Google vs. OpenAI vs. Visa: competing agent protocols threaten the future of AI commerce
When Walmart and OpenAI announced that the retailer would integrate with ChatGPT, the question became how quickly OpenAI could deliver on the promise of agents buying things for people. In the battle for AI-enabled trading, one of the biggest hurdles is getting agents to complete trades safely.
These are increasingly chat platforms such as ChatGPT replace browsers and become very good at unearthing information that people are looking for. Users will ask ChatGPT about the best humidifiers on the market, and when the model delivers results, people will have no choice but to click on the item link and complete the purchase online.
AI agents currently do not have the ability or trust infrastructure to make people and banking institutions feel safe enough to unleash their money on someone. Companies and other industry players understand that in order to enable agents to pay for purchases, there must be a common language shared between the model providers and the agents, the bank, the merchant and, to a lesser extent, the buyer.
And so, three competing standards for brokerage trading have emerged in recent weeks: Googling announced the Agent Payment Protocol (AP2) with partners such as PayPal, American Express, Mastercard, Salesforce and ServiceNow. Shortly afterwards, OpenAI And Stripe debuted the Agentic Commerce Protocol (ACP), just this week Visa launched the Trusted Agent Protocol (CRANE).
All of these protocols aim to give agents the layer of trust they need to convince banks and their customers that their money is safe in the hands of an AI agent. But this can also create walled gardens, showing just how immature the agency business really is. This is one problem that could cause companies to rely on a single chat platform and the agentic payment protocol it runs on, rather than interoperability.
How are they different?
It is not new for players to propose different standards. It typically takes years for the industry to coalesce around a single standard, or even adopt different protocols and find a way to harmonize them. However, the pace of innovation in business has continued to do so.
Soon, MCP became the de facto channel for tool usage identification, and most companies began setting up or connecting to MCP servers. (To be clear, it’s not a standard yet.) But having three different potential standards can slow that process down a bit, because it’s harder to settle on one standard when there’s so much to choose from.
These protocols are all intended to prove authorization. Both AP2 and TAP rely on cryptographic proofs to demonstrate that an agent is acting on behalf of an individual. For TAP, agents are added to an approved list and given a digital key that identifies them. AP2 uses a digital contract that serves as a proxy for human approval of the agent. OpenAI’s ACP doesn’t require too much infrastructure change, with ACP essentially acting as a courier for the trader as the agent relays information to the trader.
Walled gardens
These three protocols will ideally work on different chat platforms, but that is never guaranteed, especially if the chat platform’s biggest competitor has its own protocol. One danger of competing protocols is that they can create wall gardens, where they only work on specific platforms.
Businesses face the problem of being locked into a platform and an agentic payment standard that cannot work with another platform. Organizations not only receive the product recommended by the agent, but are usually the known sellers and must be able to trust that the agent contacting them is acting on behalf of a customer.
Louis Amira, co-founder and CEO of agent commerce startup Circuit and Chisel, told VentureBeat that while this presents opportunities for interoperability layer companies like his, it could cause confusion for enterprises.
“The better the protocol proposals become, the more likely they are to end up being walled gardens that are very difficult to interoperate,” Amira said. “We suspect they’re going to duke it out over the next few years, and the more they duke it out, the more you kind of need someone to be among them all.”
Unlike the Internet, where anyone can use any browser to access a website, chat platforms remain very segregated, thanks in large part to the TCP/IP standard. I usually use ChatGPT (because it’s installed on my laptop and I don’t need to open a new tab), so if I want to see how Gemini will handle my query, I have to open Gemini for that – the same works for anyone who shops via a chatbot.
The number of protocol proposals underlines how far we are from enabling retail agents. The industry is still deciding which standard to follow, and no matter how many Walmarts integrate with ChatGPT, it’s all moot if people don’t trust the model or agent to handle their money.
Hopefully take the best features
The best thing companies can do now is experiment with all the protocols and hope that a winner emerges. Ultimately, there could be one agentic commerce protocol that makes the most of every proposition.
For Wayne Liu, chief growth officer and president of the Americas at Perfect Corp., having multiple protocol proposals just means more learning.
“This is where the importance of open source exists, because it will be the driving force to bring everything together,” Liu said.
It would of course be interesting to see in the coming weeks if there would be just three competing trading protocols. After all, there are some major retailers and chat platforms that can still throw a spanner in the works.
