admin
Throughout 2025 and the early months of 2026, major players in the artificial intelligence landscape, namely OpenAI and Google, have been steadily introducing updates to their flagship conversational AI models, ChatGPT and Gemini, respectively. These advancements have increasingly focused on enhancing their capabilities within the realm of e-commerce. Meanwhile, their prominent competitor, Anthropic, creator of the generative AI model Claude, appeared to be charting a different course, focusing on priorities beyond immediate commercial applications. This perception, however, was significantly altered on April 24, when Anthropic published the findings of an early-stage experiment that offered a compelling glimpse into its burgeoning ambitions in agentic commerce.
The experiment, internally codenamed "Project Deal," was described by Anthropic as a "pilot experiment with a self-selected participant pool" comprising its own employees. Sixty-nine staff members, all based in San Francisco, were tasked with utilizing AI agents to negotiate on their behalf within a simulated marketplace, which Anthropic likened to a scaled-down version of Craigslist. Each participant was allocated $100 in virtual currency and was instructed to list items for sale, ranging from mundane objects like ping-pong balls to more substantial goods such as a snowboard. Crucially, Anthropic withheld certain details of the experiment from the human participants, a strategic decision that aimed to isolate the performance of the AI agents. The outcomes of Project Deal have not only raised significant questions about the future applications of Claude in e-commerce but also underscore why established giants like OpenAI, Google, and even e-commerce behemoths such as Amazon and eBay, must pay close attention to Anthropic’s evolving capabilities.
Project Deal: A Novel Approach to AI-Driven Transactions
The results of Anthropic’s "Project Deal" initiative revealed a dynamic landscape of AI-mediated transactions. In total, 186 distinct deals were successfully concluded, with a cumulative transaction value exceeding $4,000. Perhaps more significantly, the experiment indicated a strong user appetite for such services, with participants expressing a "willingness to pay for a similar service in the future." While acknowledging that these participants, as both buyers and sellers in the experiment, might not be entirely neutral observers—potentially positioned to benefit from the launch of such services—Anthropic’s findings warrant serious consideration.
The initial interactions within Project Deal took place on Salesforce’s Slack messaging platform. Participants were not informed that the marketplace was segmented into four distinct versions, each utilizing different iterations of the Claude models. This controlled environment allowed Anthropic to gather granular data on AI agent performance under varying conditions. Following the completion of transactions and the exchange of virtual goods, participants were asked to rate the perceived fairness of their AI agents’ negotiation tactics on a scale of 1 to 7. The average score of four suggests a balanced negotiation, where neither the buyer nor the seller appeared to gain a disproportionate advantage.
The implications of this level of satisfaction, if replicable in broader commercial settings, are substantial for the entire online marketplace ecosystem. Platforms like Amazon and eBay, which currently have policies in place that restrict the use of certain AI agents on their sites, may need to re-evaluate their stances. Project Deal distinctly highlights Anthropic’s strategic focus on multi-agent scenarios, where both buyers and sellers rely on AI intermediaries. This approach diverges from the public-facing strategies of OpenAI and Google, suggesting a potentially different pathway for AI integration into the commerce sector.
OpenAI’s Strategic Pivot in Agentic Commerce
OpenAI made a notable strategic shift in March, announcing a redirection of its agentic commerce strategy for ChatGPT. The company announced it would move away from its previously conceived in-ChatGPT checkout experience. Instead, OpenAI and its partners are now focusing on commerce functionalities powered by ChatGPT applications. For merchants on the Shopify platform, this translates to a deeper integration, a Shopify-ChatGPT integration, which is designed to grant online retailers and their e-commerce platforms greater control over their data.
While Project Deal is not a publicly accessible service, its experimental nature raises important questions about the data exchange and permissions required for Claude to engage in negotiation. It remains to be seen whether Claude would be empowered to execute final transactions independently. The broader challenge for major online marketplaces lies in their willingness to accommodate third-party negotiators interacting with their sellers, or to permit sellers’ AI agents to engage in direct negotiation. This experiment also prompts speculation about whether Anthropic, OpenAI, or Google might venture into establishing their own agent-assisted marketplaces, thereby introducing new competitive dynamics. It is also pertinent to note that Meta, with its existing Facebook Marketplace and associated AI models, is already positioned to potentially replicate Project Deal on a larger scale should it choose to do so.
Google’s Gemini and the Retailer-Centric Approach to Agentic AI
Google’s recent advancements in leveraging Gemini for e-commerce have largely centered on retailer-specific collaborations. A key development includes the integration of Ulta Beauty products through Google’s Universal Commerce Protocol (UCP) within the search engine’s AI Mode. Similarly, the "Ask Macy’s" AI agent utilizes Google’s proprietary technology. Google is actively partnering with other major retailers, including Walmart and Home Depot, to develop and implement similar AI-driven commerce solutions.
Ashish Gupta, Vice President and General Manager of Merchant Shopping at Google, articulated the company’s vision, stating in an April announcement that agentic AI "has great potential to make online shopping easier for everyone." In anticipation of the 2025 holiday season, Google introduced a suite of AI Mode features designed to enhance the shopper experience. At that time, the company placed a significant emphasis on streamlining the checkout process. Google’s extensive relationships with e-commerce platforms and retailers through its Google Cloud services provide a distinct advantage, offering a different entry point into the agentic commerce space compared to OpenAI.
Currently, OpenAI and Google appear to be the most established entities among large language model providers actively pursuing agentic commerce. However, Anthropic, Meta, and particularly Amazon, possess significant existing interests and capabilities within the commerce sector, positioning them as formidable players in this evolving landscape.
The Broader Implications for the E-commerce Landscape
The distinct approaches taken by OpenAI, Google, and Anthropic in the burgeoning field of agentic commerce signal a multi-faceted evolution of online retail. OpenAI’s pivot towards app-centric commerce and Shopify integrations suggests a focus on empowering existing e-commerce infrastructure and data ownership for retailers. This strategy aims to build trust and foster collaboration by keeping data within established retail channels.
Google, with its deep integration into search and its Universal Commerce Protocol, appears to be aiming for a more unified and standardized approach to product discovery and transaction across the web. By working directly with retailers and embedding AI capabilities within its core search experience, Google is positioning itself as a central facilitator of online commerce, leveraging its existing user base and technological infrastructure.
Anthropic’s Project Deal, on the other hand, ventures into a more direct and potentially disruptive territory. By simulating a marketplace where AI agents autonomously negotiate on behalf of users, Anthropic is exploring the potential for AI to act as a true intermediary, capable of complex decision-making and transaction finalization. This raises profound questions about the future role of human involvement in online shopping and the potential for AI-driven agents to become sophisticated buyers and sellers in their own right.
The success of Project Deal, indicated by user satisfaction and a willingness to pay, suggests that consumers may be receptive to AI agents handling the intricacies of negotiation. This could lead to more efficient, potentially more equitable, and certainly more convenient shopping experiences. However, it also presents significant challenges for incumbent marketplaces. Policies that currently prohibit AI agents may need to be re-examined as the technology matures and its benefits become more apparent. The question of how to regulate and ensure fairness in AI-driven negotiations will become paramount.
Furthermore, the potential for these AI developers to launch their own marketplaces cannot be overlooked. If Anthropic, OpenAI, or Google were to create agent-assisted marketplaces, they would be entering direct competition with giants like Amazon and eBay, leveraging their advanced AI capabilities to attract both buyers and sellers. This could fundamentally alter the competitive dynamics of the e-commerce landscape, creating new platforms that are built from the ground up with AI at their core.
The involvement of Meta, with its existing Facebook Marketplace, adds another layer of complexity. The social media giant’s vast user base and data analytics capabilities could enable it to rapidly scale a similar agentic commerce initiative, potentially integrating it seamlessly into its existing social and e-commerce offerings.
Ultimately, the current period represents a critical juncture in the development of agentic commerce. While OpenAI and Google have established a strong presence, Anthropic’s bold experimentation with Project Deal, coupled with the inherent strengths of Meta and Amazon in the e-commerce domain, indicates that the race to define the future of AI-powered online shopping is far from over. The coming years will likely witness further innovation, strategic partnerships, and potentially disruptive shifts as these technology titans vie for dominance in this rapidly evolving frontier. The ability of these platforms to navigate ethical considerations, ensure user trust, and deliver tangible value will be key determinants of their success in shaping the future of how we buy and sell online.
