# Ant LingGuang 2.0: Proactive Tasks, Permissions and China Availability

Ant's personal agent keeps tasks running between conversations, making cancellation and permission controls as important as memory.

By Paras Madan | 2026-10-28


Ant Group announced LingGuang 2.0 on October 27, upgrading its assistant to manage ongoing tasks after a conversation ends. The China rollout covers iOS, Android and the web. With authorization, the agent can monitor connected information and notify users when something needs attention.

## What changed in LingGuang 2.0

Ant's [company announcement](https://via.tt.se/pressmeddelande/4616995/ant-group-upgrades-lingguang-to-20-turning-its-ai-assistant-into-a-proactive-personal-agent?lang=en&publisherId=259167) describes cross-session memory, long-running scheduling and tool calls operating on independent cloud instances. Examples include spotting recurring subscriptions and watching travel prices after booking.

Connections include Alipay, email, DingTalk, Lark and Mi Home, with MCP and command-line tool support. Ant says its APASS security layer separates authorization keys from the agent; users can review, change or revoke permissions.

These are company-described capabilities, not independently established completion rates. The release does not specify an international rollout date, public developer API pricing or long-running task quotas.

## Persistence changes the product contract

A chat answer is usually evaluated when it arrives. An ongoing task needs to remain useful after the original circumstances change. A travel watch, for example, should stop being useful when the trip is cancelled, not keep generating alerts because the initial request still exists.

That makes **the lifecycle of a task** a central product feature. Users need to understand what is running, what it is watching, when it last checked and how to stop it. A polished conversation cannot substitute for those controls.

This is also why memory and permission are different things. Remembering an earlier preference should not silently extend the authority to act on a connected account. The product needs to keep the user's goal, available information and allowed actions aligned over time.

## What to check before connecting accounts

Start with a narrow, reversible task. Check whether the assistant's activity is visible and whether revoking access stops future work rather than merely hiding notifications.

For a price watch, a useful test is whether the system distinguishes a genuinely comparable offer from one with different conditions. For a recurring-payment review, identifying a charge and cancelling it are separate decisions. An assistant should make that boundary understandable.

These are evaluation criteria, not reports of failures in LingGuang. The available launch evidence does not establish how the product behaves in every interruption or changed-permission scenario. The [public product entry point](https://www.lingguang.com/) is where eligible users can check the current experience and account-specific availability.

## What matters after the first successful task

Persistent agents can reduce the need to reopen an app and ask the same question. They also create a new burden if users must repeatedly inspect whether yesterday's instructions still make sense.

The meaningful test is whether the assistant makes ongoing work easier to supervise as well as easier to start. A personal agent earns trust by keeping its authority visible when the user is no longer looking at the chat.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help AI founders explain product launches and technical decisions through clear content that developers and founders understand and share.

[Book a 30-minute call](https://cal.com/paras-madan/30min)