By Deepak Bagada, CEO at SaaSNext & Principal AI Architect.

Alipay's Agentic Commerce: Ah Bao, AHA Protocol & the Agent Shelf

On August 17-18, 2026, in Hangzhou, Alipay unveiled what it is calling China's first full-stack agentic commerce platform. This is not a chatbot bolted onto a payments app, and it is not another "AI assistant inside an app." It is a claim about how goods and services will be discovered, compared, and transacted in the agent era: the unit of distribution is shifting from the app icon to the agent-callable skill.

Here is the setup. Any merchant — whether they have an engineering team or not — can now convert their pages, products, and service workflows into agent-ready Skills and MCP tools. Any merchant that does have AI capability can go further and build and manage agents that handle recommendations, memberships, and conversion. On the consumer side sits Ah Bao, the Alipay consumer AI agent that launched in June 2026, already connected to more than 10,000 AI-enabled services across eight categories — transport, dining, culture and tourism, government livelihood services, and more — with marquee brands like KFC, McDonald's, Mixue, Luckin, Gaode Maps, Didi, Deppon, YTO, and Hive Box live on the shelf. Ah Bao is already embedded in five smartphone brands that together hold a reported 70%+ share of the Chinese handset market, and in 16 automakers' vehicles.

What does the platform actually do, and why should a merchant care? Let's walk the stack.

The two on-ramps for merchants

The clever part of the launch is that it meets merchants where they are. Alipay defines two distinct on-ramps:

No AI, no problem. A mom-and-pop restaurant or a regional logistics firm has no ML team and no budget for one. The platform lets them describe their existing pages, catalogs, and service workflows, and the system converts those into agent-ready Skills and MCP tools. The merchant's ordinary website or mini-program effectively becomes an API surface that any agent can call. This is the "digitalize your workflow once, get discovered by a million agents" pitch, and it is the piece that makes the network effect possible — most of the world's commerce runs on businesses that will never write a tool definition by hand.

AI-first merchants go deeper. A brand with an AI team can build and manage its own agents on the platform: recommendation agents, membership agents, customer-service agents, conversion agents. The platform provides the runtime, the data plumbing, the payment rails, and the audit trail; the merchant provides the agent's strategy. In effect, Alipay is offering every brand its own autonomous storefront that lives inside a million other agents' toolboxes rather than inside an app store.

The framing that matters is the second one. Twenty years ago the fight was for shelf space in a physical store; ten years ago it was for app-store ranking; in 2026 the fight is for presence in the agent's tool list, and Alipay is building the shelf the agents shop from.

Ah Bao: the consumer agent at the center

Ah Bao launched in June 2026 as Alipay's consumer AI agent, and the new platform is designed to make it the default way Chinese consumers reach services. The reported numbers are the story: 10,000+ AI-enabled services across eight categories, spanning transport, dining, culture and tourism, and government livelihood services, with household brands such as KFC, McDonald's, Mixue, Luckin, Gaode Maps, Didi, Deppon, YTO, and Hive Box already connected. The distribution footprint is the aggressive part: five smartphone brands with a combined 70%+ market share ship Ah Bao on the device, and 16 automakers embed it in their vehicles. When the assistant lives in the phone's native layer and in the car, it stops being an app the user opens and becomes an environment the user lives in.

The demo Alipay ran at the Hangzhou event is the clearest picture of where this goes. A user asks a StepFun phone agent: "find the nearest EV charger and order coffee to my home." The agent consults the charging network, finds the nearest available station, places the coffee order through the coffee merchant's skill, confirms delivery to the home address, and presents a single ready-to-confirm order combining both transactions. No app switching, no jumping between nine mini-programs, one confirmation. That one interaction is the entire thesis: the agent coordinates across merchants, and Alipay sits underneath providing identity, payment, and trust.

The AHA protocol suite and ACT 2.0

What makes the coordination possible is a protocol layer, and this is the part technically-minded readers should study. Alipay is releasing the AHA protocol suite, three related standards:

On top of the suite sits ACT 2.0, the agent-commerce transaction protocol, which adds the four things a commerce transaction needs that a plain tool call does not: delegation (the agent acting on the user's behalf), audit trails (a complete record of what the agent did and why), intent verification (confirming the user actually wants this transaction before money moves), and payment channels (the rails the money flows over). ACT 2.0 is where the platform earns its keep: without audit trails and intent verification, an agent that can spend money is a liability; with them, it is a checkout clerk that never sleeps.

There is also the Agentar ecosystem edition, aimed at enterprises: 200+ specialized skill packages, digital expert agents for verticals, and A2A multi-merchant coordination so that agents from different merchants can negotiate with each other — the coffee shop's delivery agent talking to the courier's pickup agent without a human in the middle.

The distribution math for a merchant

The economic question a merchant asks is simple: does this beat my current customer acquisition cost? The platform's pitch is that the agent shelf replaces paid acquisition with presence. The table below lays out the comparison a CFO in Mumbai, Delhi, or Bangalore should actually price against when evaluating an agentic shelf for their own market:

The unit-economics shift is real and measurable. On the old model, CAC compounds: every rupee of growth buys more of the same rented attention. On the agent shelf, the skill is built once and re-discovered by every agent, every time, at near-zero marginal distribution cost. The trade-off is margin: the platform takes its cut through payment fees, token incentives, and platform fees — which is why the incentive package matters.

The incentive package

Alipay is spending to seed the flywheel. Reported incentives include 100 million free tokens per user for building and testing skills, direct subsidies for merchants who onboard early, and reduced payment fees for agent-originated transactions. Twenty-plus partners are already in, including Qwen, Huawei, and BYD — model provider, device maker, and carmaker respectively, which is a tidy summary of the three vectors the platform needs to cover. The strategic intent is stated plainly by Ant Group CEO Cyril Han, who told the audience that agentic commerce will grow quickly over the next 6-12 months. The platform is a land-grab timed for a market that is about to tip.

The stack, end to end

+-----------------------------------------------------------------+
| CONSUMER LAYER                                                   |
|  Ah Bao in phones (5 brands, 70%+ share) + 16 automakers' cars   |
|  + third-party agents (StepFun, Qwen, ...)                       |
+-----------------------------------------------------------------+
              |  natural-language requests
              v
+-----------------------------------------------------------------+
| AGENT LAYER                                                      |
|  Agent Router / Planner   ->  AHA Agent Hub Access (discovery)   |
|  ACT 2.0: delegation | audit trail | intent verification | pay   |
+-----------------------------------------------------------------+
              |  standardized skill calls
              v
+-----------------------------------------------------------------+
| SKILL SHELF (Agentar)                                            |
|  200+ skill packages | digital expert agents | A2A coordination  |
|  Merchant Skills: no-code conversions + AI-built agents          |
+-----------------------------------------------------------------+
              |
              v
+-----------------------------------------------------------------+
| MERCHANT LAYER                                                   |
|  KFC | McDonald's | Mixue | Luckin | Gaode | Didi | Deppon      |
|  YTO | Hive Box + 10,000+ services across 8 categories          |
+-----------------------------------------------------------------+

Every layer above the merchant is designed to be model-agnostic and protocol-driven. The skills are MCP-compatible, so they are callable not just by Ah Bao but by any agent that speaks the protocol — which is the same bet the MCP directory has been tracking since the standard went mainstream. If you are building agent infrastructure, this is the architecture pattern to copy: a neutral protocol layer, an audited transaction layer, and a permissive merchant on-ramp.

A merchant skill manifest

For the technically inclined, here is what a merchant skill actually looks like — a JSON manifest that describes a service, which the platform then exposes as an MCP tool. This is the "no-code merchant" path reduced to its artifact:

{
  "skill": "mixue_order_coffee",
  "version": "1.0.0",
  "protocol": "aha-skill",
  "category": "dining",
  "description": "Order a drink for pickup or delivery to a given address.",
  "auth": "alipay-user-token",
  "payment": { "channel": "alipay-act", "intent_required": true },
  "tools": [
    {
      "name": "create_order",
      "description": "Place an order for a menu item.",
      "input_schema": {
        "type": "object",
        "properties": {
          "item": { "type": "string", "description": "Menu item identifier" },
          "qty": { "type": "integer" },
          "delivery_address": { "type": "string" }
        },
        "required": ["item", "qty"]
      }
    }
  ],
  "observability": { "audit_retention_days": 180, "trace": true }
}

Any MCP-aware agent can now discover, invoke, and — because of ACT 2.0 — safely pay through this skill. The same file, published to the hub, is what makes a merchant discoverable across every agent that connects. That is the whole revolution in one artifact: a merchant that could never afford an app now owns an autonomous storefront.

Why this matters beyond China

There are three reasons to study this launch even if you never sell a cup of tea in Hangzhou. First, the protocol-first architecture — skill standards, hub access, device rendering, and a transaction layer — is the same four-layer pattern every agentic commerce platform will converge on; the workflow library at Daily AI World has been mapping these patterns as they emerge. Second, the merchant on-ramp asymmetry: serving merchants with zero AI engineering is what unlocks a marketplace; platforms that only serve AI-savvy brands will stay niches. Third, the economics: when 70% of a market's devices ship the agent pre-installed, distribution stops being something you buy and becomes something you're included in — and the CAC math every growth team has run for a decade stops applying. The latest AI news desk will be watching the 6-12 month window Cyril Han flagged, because if agentic commerce tips in China at this scale, the rest of the world's commerce stack is already behind.

Disclaimer: Figures on device reach, service counts, token incentives, and partner numbers are as reported by Alipay and event coverage of the August 17-18, 2026 Hangzhou launch; validate specific terms against the official platform documentation before committing budget.