Ant Digital Unveils Agentar 2.0 at WAIC 2026, a "Super-Factory" for Business Agents

On July 17, 2026, at WAIC 2026 in Shanghai, Ant Digital Technologies (蚂蚁数科, the digital arm of Ant Group) unveiled a "super-factory for business agents" and the Agentar 2.0 platform, with nearly 200 role-level digital expert templates and hundreds of pluggable, ready-made agent tools.

What Is Agentar 2.0

Agentar 2.0 is an Ant Digital platform where the company assembles ready-made AI agents for specific job roles, so businesses don't have to build them from scratch. The first release includes role-level digital expert templates and subscribable Skills-level tools, each with its own professional knowledge, business workflow, and ability to call tools.

• Announcement date: July 17, 2026, at WAIC 2026 in Shanghai
• Nearly 200 preinstalled role-level digital expert templates
• Hundreds of subscribable Skills tools out of the box
• Three pillars of the approach: industry focus, scenario focus, ecosystem focus
• Foundation: Ant Group's industry-specific vertical models and more than 20 years of technology experience

WAIC 2026 ran under the theme "Smart Partners, Shared Future," and, according to 36Kr, the industry's focus is shifting from a race over model size to the value of applying models in real business processes. Ant is betting on turning components into APIs so that partners can plug in quickly and build out solutions for their own industries.

Why It Works in Finance

For financial scenarios, Ant Digital claims agent answer accuracy of 90–95% and agreement with a human expert's answer above 85%, according to company vice president Sun Lei. In his words, finance presents three barriers: security and compliance, professional accuracy (the hallucination problem), and proof of impact, i.e. ROI.

Ant's answer combines an industry-specific vertical model with engineering "scaffolding" built around the business process, plus a pay-for-results principle: the client pays for verifiable impact, such as higher conversion in marketing or lower losses in risk control. The company draws on its own experience in payments, wealth management, and insurance.

"Better to do nothing than to make a mistake," said

Sun Lei, vice president of Ant Digital Technologies, on financial regulators' requirements for AI deployment.

A Chinese Version of Harness

Ant Digital wants to build a Chinese, industry-specific version of Harness, the engineering scaffolding that turns a model's capabilities into a working business process. Where finance previously followed a "three-workshop" logic, the company defined four stages in Harness: intent recognition, planning, execution, and response formulation. Intent recognition is the key one: a customer asks about insurance, but the real request is portfolio selection or wealth management.

Ant Digital is extending the approach beyond finance into healthcare, energy, transportation, and aviation. In solar power, for example, the company uses a time-series model that forecasts electricity output based on weather and conditions at the site. The evaluation system plays a key role: on one large city commercial bank's project, Ant ran 28 rounds of testing before confirming the agents were ready.

What This Means

Ant Digital is shifting the competition from "model size" to how well a model is embedded in real business processes. The company is betting not on a universal chatbot but on industry-specific agents with verifiable ROI and pay-for-results pricing; over the next 6–12 months it promises to expand its agent matrix and gradually open up industry-specific Harness capabilities.

Frequently Asked Questions

What is Agentar 2.0?

Agentar 2.0 is an Ant Digital Technologies platform for quickly launching business agents. The first release includes nearly 200 digital expert templates for specific job roles and hundreds of subscribable Skills tools that are ready to use out of the box.

What accuracy does Ant claim for its industry agents?

According to the company, in professional financial scenarios agent answer accuracy reaches 90–95%, and agreement between their answers and those of a human expert is above 85%.