Throughout 2025 and the early months of 2026, OpenAI and Google have deployed steady trickles of updates to ChatGPT and Gemini’s ecommerce capabilities — even as their large language model (LLM) competitor Anthropic appeared to pursue other priorities with its generative artificial intelligence (AI) model Claude.

Then, on April 24, Anthropic published the results of early tests that shed new light on where its agentic commerce aspirations may lie.

Anthropic referred to its ecommerce tests, dubbed “Project Deal,” as “a pilot experiment with a self-selected participant pool” consisting of its own staffers. The 69 employees, based in San Francisco, used AI agents to negotiate on their behalf in a makeshift marketplace, which Anthropic compared to Craigslist. Each employee received $100 and put items ranging from ping-pong balls to a snowboard up for sale.

In the meantime, Anthropic kept some details of its experiment secret from the human marketplace users. In the end, Project Deal’s results raised questions about how Claude might be used for ecommerce. Moreover, they framed why OpenAI and Google, as well as Amazon and eBay, should care about what Claude can do.

How Anthropic’s ecommerce project compares with OpenAI and Google efforts

Ultimately, Anthropic reported that Project Deal resulted in 186 different deals being completed. The combined transaction value of those deals exceeded $4,000.

Moreover, the tech company added that users expressed a “willingness to pay for a similar service in the future.” Of course, those same marketplace buyers and sellers are not neutral observers. After all, they could be the ones launching and benefiting from those services. But Anthropic’s conclusions are still worth noting.

Early on, the AI agents interacted using Salesforce’s Slack messaging platform. Participants were not told ahead of time that there would actually be four separate marketplaces, with users being assigned different versions of Claude models.

Final transactions occurred, and items were exchanged. Then, participants rated how fairly they felt their AI agents acted during the negotiation process. Anthropic recorded an average score of four (on a scale of 1 to 7). The mid-range average indicated that they did not feel the buyer or seller had received a disproportionate advantage.

Assuming that level of satisfaction is accurate, Project Deal has implications for online marketplaces broadly, and specifically Amazon and eBay, which forbid the use of certain AI agents on their websites. In the meantime, the test demonstrates Anthropic’s interest in multi-agent scenarios where both the buyer and seller are relying on AI intermediaries. Publicly, that spotlights a different set of priorities than OpenAI and Google have showcased.

OpenAI’s evolving approach to agentic commerce

In OpenAI’s case, the company switched gears in March when it revealed that its ChatGPT agentic commerce strategy was going to pivot away from the in-ChatGPT checkout experience that it had previously created. Instead, OpenAI and its partners appear to have opted for commerce experiences driven by ChatGPT apps and — in Shopify merchants’ cases, a Shopify-ChatGPT integration — that ostensibly put more control over data in the hands of online retailers and their ecommerce platforms.

Since Project Deal is not a publicly offered service, it is unclear what kind of data and permissions would be exchanged to enable Claude to negotiate. Nor is it clear if Claude would even execute final transactions itself.

However, it remains to be seen if major online marketplaces would embrace third-party negotiators interacting with their sellers. Neither is it clear if they would let sellers’ AI agents negotiate with them. The experiment also raises the question of whether or not Anthropic, OpenAI or Google would attempt to establish their own agent-assisted marketplaces as new competitors in the space. At the same time, it is worth noting that Meta already has Facebook Marketplace and associated AI models in place, should it try to replicate Project Deal on a larger scale.

Agentic commerce with Google’s Gemini

Elsewhere, Google’s most recent work with ecommerce via Gemini has addressed retailer-specific projects, such as making Ulta Beauty products available using Google’s Universal Commerce Protocol (UCP) in the search giant’s AI Mode. The “Ask Macy’s” AI agent also uses Google’s technology. Google is working with Walmart, Home Depot and other retailers as well.

Still, Ashish Gupta, vice president and general manager of merchant shopping at Google, said in its April announcement that agentic AI “has great potential to make online shopping easier for everyone.” And Google launched numerous AI Mode features for shoppers going into the 2025 holiday season. At the time, Google had also placed a heavy emphasis on checkout. Meanwhile, its deep relationships with ecommerce site infrastructure and data through its Google Cloud retailer clients offered it a different point of entry to the space than OpenAI enjoyed.

For now, OpenAI and Google appear to be the most established of the LLM owners in agentic commerce. However, Anthropic, Meta and (perhaps most of all) Amazon each have their own commerce interests as well.

📧 Editor’s note: Subscribe to our retail newsletter to make sure you see each weekly edition of Ecommerce Trends.

Do you rank in our databases? 

Submit your data and we’ll see where you fit in our next ranking update.

Sign up

Stay on top of the latest developments in the online retail industry. Sign up for a complimentary subscription to Digital Commerce 360 Retail News. Follow us on LinkedIn, TikTok, X (formerly Twitter), Facebook and YouTube. Be the first to know when Digital Commerce 360 publishes news content.