AI Wiki · 16 条来源

AI Agents

AI agents are model-driven systems that plan, call tools, use memory, and iterate across multi-step tasks。

一句话定义： AI agents are model-driven systems that plan, call tools, use memory, and iterate across multi-step tasks.

页面状态

  • 状态：source-backed
  • 来源数量：16
  • 更新方式：由 source-crawler 资料池生成，人工/LLM 综合写入 当前综合。

这是什么

AI Agents 是 AI Wiki 中的一个长期知识节点。它不是一次性新闻，而是持续汇集官方博客、工程实践、newsletter、benchmark 和人物观点的主题页。

当前综合

  • Agent pages should connect protocols, memory, evals, and product workflow evidence instead of treating each tool announcement as a separate trend.
  • The durable question is not whether agents can act, but which harness, permission, trace, and review patterns make their actions dependable.

为什么值得关注

这个主题同时出现在 16 条资料中，说明它已经跨越单篇文章，成为一个需要持续跟踪的知识簇。关键词包括：AI agents、agentic systems、tool use、agent memory、agent workflow。

近期信号

  • Shared Memory vs Isolated Memory Architecture for AI Agents · Wolbarg：来自 wolbarg.com link target，约 13437 字符。
  • AI Agents Searching Online Can Struggle to Retrieve Correct Info, Resea…：来自 The Batch，约 5412 字符。
  • Evaluating AI Agents: A production blueprint with Strands and AgentCore…：来自 AWS Machine Learning Blog，约 25973 字符。
  • OpenAI introduces Presence to help enterprises build AI agents - Silico…：来自 siliconangle.com link target，约 5703 字符。
  • How AI Agents Leak Data: The Exfiltration Channel：来自 particula.tech link target，约 17303 字符。
  • Most “self-improving” AI agents don’t actually improve：来自 loadbearingtech.substack.com link target，约 11021 字符。

关键问题

  • What makes a system an agent? An agent combines model reasoning with tools, state, goals, and feedback loops rather than returning one isolated answer.
  • Why track agents separately? Agent systems expose reliability, evaluation, security, and product workflow issues that simple chat interfaces do not show.

待追踪问题

  • Which agent workflows show repeatable production value beyond demos?
  • What trace and review evidence should be required before calling an agent reliable?

来源覆盖

当前页面引用了 16 条资料，主要来自：siliconangle.com link target 3 篇、AWS Machine Learning Blog 2 篇、AlphaSignal 2 篇、particula.tech link target 2 篇、The Batch 1 篇、blog.jetbrains.com link target 1 篇、coreweave.com link target 1 篇、loadbearingtech.substack.com link target 1 篇。

证据类型

类型 数量 阅读建议
一手来源 2 官方或研究机构来源，适合支撑模型发布、方法、产品和政策相关事实。
发现信号 3 适合发现新主题和补充背景，重要事实应回到一手来源核对。
背景资料 11 可作为补充上下文，阅读时需要留意发布时间和来源权威性。

来源列表

一手来源