Risks and Rewards for CIOs

It is now possible to complete retail purchases without ever leaving the ChatGPT platform, thanks to Instant Checkout, a new transaction protocol that OpenAI launched this week. Shoppers were already using the AI platform to collate product recommendations based on their search requests, but they can now choose an item and pay for it in the same ChatGPT interface, without directly engaging with the merchant itself.

“It’s sort of an inevitable move from using generative AI capabilities to create enhanced search, to that next step of execution,” said Robert Hetu, an analyst at Gartner.

ChatGPT supports purchases from sellers on Etsy, with OpenAI announcing that over 1 million Shopify merchants would also be available “soon.” Instant Checkout builds on previous AI shopping capabilities that have entered the market, such as Perplexity’s in-chat shopping and payments feature, and Microsoft’s Copilot Merchant Program.

For shoppers, these new transaction protocols may look like an efficiency upgrade with minimal downsides. But for businesses, the technology offers a more complex path forward.

Agentic Shopping: An Opportunity for B2B Automation

When the agentic shopping system takes over that interface layer, the dynamic between vendor and customer shifts: There is less friction between browsing and purchasing, but also less direct engagement and relationship-building between brand and buyer. OpenAI also charges merchants a fee on every sale they make on the platform, adding another layer to the cost structure.

Related:Salesforce’s Benioff Says Vendors Have an Agentic AI Pricing Problem

CIOs across different sectors are looking at this announcement with varying degrees of interest, trepidation, and excitement. While the current use case is retail-focused, agentic shopping is particularly suited to B2B environments across various industries, and could substantially disrupt how vendors interact with each other.

“I do believe that we’re going to move increasingly to the automation of interactions and transactions, particularly in categories of business where the objects are very well defined,” Hetu said. “That’s where in a B2B process it makes a lot of sense, because you’re negotiating many known variables.”

Mass adoption appears to be several years away. The Gartner hype cycle shows AI shopping agents and machine customers as pre-peak of inflated expectations, with significant movement expected to occur over the next five to 10 years. This is because there is still so much that must be figured out before agentic AI transactions can be executed securely, efficiently, and at scale.

Related:4 Ways to Redefine Resilience for the AI Era

Genuine Breakthrough or LLM Damage Control?

Skeptics believe that this isn’t so much of a groundbreaking transformation as it is a gimmick, designed to distract from slow progress in core large language model (LLM) capabilities. David Linthicum, a cloud and AI subject-matter expert and founder of Linthicum Research, said he believes OpenAI is using the new agentic shopping system to “give an appearance of continuous innovation,” while they are “running out of ways to move the needle on the core technology itself.”

“The reality is that LLM providers, including OpenAI, have encountered a data wall,” Linthicum said. “The explosive progress in language models over the last couple of years was driven in large part by the ingestion of enormous amounts of new data. But at this point, we’ve reached saturation — there just isn’t that much novel, high-quality data left to train on.”

Others argue that, while gimmick may be too strong a word, there are still limitations to the current use case of agentic shopping, and many are tied to data handling. For CIOs to take full advantage of these agentic systems, there needs to be a complete integration of back-end data to provide full functionality. Real-time inventory, supply chain logistics, endpoint delivery and payment processing must all be supported within the automated system, while also providing a frictionless customer experience.

Related:Gartner: Disillusionment Around AI Presents a ‘Hero Moment’ for CIOs

Worth Watching But Not Yet Enterprise-Ready

For Keith Townsend, founder of The Advisor Bench and executive strategist, this protocol is something to watch with curiosity — but not something for CIOs to act on right now.

“OpenAI’s agentic shopping system is interesting in the same way Salesforce or ServiceNow’s early agentic AI features are interesting: It provides an ‘easy button’ to showcase what agent-driven interactions could look like,” he said. “It lowers the barrier for experimenting with agentic AI in a familiar context. That said, I don’t see CIOs rushing to embrace it.”

A key reason for hesitancy is the anticipated resistance of CIOs to handing over this data to a third-party AI platform, Townsend argued. Hetu added that cybersecurity concerns alone pose a real risk to business operations, opening up new vulnerabilities for attackers to potentially exploit. Then there are privacy concerns, compliance issues, even ethical implications. In short, a data minefield that executives may want to avoid.

“Agentic systems only become valuable when they can reason over and act on enterprise data and processes,” Townsend said. “But until CIOs are comfortable that their data is being handled under their terms, these offerings are more proof of concept than strategic priority.”

When in Doubt, Be Prepared

Still, others see a path ahead that ends with AI agents executing a large quantity of transactions. Isaac Sacolick, president of StarCIO and a bestselling author, said he believes that executives would be wise to consider a future where LLMs and AI agents emerge as a new buying channel. Even if they’re not adopting the technology right now, they can still lay the groundwork for future adoption when the ecosystem is more developed.

“For CIOs, this integration highlights the importance of having APIs and detailed metadata about their products and services,” Sacolick said. “As CIOs develop AI agents, it will be crucial to support MCP [Model Context Protocol]and other agent-to-agent protocols to enable participation in AI-enabled ecommerce ecosystems.”

Gartner’s Hetu agreed. He explained that only a unique organization would be positioned to gain a competitive advantage by acting on this technology right now — but that everyone should be monitoring it and getting ready.

In practice, this means that CIOs should be asking themselves a series of questions: What is the potential for agentic AI in their organization? Where does this potential come from? Is it from driving revenue, or displacing costs? Then they need to assess the impact on security, compliance, workforce disruption — and only then make a very calculated decision. Still, it’s a discussion worth having, he said.

“I think a CIO would ignore this at their peril,” Hetu said. “They’ve got to help the business leaders understand what is coming and be prepared.”