百度提出产业智能体操作系统,百度搭子开放企业版Baidu proposes industrial intelligent agent operating system, Baidu Dazi opens enterprise edition


AI free token/AI免费积分
0

9月16日,2026智能经济论坛上,百度集团执行副总裁、百度智能云事业群总裁沈抖介绍了"产业智能体操作平台"理念。

沈抖表示,产业智能体操作平台能把智能体、企业经营体系与云资源组织在一起,让智能体真正跑在企业业务里。现场,沈抖援引一组比较:过去一年全球AI资本开支约5000亿美元,仅占全球GDP的0.4%;同周期内电力消费与电力平台投资已接近GDP的5%。他指出,AI尚未平台进入真正的基础设施周期,产业智能体有机会把模型能力的进步转化为企业经营成果,让大模型从一项工艺变成经济力量。

沈抖将企业任务拆为三类:20%为日常运营、30%为专业领域、50%为真正决定企业经营的核心任务,将来都将由对应的产业智能体内化。据此,他提出三类智能体:通用、专业与定制,并特别强调,通用智能体不能简单理解为"办公智能体",将来办公能力将变为随时可调用的基础能力,让员工将注意力放回专业与核心情形。

专业智能体层面,百度智能云已布局伐谋、百度胜算、秒哒、文心快码、一见、Hogee、一镜等矩阵。泰康在与百度智能云合作中将智能体应用于医疗健康、风控合规、销售运营等情形,今年4月部署智能体开发平台后已上线3500多个智能体,日均消耗Tokens达200亿。

当日发布的百度搭子开发平台,将百度搭子同款工艺开放给企业,覆盖智能体开发、运行与管理的全生命周期API,企业可在现有平台中集成并定制产业智能体。百度智能云同步宣布,百度搭子将构建为完全开放的生态平台,伙伴技能、套件、应用可便捷接入,企业既能使用第三方能力,也能将自身能力放到搭子上服务更多客户。百度智能云维度表示,将围绕产业智能体落地建立开放生态,且不会向生态伙伴抽佣。

基础设施层面,沈抖介绍,Agent Infra已接入超过150个顶尖模型,过去6个月借助推理引擎深度优化,智能体长链路推理表现提高3倍以上;Agent Cloud兼容万级沙盒百毫秒级热启动,覆盖智能体身份权限统一管理与执行链路全程可追溯。算力维度,昆仑芯已在互联网、能源、金融、制造、汽车、运营商、教育等行业建设全国产集群;天池超节点已达成全栈国产化。沈抖预计,到2030年将有30亿个产业智能体持续运行,带来超过100吉瓦级推理需要。(袁宁)

 

本文来源:https://www.163.com/tech/article/L6VC979T00098IEO.html
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

0

On September 16, at the 2026 Intelligent Economy Forum, Shen Dou, Executive Vice President of Baidu Group and President of Baidu Intelligent Cloud Business Group, introduced the concept of the "Industrial Agent Operating Platform."

Shen Dou said that the Industrial Agent Operating Platform can organize agents, enterprise operating systems, and cloud resources together, enabling agents to truly run within enterprise business. At the event, Shen Dou cited a comparison: over the past year, global AI capital expenditure was about $500 billion, accounting for only 0.4% of global GDP; in the same period, electricity consumption and investment in electricity platforms had approached 5% of GDP. He noted that AI has not yet entered the true infrastructure cycle as a platform, and industrial agents have the opportunity to turn advances in model capabilities into enterprise operating results, transforming large models from a craft into an economic force.

Shen Dou divided enterprise tasks into three categories: 20% daily operations, 30% professional fields, and 50% core tasks that truly determine business operations. In the future, all of these will be internalized by corresponding industrial agents. Based on this, he proposed three types of agents: general, professional, and customized. He particularly emphasized that general agents should not be simply understood as "office agents." In the future, office capabilities will become basic capabilities that can be invoked at any time, allowing employees to refocus on professional and core scenarios.

At the professional agent level, Baidu Intelligent Cloud has deployed a matrix including Famou, Baidu Shengsuan, Miaoda, Wenxin Kuaima, Yijian, Hogee, and Yijing. In cooperation with Baidu Intelligent Cloud, Taikang has applied agents to scenarios such as healthcare, risk control and compliance, and sales operations. After deploying the agent development platform in April this year, it has launched more than 3,500 agents, with daily token consumption reaching 20 billion.

The Baidu Dazi development platform released that day opens Baidu Dazi's same technology to enterprises, covering full-lifecycle APIs for agent development, operation, and management. Enterprises can integrate and customize industrial agents on their existing platforms. Baidu Intelligent Cloud also announced that Baidu Dazi will be built into a fully open ecosystem platform, where partner skills, suites, and applications can be conveniently integrated. Enterprises can not only use third-party capabilities but also put their own capabilities on Dazi to serve more customers. Baidu Intelligent Cloud said it will build an open ecosystem around the deployment of industrial agents and will not take commissions from ecosystem partners.

At the infrastructure level, Shen Dou introduced that Agent Infra has integrated more than 150 top-tier models. Over the past six months, with deep optimization of the reasoning engine, agents' long-chain reasoning performance has improved by more than 3 times. Agent Cloud supports 10,000-level sandboxes with 100-millisecond-level hot starts, covering unified identity and permission management for agents and full traceability of execution chains. In terms of computing power, Kunlun Chip has built fully domestic clusters in industries such as internet, energy, finance, manufacturing, automotive, telecom operators, and education; Tianchi supernodes have achieved full-stack domestic localization. Shen Dou predicts that by 2030, 3 billion industrial agents will be continuously running, generating more than 100 gigawatts of inference demand. (Yuan Ning)

Source: NetEase Smart Editor: Han Lei_NBJS32996

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry