When Walmart and OpenAI announced that the retailer would integrate with ChatGPT, the question arose of how quickly OpenAI could deliver on its promise of agents buying things for people. In the fight for AI-powered commerce, one of the biggest hurdles is getting agents to complete transactions safely.
There are more and more chat platforms like ChatGPT browser replacement and does a great job of revealing the information people are looking for. Users will ask ChatGPT about the best humidifiers on the market, and when the model returns the results, they have no choice but to click on the item link and complete the purchase online.
Artificial intelligence agents currently do not have the capabilities or trust infrastructure that would allow people and banking institutions to feel sheltered enough to allow themselves to exploit someone else’s cash. Enterprises and other industry players understand that to enable agents to pay for purchases, there must be a common language between agent models and providers, the bank, the seller and, to a lesser extent, the buyer.
Over the past few weeks, three competing standards for agent trading have emerged: Google announced Agent Payment Protocol (AP2) with partners including PayPal, American Express, Mastercard, Salesforce and ServiceNow. Shortly thereafter OpenAI AND Stripe debuted the Agentic Commerce Protocol (ACP) this week Visa launched Trusted Agent Protocol (GET).
All of these protocols are intended to provide agents with the layer of trust they need to convince banks and their customers that their money is sheltered in the hands of an AI agent. But they can also create walled gardens, showing how immature agent trading really is. This is a problem that could result in enterprises relying on a single chat platform and the agent payment protocol it runs on rather than interoperability.
How do they differ?
It’s nothing up-to-date for players to propose several standards. It usually 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 enterprises has moved the needle in this regard.
Quite quickly, MCP became the de facto channel for identifying tool usage, and most companies began setting up or connecting to MCP servers. (To be clear, this is not standard yet). However, having three different potential standards can leisurely down the process somewhat because it is more hard to combine one standard when there are so many to choose from.
All these protocols are intended to prove authorization. Both AP2 and TAP rely on cryptographic evidence demonstrating that an agent is acting on behalf of an individual. With TAP, agents are added to an approved list and given a digital key identifying them. AP2 uses a digital contract that serves as a substitute for human agent acceptance. OpenAI’s ACP does not require much change to the infrastructure, where ACP essentially acts as a courier for the seller as the agent forwards the information to the seller.
Walled gardens
These three protocols work perfectly across different chat platforms, but this is never guaranteed, especially when the chat platform’s largest competitor has its own protocol. The danger with competing protocols is that they can create wall gardens where they only work on specific platforms.
Enterprises struggle with being stuck in a platform and an agent-based payment standard that won’t interoperate with another. Not only do organizations receive the product recommended by an agent, but they are also most often a registered reseller and need to be sure that the agent contacting them is acting on behalf of the customer.
Louis Amira, co-founder and CEO of agent trading startup Circuit and Chisel, told VentureBeat that while this creates an opportunity for interoperability layer companies like his, it could create confusion for businesses.
“The better the protocol proposals, the more likely they are to turn into walled gardens and it will be very difficult to work together,” Amira said. “We suspect they will be fighting each other for the next few years, and the more they fight, the more they will need someone to be under them all.”
Unlike the Internet, where anyone can access a website using any browser, thanks in gigantic part to the TCP/IP standard, chat platforms tend to remain very separate. I mostly exploit ChatGPT (since it’s installed on my laptop and I don’t need to open a up-to-date tab), so when I want to see how Gemini will handle my query, I have to open Gemini to do so – the same goes for anyone shopping through the chatbot.
The number of protocol proposals highlights how far we are from enabling commercial agents. The industry still needs to decide which standard to adopt, 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 cash.
Let’s hope you take advantage of the best features
The best thing that enterprises can do for now is to experiment with all the protocols and hope that a winner emerges. Ultimately, there could be a single agent trading protocol that leverages the best of each proposition.
For Wayne Liu, chief development officer and president of the Americas at Perfect Corp., having multiple protocol offerings simply means more learning.
“This is where open source plays such an important role because it will be the driving force in bringing everything together,” Liu said.
Of course, it would be captivating to see over the next few weeks if there will be just three competing agent trading protocols. After all, there are a few gigantic retailers and chat platforms that could still weigh in on the whole thing.