内容来源：https://aibusiness.com/agentic-ai/aws-s-new-agentic-tools-trail-rivals-respond-real-problems

内容总结：

AWS纽约AI峰会：云巨头加速布局智能体AI，聚焦安全与企业落地

在周三于纽约举行的AWS AI峰会上，亚马逊云科技密集发布了一系列新功能，涵盖其AI智能体构建平台Amazon Bedrock AgentCore、AI助手Amazon Quick、自主智能体以及Adobe、Shopify等第三方应用集成。尽管这些能力在市场上并非首创，但AWS试图通过“组合拳”式的大规模更新，回应企业在构建、部署和保障智能体AI安全性方面的核心关切。

从“前沿模型”转向“上下文工程”

当前AI行业趋势已从追逐最新模型转向解决实际落地难题。AWS新推出的Amazon Bedrock托管知识库和AWS Context服务，旨在帮助AI智能体精准获取所需知识背景，减少“幻觉”现象。Omdia分析师Torsten Volk指出：“成功与失败的AI项目之间，核心区别在于智能体能否说出‘我不知道答案’，而不是凭空编造。”

安全成为智能体时代新焦点

为增强企业信任，AWS预览了全新安全工具AWS Continuum。该工具可针对编码智能体中的漏洞，主动识别风险、生成优先级修复列表并自动执行补丁更新，支持基于多个前沿模型的智能决策。

追赶中寻求差异化

尽管AWS在部分能力上落后于谷歌、微软及OpenAI等竞争对手，但分析师Lian Jye Su认为，此次更新表明AWS正在“认真倾听客户需求”。他同时指出，AWS还需在连接器生态、硬件基础及智能体运维（AgentOps）能力上持续构建差异化优势，才能在智能体AI时代真正脱颖而出。

中文翻译：

由 Google Cloud 赞助
选择您的首个生成式AI应用场景
要开始使用生成式AI，首先应聚焦于能够改善人类与信息交互体验的领域。

凭借其最新的AI工具与产品，该供应商表明，尽管未能为市场带来新意，但仍在倾听客户需求。

AWS正试图重振其智能体AI产品线，同时在Amazon Quick、Bedrock、AgentCore以及全新安全产品中发布新功能。

尽管其AI产品与服务可能被视为落后于谷歌、微软等竞争对手以及OpenAI、Anthropic等AI实验室，但这反映出AI供应商不再仅专注于构建新工具，而是着力解决企业在构建、部署及保障智能体AI安全方面的关切。

在周三于纽约举行的AWS AI峰会上，AWS在多个平台推出新功能，包括用于构建、连接和优化AI智能体的平台Amazon Bedrock AgentCore；AI驱动助手Amazon Quick；自主智能体；以及与Adobe、Shopify、Smartsheet和Snowflake等应用的新集成。这家云巨头还预览了一项新服务，该服务可帮助智能体将现有数据间的关系映射到知识图谱中。除推出新的智能体能力外，AWS还聚焦于AI智能体的安全保障。

“他们正在增强自身（产品），为AI智能体的普及做好准备。”Omdia（Informa TechTarget旗下机构）分析师Lian Jye Su表示。

然而，其中许多发布内容在市场上并非新事物。

“其他超大规模云服务商和前沿模型供应商已提供这些能力，但只有像AWS这样的超大规模云商才能同时推出如此多的增强功能。”Su说道。

例如，新的Amazon Bedrock托管知识库提供了类似Google Cloud Vertex AI Search与Vertex AI Grounding的功能。知识库可连接SharePoint、Google Drive和Confluence等应用中的非结构化数据源，并具备智能体检索器，可规划查询、关联文档间概念并对结果重新排序。尽管知识库在运用检索增强生成方面与Vertex AI Search相似，但AWS提供的第三方AI模型范围比谷歌更广。

另一项新功能是AgentCore上的网络搜索，它能在保持数据处于AWS安全边界内的同时提供实时网络信息。AWS还预览了AgentCore支付功能，使智能体能够发现、访问并支付优质内容与服务。

除知识库外，AWS还推出了AWS Context。这项新服务将现有数据间的关系映射到知识图谱中，使AI智能体能够访问受治理的数据关系、业务规则及领域知识。

“只需在管理控制台中点击几下，您的智能体现在就能获取上下文信息。”AWS智能体AI副总裁Swami Sivasubramanian在周三的主题演讲中表示。

知识库与AWS Context表明，AI趋势已从持续创新的需求转向关注开发者面临的挑战以及阻碍客户有效推进AI项目的障碍。其中一些挑战在于确保AI智能体能够获取正确的知识与上下文，从而避免在执行任务时产生幻觉。

“重点已不再是前沿模型。”Omdia分析师Torsten Volk表示，“关键在于上下文工程。失败AI项目与成功落地的AI项目之间的重大区别在于，智能体能否说‘不，抱歉。我不知道这个问题的答案。’而不是凭空编造。”

他补充道，许多企业成功应用AI的另一障碍是对技术的信任。

AWS通过AWS Continuum解决这一问题。Continuum是一款新型安全工具，可应对编码智能体中的漏洞。AWS表示，Continuum通过推理、行动、遥测和上下文等元素，提供主动且以结果为导向的安全防护。该工具可吸收现有漏洞积压，生成带有证据的优先级列表，识别误报，评估现有防御与控制措施，并通过策略更新或代码补丁提供修正方案。该工具根据应用场景使用多个前沿模型。

Su表示，尽管AWS在某些领域落后于竞争对手，但这些新发布应能帮助希望构建智能体的企业，并表明供应商正在倾听客户需求。

“这表明AWS收到了大量关于支持智能体AI能力的需求与请求。”他说。但他补充道，要在智能体AI时代取得成功，供应商仅靠自身一系列类似能力还不够。

“他们需要更好地实现差异化。”Su表示，“AWS需通过添加连接器以便捷访问、集成功能作为技能、准备合适的硬件基础，并引入更多涵盖FinOps、可观测性及治理的AgentOps能力，持续让云平台为智能体做好准备。”

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.
With its latest AI tools and products, the vendor showed it is listening to its customer base even while it fails to offer novelty in the market.
AWS is trying to revive its lineup of agentic AI offerings while releasing new capabilities across Amazon Quick, Bedrock, AgentCore, and new security products.
While its AI products and services may be perceived as behind competitors such as Google and Microsoft and AI labs such as OpenAI and Anthropic, they illustrate a trend in which AI vendors are no longer focusing solely on building new tools but on addressing enterprises' concerns about building, deploying and securing agentic AI.
At its AWS AI Summit New York City conference on Wednesday, AWS introduced new capabilities across platforms such as Amazon Bedrock AgentCore, its platform for building, connecting and optimizing AI agents; Amazon Quick, its AI-powered assistant; Autonomous Agents; and new integrations with applications such as Adobe, Shopify, Smartsheet and Snowflake. The cloud giant also previewed a new service that helps agents map relationships across existing data into a knowledge graph. Beyond introducing new agentic capabilities, AWS focused on securing AI agents.
“They are enhancing their [products] to be ready for the proliferation of AI agents,” said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget.
However, many of the releases are not new to the market.
“These capabilities are already available from other hyperscalers and frontier model vendors, but only a hyperscaler like AWS can launch so many enhancements at the same time,” Su said.
For example, the new Amazon Bedrock Managed Knowledge Base offers features similar to Google Cloud’s Vertex AI Search and Vertex AI Grounding capabilities. Knowledge Base connects unstructured data sources across applications such as SharePoint, Google Drive and Confluence. It features an agentic retriever that plans queries, connects concepts across documents and reranks results. While Knowledge Base is like Vertex AI Search in its use of retrieval augmented generation, AWS offers a wider range of third-party AI models than Google.
Another new feature is Web Search on AgentCore, which provides real time web information while keeping data within AWS security boundaries. AWS also previewed AgentCore Payments, which enables agents to discover, access and pay for premium content and services.
Along with Knowledge Base, AWS introduced AWS Context. The new service maps relationships across existing data into a knowledge graph, allowing AI agents to access governed data relationships, business rules and domain knowledge.
“With just a few clicks in the managed console, your agents now have context,” said vice president of a Swami Sivasubramanian, VP, AWS agentic AI, during a keynote presentation on Wednesday.
Knowledge Base and AWS Context show that AI trends have has shifted from the need to innovate constantly to the need to focus on developers’ challenges and the obstacles that keep customers from effectively pursuing AI initiatives. Some of those challenges are about ensuring AI agents have access to the right knowledge and context, so they don't hallucinate when asked to perform a task.
“It’s not about frontier models anymore,” said Torsten Volk, an Omdia analyst. “It’s about context engineering. The big difference between a failed AI project and an AI project that gets across the finish line is that the agent can say, ‘No, sorry. I don’t know the response to this question,’ instead of just making something up.’”
He added that another barrier for many enterprises to succeed with AI is trust in the technology, he added.
AWS addresses this issue with AWS Continuum. Continuum is a new security tool that addresses vulnerabilities in coding agents. AWS said Continuum provides active, outcome-driven security with elements such as reasoning, actions, telemetry and context. The tool can ingest the existing vulnerability backlog, generate evidence-backed priority lists, identify false positives, assess existing defenses and controls, and provide corrective measure using policy updates or code patches. The tool uses multiple frontier models depending on the application.
While AWS is trailing competitors in some of these areas, the new releases ought to be helpful for enterprises looking to build agents and show the vendor is listening to its customer base, Su said.
“It shows AWS is getting a lot of demands and requests for supporting agentic AI capabilities,” he said. However, the vendor needs more than just its own set of similar capabilities to be successful in the agentic AI era, he added.
“They will need to differentiate themselves better,” Su said. “AWS needs to continuously make its cloud agent-ready by adding connectors for ease of access, integrating functions as skills, preparing the right hardware foundation … and introducing more AgentOps capabilities in terms of FinOps, observability and governance.”