内容来源：https://news.microsoft.com/source/asia/features/with-microsoft-agent-365-one-startup-furthers-goal-of-making-ai-agents-more-intuitive/

内容总结：

在人工智能应用浪潮中，初创企业Mainfunc推出的Genspark平台正试图重塑人机协作模式。该公司联合创始人景埃里克（Eric Jing）日前宣布与微软达成战略合作，将其核心产品“Genspark超级智能体”接入微软最新推出的Agent 365系统，标志着AI助手正式迈入企业级应用新阶段。

作为行业变革者，景埃里克提出了“直击业务成果”的人机交互理念。与传统聊天机器人不同，Genspark用户可通过自然语言指令直接生成演示文稿、电子表格等成品，目前平台已集成80余种专属智能体，覆盖视频剪辑、图像生成、网站搭建乃至餐厅预订等场景。这种“对话即生产”的模式，让知识工作者能聚焦最终产出而非操作过程。

微软在Ignite大会上发布的Agent 365系统，为智能体管理提供了关键基础设施。该系统通过Defender、Entra、Purview等安全组件构建防护体系，使企业能够像管理员工一样管控智能体的权限与行为。微软身份产品营销总经理伊丽娜·内查耶娃（Irina Nechaeva）强调：“智能体创新必须与管控工具同步推进，这对企业安全治理至关重要。”

对于Mainfunc这家成立仅一年、团队30人的初创企业而言，此次合作具有里程碑意义。景埃里克透露，公司已实现5000万美元年度经常性收入，在日本、美国、韩国市场表现突出，其客户留存率高达88%-92%。这位微软搜索引擎Bing的初创团队成员特别指出，Azure云平台为其整合OpenAI的GPT、Anthropic的Claude等8大模型提供了稳定支撑，这种多模型架构正是其“秘密武器”。

尽管AI能完成80%-90%的预备工作，景埃里克仍强调“最后一公里”必须由人类把控：“最终5%-10%的校验环节仍需人工完成，这是确保质量的关键。”这种务实态度也体现在产品设计中——Genspark生成的幻灯片可完美适配PowerPoint，让企业用户无需改变既有工作流程。

随着全球企业加速部署智能体（据微软预测，2028年企业智能体数量将超13亿），景埃里克预见未来两年AI应用主力将从开发者社群转向知识工作者。他认为当前AI革命仅渗透社会需求的“不足1%”，这片蓝海市场正孕育着无限可能。

中文翻译：

景昊的使命是推动人们从单纯与聊天机器人对话，转向将人工智能作为日常助手来完成工作，这一切都可通过单一平台实现。

这个平台正是Genspark超级智能体——由他与朱继盛在美国帕洛阿尔托联合创立的AI初创公司Mainfunc于2024年推出，目前在新加坡和日本设有分支机构。该平台致力于让用户“直接与工作成果对话，而非与聊天程序交流”，景昊在对比通用大语言模型时强调：“这将带来截然不同的用户体验。”

在Genspark平台上，用户使用自然语言向数字智能体发出指令，就能直接生成演示文稿或电子表格。“如果需要幻灯片，你就直接对幻灯片说话；想要电子表格，就直接与电子表格沟通。”景昊这样描述其创新交互模式。目前平台已部署80余款自研智能体，服务范围涵盖视频剪辑、图像生成、网站搭建、营销物料制作乃至餐厅预订。

如今，通过微软在Ignite大会上最新发布的Agent 365平台，企业可安全接入Genspark及Cognition、Glean、Kasisto、n8n等厂商的智能体。景昊表示：“我们希望让Genspark成为微软生态系统的一部分，知识工作者不仅能在Genspark平台完成任务，更能直接在Microsoft 365平台内实现无缝操作。”

据微软《2025年工作趋势指数报告》显示，全球企业已部署数以千计的智能体来自动化日常任务。微软赞助的IDC调研报告更预测，到2028年这一数字将突破13亿。微软身份产品营销总经理伊丽娜·涅恰耶娃指出：“对智能体的创建者、权限范围及行为安全进行全程管控至关重要，未受监管的智能体可能成为数据泄露或越权操作的黑箱风险。”

Agent 365作为智能体控制中枢，将企业现有人员管理体系延伸至智能体管理。它使智能体在获得定制化应用权限的同时享有同等安全防护，大幅降低业务流程整合成本。该平台集成了微软 Defender、Entra和Purview等安全生产力解决方案，IT部门可通过Microsoft 365管理中心调度智能体，使其不仅在微软应用生态中，还能在其他工作流里与人类管理者协同作业。

以时间紧迫的市场总监需要制作董事会多媒体方案为例：他们可在Agent 365中调用Genspark超级智能体，凭借其内置的Microsoft Entra智能体身份认证，安全连接所需资源。随后通过自然语言指令要求智能体完成课题研究，并生成演示文稿、文档、网站、视频及宣传海报。这些材料还能根据邮件、聊天或Teams会议中的讨论实时优化。项目结束时，Agent 365会核验智能体操作日志是否符合企业规范。

景昊对通过Agent 365将Genspark的“独特能力”融入微软生态充满信心，认为这将为用户创造“魔法时刻”。他透露，Genspark的幻灯片制作工具本就为兼容微软PowerPoint量身打造：“我们历经四轮文件导出功能迭代，确保生成的幻灯片能完美转换为PowerPoint格式，让习惯原有工作流程的用户无缝衔接。”

尽管智能体表现卓越，景昊仍强调人类在工作流中的核心地位：“我们认为AI可完成80%-90%的初步工作，但最后5%-10%的收官环节仍需人工审核把关。”对于成立仅一年的Mainfunc及其今年四月刚推出的多智能体平台Genspark而言，与微软的合作具有里程碑意义。

这支仅30人的团队专注为知识工作者赋能，现已实现5000万美元年度经常性收入，跻身全球增长最快的AI初创公司之列。景昊表示微软提供了“触达企业客户的绝佳渠道”，这正是初创公司难以独立覆盖的领域，同时盛赞其“构建了优质的软件生态系统”。

Genspark智能体融合了OpenAI的GPT、Anthropic的Claude等八大语言模型，通过微软Azure云平台为用户提供场景化最优解决方案。曾参与微软必应搜索引擎创始团队的景昊坦言：“这正是我们的核心竞争力。”基于对用户需求的专注及与微软的历史渊源，Mainfunc选择Azure作为Genspark的底层支撑：“我们信赖Azure可扩展的云解决方案，其统一接口还能协调多家人工智能模型服务。”

从衡量AI初创公司生存发展的关键指标来看，Mainfunc发展稳健：除年度经常性收入外，Genspark上线首月付费用户留存率达88%-92%，截至2025年6月30日的第二季度月均现金消耗不足百万美元。日本、美国和韩国是其核心订阅市场。

景昊预见行业将迎来爆发式增长：“当前人工智能革命仅渗透社会不足1%的领域，AI输出质量仍有巨大提升空间。两年内，整个市场重心将从开发者社区全面转向知识工作者群体。”

英文来源：

Eric Jing is on a mission to shift people from just talking with chatbots to using AI as an everyday companion to get work done, all from one platform.
That platform is Genspark Super Agent from Mainfunc, the AI startup he co-founded with Kay Zhu in Palo Alto, California in 2024, and has offices in Singapore and Japan.
“We want our users to talk directly to the business outcome, instead of talking to a chatbot. So that’s quite a different user experience,” Jing said comparing against LLMs, generative AI’s large language models commonly used to write, code or problem-solve.
On Genspark, users prompt the digital agent producing slides or spreadsheets directly, in natural language.
“If the outcome is a slide, you are talking to the slide directly. If the outcome you want is a spreadsheet, you talk to the spreadsheet,” said Jing.
There are more than 80 such homegrown agents on Genspark, delivering services ranging from edited videos and generated images to whole websites, marketing materials and even restaurant reservations.
And now, agents from Genspark and other providers, including Cognition, Glean, Kasisto, n8n and more, will be available for organizations to discover securely with Microsoft’s Agent 365, introduced today at Ignite.
“We want to make Genspark … a part of the Microsoft ecosystem,” said Jing of the move to be a partner in Agent 365. “So a knowledge worker can get tasks done not just from the Genspark site, but also within the Microsoft 365 platform without leaving it.”
Firms worldwide are already deploying thousands of agents to automate routine tasks, according to Microsoft’s 2025 Work Trend Index report. In an IDC Info Snapshot, sponsored by Microsoft, analyst firm IDC predicts that number will rise to more than 1.3 billion by 2028.
Keeping track of these agents — who created them, what they can access and if they’re behaving safely — is crucial. Unmanaged agents can turn into invisible threats that leak data or take unauthorised action, undermining a company’s security, governance and compliance framework.
Agent 365 is the control plane for agents that extends the existing infrastructure firms use for managing people, to agents. It equips agents with access to the same apps and protections, tailored to the agents’ needs, reducing time and effort spent integrating them into business processes.
Agent 365 includes Microsoft’s security and productivity solutions: Defender, Entra and Purview to protect and govern agents. Meanwhile, IT departments can manage agents with Microsoft 365 admin center and allow these agents to collaborate with their human managers not just on Microsoft 365 apps, but other workflows too.
Having these controls in place cannot be an afterthought, said Irina Nechaeva, general manager of identity product marketing at Microsoft. “All of this needs to be happening together with the agentic innovation itself. So not just the agents, but also the right tools to control, monitor, protect and make those agents more effective.”
So what will this process look like for say, a time-strapped marketing executive who needs to produce a multimedia proposal for the board?
They could go to Microsoft Agent 365, pick Genspark’s Super Agent and have that agent added to the team. And because that agent has a built-in Microsoft Entra Agent ID, it can securely connect to resources it needs to complete the job. They would then prompt that agent to research the topic, produce a presentation deck, document, website, video and posters. And the agent can refine or update these materials based on discussions via email, chat or Teams meetings.
At the end of the project, Agent 365 can compare the agent’s logged activities, to see if they align with the firm’s guidelines and policies.
Jing is confident that integrating Genspark’s “unique capabilities” into the Microsoft ecosystem via Agent 365 will create “magic moments” for the user.
Genspark’s AI tool for creating slides, for example, was already designed with Microsoft’s PowerPoint in mind.
“We wrote four iterations of the file exporting feature so you can export a Genspark slide into a perfect PowerPoint file … people already familiar with PowerPoint or with this kind of workflow in their company — they can still stick to PowerPoint,” he explained.
He stressed that the human element in this workflow is still paramount, regardless how well Genspark’s agents perform.
“In our world, we believe that 80% to 90% of the preliminary work can be done by AI. But the last mile – 5% to 10% of work – should still be done by humans manually. You need to verify everything,” he said.
Partnering with Microsoft is a big step for one-year old startup Mainfunc, and multi-agent Genspark, which was just launched in April 2025.
With a team of just 30 people, Jing says Mainfunc wants to focus on doing one thing really well — helping knowledge workers accomplish more.
Jing said Microsoft offers a “great channel to enterprise,” a user segment that Mainfunc has limited ability to service. “And it’s a great marketplace for different software,” he added.
So far, pouring all its effort into deciding which AI model can best solve pain points or tasks for knowledge workers at lowest cost has earned Mainfunc an annual recurring revenue (ARR) of $50 million, and a ranking as one of the world’s fastest-growing AI startups.
Genspark’s agents are powered by a mix of eight LLMs— comprising household names like OpenAI’s GPT and Anthropic’s Claude—giving users a best-for-purpose service curated by Genspark and underpinned by Microsoft Azure.
“This is definitely our secret sauce,” said Jing, an industry veteran who was a founding team member for Microsoft’s search engine Bing.
This laser focus on user solutions, and Jing’s history with Microsoft, also led Mainfunc to adopt Microsoft Azure as a reliable and secure infrastructure for Genspark.
“Basically, I want to … solve the pain point from our users, not solve the pain point of different services,” Jing said. “We trust Azure to host scalable cloud solutions.”
Azure also enables Genspark to interface with multiple AI models, including from OpenAI and Anthropic, via a single platform.
Going by AI startups’ “true north” metrics — measures of survival and growth — cited by Jing, Mainfunc seems on track. In addition to ARR, which measures predictable income, Mainfunc had a paid retention rate of 88% to 92% in Genspark’s first month, indicating customer loyalty. Meanwhile its monthly burn rate, or how quickly it’s spending cash, averaged less than $1million for the second quarter ended June 30, 2025.
Jing said its top subscriber markets are Japan, the U.S. and South Korea.
He sees huge growth for the sector, because he believes the AI revolution has penetrated “less than 1% of society” and there is still much improvement to come in the quality of work that AI can deliver.
“Two years from now, I see the entire market will shift more and more from developer communities to the knowledge workers,” he said.