内容来源：https://aibusiness.com/agentic-ai/meta-joins-enterprise-agent-race-with-muse-wonder

内容总结：

Meta近日发布两款新型人工智能模型Muse Wonder与Muse Wonder Full，进一步拓展其个人AI代理Muse的能力边界。新模型基于Muse Spark模型驱动，可将法律、金融、健康等领域的专业知识应用于用户的工作场景，被视为Meta在消费级AI市场与Anthropic、OpenAI和谷歌展开竞争的重要举措。

Meta表示，上述模型是其推进“个人超级智能”战略的下一步。Muse作为一款AI代理，可为个人用户执行网页浏览、行程规划、购物及邮件整理等任务。新模型引入“主动记忆”和“主动聚焦”功能：前者可基于关键信息记住最重要内容并跨对话或任务延续，后者允许模型在首次尝试失败时探索替代方案。

华盛顿大学教授Chirag Shah指出，过去一两年AI行业正从聊天机器人转向代理化世界，用户可将部分任务委托给系统完成。随着Meta Muse、OpenAI Dots和Gemini Agent等产品涌现，AI正接近下一个重大里程碑。Gartner分析师Arun Chandrasekaran认为，Meta正努力引导模型在个人代理工作流中更高效，使其在日常任务中更加专业化。

然而，尽管面向消费者和企业的个人代理发布节奏加快，企业客户仍持谨慎态度。Shah表示，许多企业不确定能否信任代理正确行事，以及是否会做出有害或意外之举，同时对数据安全存有顾虑。此外，企业还需评估这些模型能否与现有应用及网络安全架构集成。Tekonyx创始人Sid Nag强调，集成能力是企业关注的核心。

对Meta而言，新模型有助于建立AI代理的盈利模式。此前Meta主要依靠免费开源AI模型支撑广告业务，Chandrasekaran指出，Meta无法持续免费提供模型算力，向消费者收费需要更强大、更个性化的AI代理。开发者使用该模型每百万输入token收费4美元，每百万输出token收费20美元；如需使用Muse Wonder Full，须订阅Meta首个专用编码代理Muse Code，月费分别为5美元、15美元、50美元和200美元不等。

除消费者市场外，Meta还瞄准使用Instagram、Facebook和WhatsApp运营的大量小企业，可在这些平台上为其提供AI代理服务。对于大型企业，Meta仍需与全球系统集成商及Salesforce等供应商合作，建立完整的分销渠道。

中文翻译：

由谷歌云赞助

选择你的首批生成式人工智能用例

要开始使用生成式人工智能，首先应聚焦于那些能够改善人类信息体验的领域。

该公司的新产品发布指向一个正变得更加个性化的AI市场，智能体和模型能够完成任务，但企业仍未准备好迎接AI的这一新阶段。

Meta于周一推出了面向个人和专业智能体任务的模型，借助其上月成功发布的Muse个人智能体的势头。

Muse Wonder和Muse Wonder Full扩展了Meta Muse的发布成果。Muse由Muse Spark模型驱动，是一款为个人执行任务的AI智能体，例如网页浏览、规划旅行、购物和组织电子邮件。

Meta表示，新模型通过将专业知识应用于用户在法律、金融和健康等领域工作中遇到的情境，扩展了Muse的能力。Meta称，这是这家拥有Facebook、Instagram和WhatsApp的母公司为与Anthropic、OpenAI和谷歌竞争而发布的首批前沿模型。Meta表示，这是将个人超级智能带给客户的下一步。

虽然这些模型主要面向消费者，但它们也是一股快速发展的趋势的一部分——能够在企业环境中运作的AI智能体。

“过去一两年一直是重点所在，我们正在远离聊天机器人，走向这个智能体世界，在这个世界里（我们）能够将一些任务委托给这些系统来完成，”西雅图华盛顿大学教授奇拉格·沙阿说。

他补充说，有了Meta Muse、OpenAI Dots和Gemini Agent，AI正处于下一个重大里程碑的边缘。其他流行的新型个人智能体包括来自SpaceXSI的Grok Bot、Instinct和Hermes。

如今隶属于OpenAI的OpenClaw仅在10个月前开创了个人智能体趋势，但当时很难使用。而这些智能体则不然。

通过Muse Wonder和名字颇具奇思妙想的Muse Wonder Full，Meta融入了其所谓的“主动记忆”和“主动聚焦”。利用主动记忆，模型可以根据此前接收到的关键信息记住最重要的内容，并将其贯穿于对话或任务之中。主动聚焦使模型能够在执行任务或实现目标的第一种方式不成功时，尝试替代方法。

“他们确实在努力引导这些模型在个人智能体工作流程中更加高效，”Gartner分析师阿伦·钱德拉塞卡兰说。“他们正在使这些智能体在日常任务中更加专业化。”

沙阿表示，这是朝着正确方向迈出的一步，但尽管最近面向消费者和企业的个人智能体发布节奏紧凑，企业并未完全买账。

“他们非常犹豫，”他说，并补充说许多企业不确定能否信任智能体去做正确的事，也不确定它是否会做出任何有害或意料之外的事情。

“信任是指它会如何处理我的数据，而不仅仅是完成任务。‘我们怎么知道Meta不会以危及我或我公司立场的方式查看我的数据？’”沙阿继续说道。“我所接触的公司，他们现在还没有能力将这一切落实到位。”

企业也可能对此类模型保持谨慎，因为他们需要确定能否将其与已经在使用的应用程序连接起来。

“归根结底，一切都关乎集成，”Tekonyx创始人兼分析师西德·纳格说。“这个特定功能是否能与现有应用程序和网络安全架构集成？”

“这些接口正在幕后运行，并连接到所有不同的应用程序。所有这些集成都将是企业关注的重点，”他补充道。

钱德拉塞卡兰表示，对Meta而言，Muse Wonder和Muse Wonder Full使这家社交媒体巨头能够为其AI智能体建立收入模式。此前，该公司一直专注于免费的开源AI模型，以构建其庞大的数字广告业务并扩展其商业生态系统。

“他们不能一直免费提供模型中的计算能力，”他说。“他们为AI智能体实现收入模式的方式之一就是开始向消费者收费，但要向消费者收费，他们需要提供能力更强的模型和更加个性化的AI智能体。”

对开发者而言，该模型的价格为每百万输入token 4美元，每百万输出token 20美元。想要使用Muse Wonder Full的开发者必须订阅Muse Code——Meta首个专用编程智能体。Muse Code的日常使用费用为每月5美元，高频使用为每月15美元，强力使用为每月50美元，最高使用为每月200美元。

除了消费者市场，Meta还瞄准了众多使用Instagram、Facebook和WhatsApp来运营业务的小企业，现在它可以在这些社交媒体平台上为这些企业提供AI智能体。

钱德拉塞卡兰表示，对于更大的企业，Meta仍需通过与全球系统集成商GSI和Salesforce等供应商合作来建立完整的分销渠道。

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
The company’s new releases point to an AI market becoming more personal, with agents and models completing tasks, but enterprises still aren’t ready for this new phase of AI.
Meta introduced models on Monday that target personal and professional agentic tasks, capitalizing on the momentum of its successful Muse personal agent, released last month.
Muse Wonder and Muse Wonder Full expand on the launch of Meta Muse. Powered by the Muse Spark model, Muse is an AI agent that does tasks for individuals, such as web browsing, planning trips, making purchases and organizing emails.
The new models expand Muse’s capabilities by applying specialized knowledge to circumstances users encounter at work in domains such as law, finance, and health. Meta said. These are the first frontier models the Facebook, Instagram and WhatsApp parent company has released to compete with Anthropic, OpenAI and Google. Meta said they're the next step in bringing personal superintelligence to its customers.
While the models are geared mainly toward consumers, they are also part of a fast-developing trend of AI agents capable of operating in enterprise environments.
“This has been the focus for the last year or two that we are moving away from chatbots and going to this agentic world where [we're] able to delegate some tasks that these systems can do,” said Chirag Shah, a professor at the University of Washington in Seattle.
With Meta Muse, OpenAI Dots and Gemini Agent, AI is on the verge of its next big milestone, he added. Other popular new personal agents include Grok Bot from SpaceXSI, Instinct and Hermes.
OpenClaw, now part of OpenAI, pioneered the personal agent trend only 10 months ago, but it was difficult to use. These agents are not.
With Muse Wonder and the whimsically named Muse Wonder Full, Meta incorporated what it calls Active Memory and Active Focus. Using Active Memory, the models can remember what matters most based on critical information they received before and carry that across conversations or tasks. Active Focus enables the model to try alternative methods of carrying out a task or a goal in case the first way is unsuccessful.
“They’re really trying to steer these models to be more effective at personal agentic workflows,” said Arun Chandrasekaran, an analyst at Gartner. “They’re enabling these agents to be more specialized at daily tasks.”
It’s a step in the right direction, but despite the staccato cadence of recent releases of personal agents for consumers and enterprises, enterprises are not fully sold on them, Shah said.
“They are very hesitant,” he said, adding that many are unsure of whether they can trust the agent to do the right thing, and whether it will do anything harmful or unexpected.
“Trust in the sense of what it would do with my data beyond just doing the task. ‘How do we know that Meta is not looking into my data in a way that jeopardizes my position or my company’s position?’” Shah continued. “The companies that I talk to, they’re not there yet to be able to put this in place now.”
Businesses could also remain cautious about models like these because they need to determine whether they can connect them to applications they already use.
“At the end of the day, it’s all about integration,” said Sid Nag, founder and analyst at Tekonyx. “Does this particular functionality integrate with existing applications and cybersecurity constructs?”
“These interfaces are going behind the scenes and connecting to all these different applications. All of that integration is going to be what enterprises are going to look at,” he added.
For Meta, Muse Wonder and Muse Wonder Full enable the social media giant to establish a revenue model for its AI agents, Chandrasekaran said. Previously, it had focused on free open source AI models to build its sprawling digital advertising business and expand its commercial ecosystem.
“They can’t keep giving away the compute in models for free,” he said. “One of the ways they can enable a revenue model for the AI agents is to start charging the consumer, but for them to charge the consumers, they need to deliver more highly capable models and more personalized AI agents.”
For developers, the model costs $4 per million input tokens and $20 per million output tokens. Developers who want access to Muse Wonder Full must subscribe to Muse Code -- Meta’s first dedicated coding agent. Muse Code costs $5 per month for everyday use, $15 per month for high usage, $50 per month for power, and $200 per month for maximum.
Beyond the consumer market, Meta is targeting the many small businesses that use Instagram, Facebook and WhatsApp to run their operations, and it can now offer AI agents on these social media platforms for those businesses.
For bigger enterprises, Meta still needs to build an entire distribution channel by partnering with vendors such as global systems integrator GSI and Salesforce, Chandrasekaran said.