您正在使用IE低版浏览器，为了您的雷峰网账号安全和更好的产品体验，强烈建议使用更快更安全的浏览器
此为临时链接，仅用于文章预览，将在时失效
0
导语：实测灵光2.0，真能帮我管事
机票和酒店订好，还是忍不住隔几天查查价格。降了，算算退改费用，看要不要重订；没降，过两天再看。花不了多少时间，却总得惦记着。
为了追剧开了会员，剧看完就没再打开，自动续费却一直没停。直到某天翻账单，才发现又扣了几个月的钱。
这些事务单次处理并不复杂，却一直需要反复关注。过去，AI可以帮忙查一次价格、分析一份账单，跟进工作仍然留给用户。现在，灵光想把这部分也接过去。
10月27日，蚂蚁集团发布灵光2.0，以 “能管事的个人AI助理”为定位，成为国内大厂首批落地的Personal Agent产品。
它想带来的变化很直接：不再是被动响应指令、完成单次任务，而是能主动发现需求、持续帮你管理事务。
Muse之后，全球Personal Agent竞争开始加速，国内大厂也在密集布局。灵光率先交出产品答卷，从盯票价、理账单等日常小事入手，让AI开始替用户持续管事。
从“一次帮忙”到“长期管事”，灵光迈向下一程
去年灵光1.0上线时，让人印象最深的是它交付答案的方式——用户提一个需求，灵光30秒内就能手搓一个带3D、带图表的可交互小应用，让许多想法能快速落地。
但交付一个答案或应用之后，服务往往也就告一段落。需求变了、信息更新了、计划需要调整，用户还得回来重新交代。
灵光2.0则将服务从“办事的短跑”变成了“管事的长跑”。用户在交流中将需要持续关注的事项纳入“帮我管”，灵光就能持续跟进。
以雷峰网(公众号：雷峰网)为例，编辑每天都要持续关注海外AI公司的产品发布、技术更新和行业讨论。但是新闻事件极多、消息又复杂、热度变化也极快，同一事件还会出现大量转述。自己盯要花很多时间，但是不盯又不行。
这种需要持续关注的事情，就可以直接交给灵光，让它筛选出值得编辑进一步核查和追踪的事件。在“帮我管”中，可以看到对应的海外AI热点监控任务，点进去就能查看每天搜集到的消息进展。
生活里的事，也可以交给灵光来做。
最近深圳倡导APEC会议期间企业带薪休假，小编也蠢蠢欲动想带小孩出去逛逛，于是让灵光帮看看从深圳到北京、从深圳到哈尔滨还有没有便宜的机票。
当时灵光就告诉小编，根据它查到的记录，周六的时候机票已经在涨价，让我看到合适的直接出手。但我不信邪，让它再盯一盯。
两天后，灵光反馈，飞北京的低价机票继续在陆续跳档，哈尔滨的机票价格还比较稳定，建议尽早出手。
除了这些长周期任务，灵光还能帮忙分析小编的支付宝账单。之前支付宝花呗每个月都有一大堆账单，一直懒得去核对，灵光出来后，小编也让它给我仔细去拆了一下账单明细。
意外发现，有一个华为视频的自动订阅一直忘记取消了，幸亏灵光把它“点”出来了。
现在，灵光的管事能力已经进入了小编的日常生活。比如想买一个预算6000元左右的西门子洗碗机，它可以在京东、淘宝、拼多多、抖音上持续比价；小孩想英语启蒙，它可以帮忙做课程计划。
从用户交代的目标，到交流中逐步形成的需求，再到授权信息中尚未被留意的问题，灵光尝试接住的事务开始变多。
过去一年，Claude Code、Codex等Agent让人看到了AI执行复杂任务的能力。用户交代目标，系统自主规划步骤、调用工具，执行非常复杂的任务，以完成某个具体任务为目标。
但用户的实际生活中，大量的需求没有明确的结束时间，外部条件也会不断变化。
作为Personal Agent的灵光，正在尝试承接的正是这一段：AI能够长期记住一件事情的来龙去脉，持续关注后续的变化，并且在需要的时候主动采取行动，让用户可以少盯一些事、少操一些心。
Muse之后，灵光给出了怎样的中国解法？
灵光2.0的升级，发生在全球AI厂商重新寻找个人助理产品形态的窗口期。
9月，Meta推出Muse，将长期记忆、后台工作和跨应用执行组织进同一个产品，随后Meta继续引入购物、旅行等服务连接，扩大助理能够承接的需求范围。
紧接着国内厂商也加快了动作：字节推进代号为Spell的个人智能体项目，计划与豆包结合；腾讯则推出云端个人助理LightVela。
各家大厂开始争取同一种用户关系：AI理解需求之后，能够在用户离开对话时继续工作。但从哪里理解用户、怎样组织后续行动，各家选择并不相同。
海外用户的信息和服务分散在邮箱、日历、独立网站和订阅账户中，所以Muse 更强调“自主执行”的委托关系。例如，用户希望降低一笔服务账单，可以把目标交给Muse，由它通过浏览器查找方案、推进相关操作，遇到需要批准的步骤再联系用户。
国外用户信息和服务散落在各种App和小程序中，单次操作已经很方便，但跨平台信息、长期事项仍需要用户自己惦记。
所以灵光选择的是把散落在不同平台、需要长期记住和反复查看的事情接过来，让用户交代之后，可以少操心。
支撑这一产品的核心，是灵光2.0版本中的跨会话记忆、长程任务调度和工具连接能力的配合。
以盯低价机票为例，跨会话记忆让灵光记住出行日期、预算和时段要求；长程任务调度让它在后续合适的时间继续查询，工具连接则帮助它获取新的航班和价格信息。
记忆负责记清楚，调度负责接着做，工具负责拿到新信息。系统再根据这些信息判断，哪些变化值得提醒用户。云端独立运行环境，则让这些工作可以在用户退出聊天后继续进行。
用户不用每次从头介绍情况，也不用为了获得新进展，反复回来问一句。
灵光2.0的优势，还在于其用持续管事的技术能力，连接中国特有的数字服务生态。
它采用了两种互补的方式：
一部分常见服务直接内置，包括支付宝账单、会员、出行，以及钉钉、飞书、邮箱、米家等，降低用户的配置成本；更分散的需求，则通过独立电脑、浏览器，以及MCP、CLI等开放工具继续扩展。
内置服务让产品可以从日常需求开始，开放工具则为长尾任务留下空间。
灵光因而可以从蚂蚁熟悉的服务起步，又通过办公平台、邮箱和智能设备连接，向支付宝之外延伸。它尝试围绕用户的一件事组织工具，而非将工作限定在某个应用内部。
这也是灵光区别于Muse的独特解法：把个人智能体的能力，接进中国用户熟悉的账户、应用和服务生态，让持续管事落到真实生活里。
从“管事”到“托付”，蚂蚁的优势在哪里
当AI开始长期管事，用户与它的关系也会变重。
蚂蚁长期在支付和生活服务积累的能力，为更好连接服务与用户提供了基础。
目前，蚂蚁已有大量的商家协作关系和业务经验，灵光可以沿着现有网络推进这些改造，不需要为每项需求重新寻找服务方、建立合作，这是数字服务生态能够转化出的实际优势。
当服务进一步走向交易，面向AI的支付基础设施也会发挥作用。
AI帮用户找到合适的商品后，需要完成付款；商家向AI应用提供服务，需要收款；开发者开放工具和能力，也需要计量、对账和结算。而支付宝AI支付的布局，正在覆盖智能体支付、AI应用收款，以及工具调用的计量与结算，为这些需求提供支持。
从这个角度看，蚂蚁的机会来自几项积累的共同作用：灵光探索个人智能体的产品与技术，可信机制支撑授权和执行，服务生态提供实际供给，AI支付能力衔接交易。
个人智能体要走进中国人的日常生活，需要将前沿技术接入本土的账户体系、应用与服务网络，建立用户愿意长期托付的服务关系。
而能将这些环节串起来，正是蚂蚁多年积累形成的差异化优势。
当AI开始管事，服务有了新的新想象
进入Personal AI时代，人与服务的连接方式也有了新的可能。
过去，用户先明确需求，再寻找应用和商家；未来，AI可以长期理解目标、关注变化，在合适时机组织服务。
商业竞争也将从模型和单个应用，延伸到技术能力、服务生态与用户信任的协同。
所以，灵光是观察AI Agent行业变化的一个重要切口。它探索的“持续管事”，有机会把蚂蚁过去围绕交易建立的连接，延伸为围绕用户长期需求展开的服务。
Personal AI的商业想象，正来自这种更持续的用户关系。
对蚂蚁而言，这是多年服务生态积累被进一步激活的机会；对普通人而言，价值则更朴素：那些一直挂在心上的小事，有人记着、接着办，自己可以少操一点心。
雷峰网原创文章，未经授权禁止转载。详情见转载须知。
为了您的账户安全，请验证邮箱
请验证您的邮箱
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
You are using a low-end IE browser and strongly recommend a faster and safer browser for your Thunder Peak account security and better product experience.
This is a temporary link, only for article preview, which will expire at the time
0
Introduction: Visual light 2.0.
The airline tickets and the hotel were booked and prices were checked several days later. It's down, it's a refund, see if it's rescheduled; it's not down, it'll be read in two days. It won't take much time to remember.
It's been open to members in pursuit of the show, it's never opened again, and the automatic renewal is still on. It was not until one day that the bill was turned over that it found that it had been withheld for months.
These services, which are not complex on an individual basis, continue to require repeated attention. In the past, AI had been able to help check prices and analyse a bill, and follow-up was still left to users. Now, Spirit Light wants to take this part.
On 27 October, the ants group publish luminous 2.0, positioned as a “manager's personal AI assistant”, became the first PERSONAL Agent product to land in large factories in the country.
The changes that it wants to bring are straightforward: it is no longer passive response to instructions, individual assignments, but the ability to proactively identify needs and continue to manage your affairs.
After Muse, global competition for PERSONAL Agent began to accelerate, and large domestic factories are also in an intensive configuration. Luang has taken the lead in handing over product responses, starting with day-to-day minor issues such as ticket prices, billing, etc., and allowing AI to continue to supervise users.
From "one-time help" to "long-term management", the light goes to the next.
Last year, when the light came on line, the most impressive was the way it delivered the answer -- – Users demand a 3D, 3D-based, graph-based, small, interactive application within 30 seconds, so that many ideas can land quickly.
However, when an answer is delivered or applied, the service often ends. Demand has changed, information has been updated, plans need to be adjusted and users have to come back and reciprocate.
The light 2.0 transforms the service from a “do-good sprint” to a “manager's run”. Users will need to keep their concerns under “help me through” in their communication so that the light can follow up.
Using the example of the Thunder Peak Network (WeChat Official Account:Rape Peak Network), editors are constantly following the products of Overseas AI publish, technology upgrading and industry discussions on a daily basis. However, the number of news events, the complexity of the news, and the speed of the heat changes, and there will be a lot of recitation of the same event. It takes a lot of time to look at yourself, but not without it.
Such a matter requiring continued attention can be referred directly to the light, allowing it to screen events that merit further editorial verification and tracking. In “Help me take care of” you can see the corresponding AI hotspot surveillance mission from overseas, and then you can click in and see the progress of the information collected every day.
Things in life can also be given to the light.
Recently, Shenzhen advocated for a paid business vacation during the APEC conference, and Xiao Jing wanted to take the kids out to see if there were any cheap tickets from Shenzhen to Beijing and Shenzhen to Harbin.
At that time, Liang Xinjiang told him that, according to the records it had found, the airfare was already rising on Saturday, so I saw the right direct hand. But I don't believe in evil, let it stare.
Two days later, light feedback continued to show that low-priced air tickets to Beijing continued to drop, and that the price of Harbin ' s air tickets was stable, and it was recommended that they should be available as soon as possible.
In addition to these long-cycle missions, the light can also assist in the analysis of the small authors ' payment bills. I've been paying a lot of bills every month, and I've been tired of checking it out, and when the light came out, the editor let me take it down.
Unfortunately, an automatic subscription to a Chinese video has been forgotten to be cancelled and fortunately the light has come out of it.
Now, the management of the light has entered the daily life of the children. For example, a Siemens dishwasher, with a budget of about $6,000, can compete continuously on Kyoto, poaching, and Douyin; children can help with the curriculum if they want to start with English.
From the point of view of the user, to the need to evolve in the communication and to the issues that have not yet been noticed in the authorized information, the number of services that the light attempts to catch has grown.
Over the past year, Agent, Claude Code, Codex and others have shown AI ' s ability to perform complex tasks. Users are targeted, the system is autonomous in planning steps, using tools and carrying out very complex tasks, with the objective of achieving a specific task.
In practice, however, a large number of needs do not have a clear end time and external conditions change.
As the inspiration for the Personal Agent, it is the paragraph that is being tried: AI is able to keep track of one thing for a long time, keep an eye on subsequent changes, and take action when needed, so that users can look less at things and worry less.
What kind of Chinese solution did Shingwang give after Muse?
The upgrade of Psychiatry 2.0 took place in the window period when global AI manufacturers were re-searching the personal assistant product form.
In September, Meta launched the Muse, bringing long-term memory, back-office work and cross-application implementing organizations into the same product, after which Meta continued to introduce services such as shopping, travel and so forth to expand the range of needs that the assistants can take on.
The move was also accelerated by domestic manufacturers: the byte Propulsion Personal Z42Z Project, code-named Spell, planned to be combined with Doubao, and the Tencent introduced LightVela, cloud personal assistant.
Major plants are beginning to seek the same user relationship: AI is able to continue its work when users leave the dialogue after it understands the needs. However, the choice is not the same from where users are understood and how follow-up is organized.
The information and services of overseas users are scattered in mailboxes, calendars, stand-alone websites and subscription accounts, so Muse places more emphasis on “self-execution” as a commission. For example, users would like to lower a service bill by giving the target to Muse, which will search the programme through the browser, advance the operation and contact the user with steps requiring approval.
Foreign user information and services are scattered over a variety of App and small programs, and single operations are already convenient, but cross-platform information and long-term matters still require user thinking.
So the light chooses to be less concerned when it connects things that are scattered across different platforms and that need to be remembered and viewed over time and over time.
At the core of this product is a combination of cross-session memory, long-range tasking and tool connectivity in version 2.0 of the light.
In the case of low-priced air tickets, for example, the cross-session memory allows the light to remember the travel date, budget and timing requirements; the long-range task schedule allows it to continue its queries at a suitable subsequent time, while the tool connection helps it to obtain new flight and price information.
Memory is responsible for remembering, movement is responsible for following up and tools for obtaining new information. On the basis of this information, the system will then judge what changes are worth reminding users of. The cloud operates independently of the environment, which allows the work to continue after the user exits the chat.
Users do not have to start from scratch and do not have to come back again and again to make new progress.
The advantage of luminous light 2.0 is also that it connects China ' s unique digital service ecology with the technical capacity of continuous supervision.
It follows two complementary approaches:
Some of the common services are directly embedded, including the payment of valuable bills, membership, travel, as well as nails, flybooks, mailboxes, rice houses, etc., reducing the cost of the user ' s configuration; more dispersed demand continues to expand through stand-alone computers, browsers, and open tools such as MCP, CLI.
The built-in services allow the products to start from day-to-day needs, while open tools leave space for long-term assignments.
The light can thus begin with services familiar to an ant and extend beyond the payment treasures by connecting to office platforms, mailboxes and smart devices. It tries to organize a tool around a user, rather than confining the work to an application.
This is also the unique solution of the light from Muse: The ability of the individual AI agent to feed into the well-known accounts, applications and service ecology of Chinese users so that the continuous management can be put into real life.
From "supervising" to "trusting." What's the advantage of an ant?
As AI begins to operate for a long time, the relationship between the user and it will become heavier.
The ability of ants to accumulate over time in payment and living services provides a basis for better connectivity between services and users.
At present, the ants have a great deal of business collaboration and business experience, and the light can drive these transformations along existing networks without the need to find a new provider for each demand and to establish cooperation, which is the practical advantage that digital service ecology can transform.
As services move further towards trade, the payment infrastructure for AI also works.
When AAI helps users to find suitable commodities, they need to complete the payment; businesses provide services to AI applications, requiring collections; and developers open their tools and capabilities, requiring measurement, reconciliation and settlement. The layout of the payments made by Bao AI is supporting these needs by covering AI agent payments, AI application receipts, and tool call measurement and settlement.
From this perspective, ants ' opportunities come from several cumulative co-effects: The light explores the products and technologies of the individual AI agent, a credible mechanism that underpins authorization and implementation, the ecological provision of services, and the AI capacity to pay trade-off.
In order to enter the daily lives of the Chinese, individuals AI agent need to connect forward technology to the local system of accounts, applications and service networks, and to establish service relationships that users are willing to provide over time.
And it's the differential advantage that ants have accumulated over the years.
When AI started running, the service had a new idea. Elephant.
In the era of Personal AI, there are new possibilities for people to connect to services.
In the past, users identified needs before looking for applications and businesses; in the future, AI could understand goals over time, focus on changes and organize services at the right time.
Business competition will also extend from models and individual applications to synergies between technological capabilities, service ecology and user trust.
So light is an important entry point for observing changes in the AIAgent industry. The “continuing stewardship” it explores has the opportunity to extend ant's past connection around the transaction to service delivery around long-term user needs.
Personal AI ' s business imagination is coming from this more continuous user relationship.
For ants, this is an opportunity for many years of service eco-accumulation to be further activated; for ordinary people, the value is more modest: Those little things that have been on your mind, those who remember and go on, can do less for themselves.
Original Ray Peak article, unauthorized reproduction. For further information, see the information below.
For your account security, please verify the mailbox.
Please verify your mailbox.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.