the Polish technology market and e-commerce

For Poland, this change has concrete implications. Polish e-commerce platforms need to start thinking about integration with Polish-language AI chatbots. Allegro, which is the market leader, should consider partnerships with OpenAI or Google so that Sparky or similar systems can recommend products from their catalog. Smaller platforms can look for niche solutions — for example, integration with Polish AI chatbots that specialize in specific product categories.

For developers and startups, this is also an opportunity. The Polish AI industry is relatively young but dynamic. Specialized systems can emerge — chatbots for specific industries (fashion, electronics, food) that will operate on AI platforms but with Polish expertise and knowledge of the Polish market. This is more realistic than trying to compete with OpenAI in the field of general chatbots.

For consumers, this means that within a few years they will be able to buy products in Poland through a conversation with AI — naturally, without leaving the chatbot. This will be convenient, but it also requires education. Polish society needs to learn to trust these systems, understand their limitations, and know how to use them safely.

Sparky in practice — what it will look like for the user

Imagine a scenario: you're in ChatGPT, talking about what to cook for dinner. You say you need ingredients for risotto. ChatGPT suggests a recipe and then offers: "Would you like Sparky to show you where to buy these ingredients at Walmart?" You click "yes," and Sparky appears as a mini-application in the conversation window. It shows available arborio rice, parmesan, vegetable broth — all with prices, availability, and reviews. You browse the options, add your selected products to the cart, and then proceed to checkout.

This is much more natural than Instant Checkout, but still much more convenient than traditional shopping. You don't have to leave ChatGPT, you don't have to type what you're looking for in Walmart's search bar — everything is already available in the context of the conversation you were having.

For Walmart, this means it will be able to reach users who would never visit its website. For OpenAI and Google, this means their platforms become more useful and engaging. For the user, this means convenience, but also the need to be aware that a conversation with AI can directly lead to spending money.

Lessons for other industries — does agentic AI have a future beyond e-commerce?

The story of Walmart and OpenAI also has significance for other industries experimenting with agentic AI systems. Banks are thinking about AI autonomously managing investment portfolios. Insurance companies want AI to autonomously process claims. Logistics companies want AI to autonomously plan delivery routes. Everyone is waiting for the moment when agentic AI will be advanced and trusted enough.

The lesson from Walmart is clear: even in the simplest scenarios, where AI only needs to select products and complete a transaction, autonomy is a problem. In more complex scenarios — for example, in medicine, where an AI error could cost a life — autonomy is even more problematic. The future of agentic systems will probably be in systems that support human decision-making rather than replace it.

This means that the hype around complete automation of everything through AI is overstated. Reality will be more mundane, but more practical: AI will support people, recommend, explain, but final decisions will belong to humans. This may not sound like science fiction, but it is realistic and safe.