Alipay unveils AHA protocol to link AI agents across devices and services

Alipay President Jun Li showed how the fragmentation looks in daily life. He described a scenario where he tells "Abao," Alipay's super agent, "Order a car to my company and grab a coffee from Starbucks downstairs." The next day, he adds, "Send the invoice of yesterday's trip to my email." The tasks are handled by a chain of agents behind the scenes. Li said the goal is to turn one-time product connections into a reusable protocol, so agents can discover each other and hand off tasks.

Since a pilot in June, Abao has adapted over 10,000 services into intelligent capabilities across eight categories, including transportation, dining, tourism and government affairs. The AHA protocol has been integrated with five major smartphone vendors covering more than 70 percent of the market share, and with 16 car brands, while more than 60 automakers have signed intention agreements. AI glasses, smartwatches and large-model apps are being added. The first 50-plus services have also been connected to Alibaba's Qwen model.

The connection with OPPO illustrates the model. In July, OPPO's assistant Breeno hooked into nearly 200 life services via AHA, enabling users to complete tasks such as booking movie tickets or ordering food with a single voice command. The collaboration added over two million new active users to Alipay on OPPO phones. In the technical design, Breeno handles user intent and privacy, while Abao handles the transaction, keeping the two sides separate.

To bring merchants into the agent network, Alipay released a full-stack agent commerce infrastructure. Restaurants or other businesses without AI teams can hand their existing pages, products and service flows to Alipay's AI open platform, which converts them into standards such as Skills and MCP. Merchants with development capabilities can create their own agents. Since the platform opened in July, some merchants' development requests have already been booked into next year. Alipay also launched an incentive plan with discounted payment fees, free access for individual developers, computing subsidies and transaction incentives, plus access to 57 operational capabilities such as membership, coupons and credit.

Ant Group CEO Han Xinyi offered a timeline for agent commerce: some merchants will feel value within three months, more will join in about six months, and a real inflection point may emerge in the second half of next year. He described the growth potential with a "fourth power" idea: instead of 1.4 billion people, there will be 14 billion agents if every person owns ten. "User intents are one square, atomized services from merchants are another square," he said. "We believe the inflection point of demand is coming soon."

Executives from device makers echoed the transformation. OPPO vice president Tang Kai predicted that in one to two years, agent business models will move "from selling models, tokens and tools to selling tasks, results and productivity." vivo vice president Zhou Wei said, "The mobile internet sells reach; the agent era sells results." He said the biggest challenge is trust, proposing a principle: "devices manage identity, clouds manage capabilities, and platforms manage responsibility." He warned that if the new economy fails to form, it may get stuck in the "last meter" of payment and responsibility. Geely's chief engineering scientist Ren Xiangfei defined agents as "the autonomization of intent."

Beyond industry blueprints, multi-agent workflows are already appearing in developer communities. In a recent discussion on the Matrix community of the Chinese tech site SSpai, 380 participants compared coding agents such as Codex, Pi, Grok, Kimi-Code and Hermes Agent. One user described a workflow where ChatGPT web conversations feed directly into Codex's local Work app for code review and editing; another said he runs a terminal-based workflow using Codex, Pi, Grok and Kimi-Code across his Mac mini, MacBook and phone, with agents running in the background. These examples indicate that the move toward inter-agent collaboration is not limited to large platforms.