[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-may-2026-5-agent-tech-self-evolve":3,"topics-all":36,"news-related-f70c514b-bc4e-4930-bdc2-1db3bcc65b2f":55},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"f70c514b-bc4e-4930-bdc2-1db3bcc65b2f","本周AI Agent五项关键技术：从自进化到系统安全，生产级落地加速","2026年5月19日至23日，AI Agent领域密集发布五项关键生产技术，从自我进化到系统安全形成完整链条。\n\nMOSS论文提出自进化框架：Agent识别自身逻辑弱点，直接重写源代码模块，通过自动化测试验证后部署更新。这不是Prompt调优，是真正的自主代码修复。Agent可以主动分析源码文件，找到失败模式，重新编写对应模块，通过测试套件验证，部署改进后的版本。配套的Ratchet方案提供非分歧分析，确保修改不导致基准分数下降。对生产环境，这意味着从人工排错→更新→部署的数天周期压缩为分钟级闭环——编码Agent遇到TypeScript重构失败模式，可以自主修补工具调用逻辑，无需等待开发者介入。\n\nGoogle在I\u002FO 2026发布的Managed Agents将Agent编排从客户端迁移至服务端。开发者定义工具、指令和触发器，Google在API层面维护Agent循环、持久化状态、处理调度。无需自建服务器、无需维持WebSocket连接。Agent可以在无活跃客户端的情况下全天候主动执行任务。Hosted Agent基础设施正在成为大厂标配战场，但对中小团队也带来锁入风险。\n\nCompiling Agentic Workflows into LLM Weights论文证明，多步Agent流水线可以蒸馏为单一模型，成本降低两个数量级，延迟从30秒压缩至2秒。这不是替代Agent，而是分层策略：复杂任务保留完整流水线，稳定模式交给编译后的轻量模型。\n\nIdleSpec利用工具调用空闲时间预生成候选动作，实测60-80%的情况下用户感知延迟为零。LCGuard通过潜在通信防护解决多Agent共享KV-Cache时的系统级风险——一个受损Agent不再能污染整个系统。\n\n五条技术路径共同指向一个结论：生产级AI Agent正在从能跑走向跑得好。基础设施抽象层级在提高，部署门槛在下降，但系统复杂度和安全边界也在同步扩展。技术爆发之后，真正的考验是工程化能力能否跟上。","https:\u002F\u002Fwww.requesty.ai\u002Fblog\u002Fai-agent-techniques-may-2026-self-evolving-managed-compiled","2f79a578-93e6-4aff-9d52-42ba2239b02d",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":18,"name":19,"slug":19,"description":13,"color":13},"e82b2d09-81b2-43d1-977e-e018443b3c14","coding-agent",{"id":21,"name":22,"slug":22,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"1d7b3f61-cff5-4968-ae25-7819027fd7c1","en","Five agent breakthroughs this week: self-evolution to security","During the week of May 19-23, 2026, the AI agent field saw five key production technologies released in tight succession, forming a complete chain from self-evolution to system safety.\n\nThe MOSS paper proposes a self-evolution framework: agents identify their own logical weaknesses, directly rewrite source-code modules, validate via automated tests, and deploy the updates. This isn't prompt tuning — it's genuine autonomous code repair. The agent can actively analyze source files, find failure patterns, rewrite the corresponding module, verify via test suites, and deploy the improved version. The companion Ratchet scheme provides non-divergence analysis, ensuring modifications don't degrade benchmark scores. For production environments, this compresses the cycle of manual debugging → update → deploy — from days to a minutes-level closed loop. A coding agent hitting a TypeScript refactoring failure can patch its tool-calling logic on its own, no developer intervention needed.\n\nGoogle's Managed Agents, released at I\u002FO 2026, moves agent orchestration from client-side to server-side. Developers define tools, instructions, and triggers; Google maintains the agent loop, persistent state, and scheduling at the API layer. No self-hosted servers, no WebSocket maintenance. Agents can run actively around the clock without an active client. Hosted agent infrastructure is becoming a standard battleground for the majors, but it also brings lock-in risk for small teams.\n\nThe paper Compiling Agentic Workflows into LLM Weights shows that multi-step agent pipelines can be distilled into a single model — two orders of magnitude cost reduction, latency compressed from 30 seconds to 2 seconds. This isn't replacing agents, but a layered strategy: complex tasks keep the full pipeline, stable patterns go to the compiled lightweight model.\n\nIdleSpec exploits the idle time of tool calls to pre-generate candidate actions — in 60-80% of cases user-perceived latency is zero. LCGuard uses latent communication protection to address the system-level risk of multi-agent shared KV-Cache — a compromised agent can no longer poison the whole system.\n\nThese five technical paths point to a single conclusion: production-grade AI agents are evolving from \"can run\" to \"runs well.\" The infrastructure abstraction layer is rising, deployment barriers are falling, but system complexity and security boundaries are expanding in lockstep. After the tech explosion, the real test is whether engineering capability can keep up.","may-2026-5-agent-tech-self-evolve","2026-05-24T13:10:00Z","2026-05-24T13:07:35.453926Z","2026-08-19T02:08:40.142862Z",true,"agent",157,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"c696208b-6535-4eb9-b1ed-2e4f835d2f88","NVIDIA SoL-Pi 把 coding agent 的 token 砍掉 44%,harness 开始变天","nvidia-sol-pi-harness-token-compression","2026-09-19T03:00:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"4860cd5e-0e85-458a-bb63-cf87e137332b","MCP 2026-07-28 Release Candidate：协议层正式无状态化，Agent 后端跑上普通 HTTP 基础设施","mcp-2026-07-28-stateless","2026-07-01T08:20:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"9b47c85f-7b67-4e33-be69-b98fc50af87a","Prime Intellect 押注「递归语言模型」RLM：让 LLM 主动管理自己的上下文","prime-intellect-rlm-recursive-language-model","2026-06-19T14:30:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"ad3e5dbd-2c30-43a1-bf67-a6ccd16fa11e","Databricks 开源 Omnigent：Matei Zaharia 想给 Coding Agent 之上再加一层「元 Harness」","databricks-omnigent-meta-harness-coding","2026-06-13T08:00:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"54b3961a-17a4-4bcf-af68-e4c873eeb8ca","华为云 ModelArts Next：从训练平台走向「智能体原生」的训推底座","huawei-modelarts-next-agent-native-base","2026-06-05T07:00:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"a3e9cf60-e2dc-4ac6-a1a6-1089ee721bc9","Google Antigravity CLI 全面开放：子Agent并行编排进入终端开发时代","google-antigravity-cli-sub-agent-parallel","2026-05-31T10:05:00+00:00"]