Open Source Maintainers Need a Spam Filter for AI Labor

AI is fundamentally changing how junior developers learn, but it’s also making senior developers a whole lot less valuable if all they’re…

AI Agents Will Take Over Mobile Terminals in Next Decade, Honor's Product Line President Says

一财全球Yicai Global

Using mirrord to verify AI-SRE fixes against the staging cluster

Part II of two. See Part I for the recipe. In Part I we’ve discussed how you can plug mirrord into your AI-SRE so it can autonomously test its fix in the real cluster. In this post, we’ll show this running end-to-end on a real Kubernetes cluster against HolmesGPT , the only OSS AI-SRE we found that is self-hostable. We planted two bugs to test both possible verdicts: PASS when the patched run clears the alert’s SLO without any regressions, REJECT when it doesn’t (the patched run still violates the alert’s SLO, or cleared the SLO but added a regression). Flow: HolmesGPT investigates an alert and returns a markdown report. A small Claude wrapper turns the report into a code patch. The verifier runs that patch under mirrord exec , compares against an unpatched baseline, and returns a verdict tied to the alert’s actual

SLO. What’s HolmesGPT? A prebuilt LLM-backed agent tailored for on-call work: it can run kubectl commands, fetch logs from the cluster, and read the runbooks you’ve already written (in Notion, Confluence, or markdown files) to ground its investigation. A Python service called checkout handles HTTP requests; on each request it calls a pricing service for the item price. A loadgen pod sends requests to checkout continuously to keep the system warm. Prometheus scrapes checkout ’s metrics and Alertmanager fires when SLOs are violated. Everything sits in one namespace. HolmesGPT runs as a pod in the cluster, subscribed to Alertmanager. When an alert fires, HolmesGPT investigates and hands its markdown report off to a separate verifier pod. The verifier runs the bridge (the Claude call that turns the report into a code patch), then uses mirrord exec to run the patched code inside the verifier pod with deploy/checkout ’s network identity, env, and mounts. When the patched copy calls pricing , it hits the real pricing pod. A recent change introduced a request format checkout() doesn’t handle: every request for an item_id ending in -3 raises ValueError and returns 500. Loadgen sends a mix; ~10% of requests fail. A Prometheus rule fires CheckoutErrorRateHigh once the error rate climbs over 5%. HolmesGPT Investigates This is the call HolmesGPT runs in-cluster when Alertmanager fires: holmes investigate alertmanager \ --alertmanager-url \ --alertmanager-alertname CheckoutErrorRateHigh \ --model 'anthropic/claude-sonnet-4-20250514' HolmesGPT investigated for ~30 seconds (pulled pod descriptions, fetched logs, walked the service config) and concluded: Root Cause: Application logic error causing 500 responses for specific item (item-3) Error Details: Error rate: 20.09% (above 5% SLO). Specific error: ValueError: unsupported catalog shape for item_id=item-3 . Pattern: repeated failures for item-3 checkout requests returning HTTP 500. Remediation: Fix application code to handle item-3 catalog shape or add proper validation. Clean diagnosis: HolmesGPT pulled the exception out of the logs and attributed it correctly. Bridge to a code patch. HolmesGPT outputs a markdown report. The bridging step is a small Claude call running in the verifier pod (~80 lines around the Anthropic SDK) that takes the report plus the service’s source and returns a code patch. We asked it to faithfully implement HolmesGPT’s recommendation: handle the item-3 case (Claude chose to return zero) rather than raising. Claude produced the minimal edit to . Verify. Two runs (baseline and patched), 100 requests each. The verifier compares the alert’s SLO condition against the patched run, plus a regression watchlist on the other signals. The “baseline” here is the verifier’s own load test on the unpatched code, same load and same downstream as the patched run, not the live error rate HolmesGPT observed. That’s why the number below differs from the 20.09% in HolmesGPT’s report. SLO under verification: error_rate > 5% (alert fires while this holds) Regression watchlist : Verdict: PASS. Alert condition error_rate > 5% no longer satisfied (10% → 0%). Regression watchlist clean. The error-rate bug was easy: a literal exception in the logs with an obvious fix. What does the loop look like for a less log-visible bug? We swapped in the second planted defect: fetch_price() has no client-side timeout, and pricing has a long tail: 1 in 10 calls takes ~1.5s. p99 drifts to ~2 seconds, well above the

300ms SLO. CheckoutP99High fires. holmes investigate alertmanager \ --alertmanager-url \ --alertmanager-alertname CheckoutP99High \ --model 'anthropic/claude-sonnet-4-20250514' Root Cause Analysis: Latency caused by synchronous dependency calls (each checkout request triggers pricing API calls), no apparent caching, potential bottleneck in pricing service. Caching is a reasonable optimization. It’s also generally a good idea. The question is whether it resolves this specific alert . Bridge: Of the four recommendations, only caching is a single code change; the others are runtime tuning or architectural refactors. Claude implemented it as an in-memory TTL cache around fetch_price() . SLO under verification: p99 latency (ms) > 300 Regression watchlist: Verdict: REJECT. Alert would still be firing post-merge. SLO p99_ms > 300.0 (baseline 2162, patched 2010). p50 dropped 99.8% (cache hits return instantly), but the 99th-percentile request is still a cache miss hitting the 1.5s tail. The alert-aware classifier says the thing the on-call needs to hear: the alert that fired would still fire on this code; SLO threshold 300ms; patched value 2010ms. The fix HolmesGPT didn’t propose: a short client-side timeout on fetch_price() to fail-fast under the SLO instead of waiting out the 1.5s tail. We re-ran the bridge with a hand-crafted patch doing that; verifier returned PASS at patched p99 ~330ms. The point isn’t that HolmesGPT missed the right fix, it’s that the verifier surfaced the gap before the wrong patch shipped. The loop closes against a real OSS AI-SRE you can install yourself. HolmesGPT investigates, the bridge turns its report into a patch, mirrord runs it against the live cluster, the classifier returns a verdict tied to the actual alert. PASS and REJECT both have to be useful or the verifier becomes decorative. Scenario 1 signs off on a real fix. Scenario 2 stops one that looks good (p50 −99.8%) but doesn’t clear the alert. The sample checkout / pricing services, Prometheus rules, verifier code, and bridge script are small (~500 lines of Python plus the ~80-line bridge). The code lives at . You’ll need the mirrord operator installed in your cluster. What are the alternatives to HolmesGPT? HolmesGPT is the open-source, self-hostable AI-SRE used in this demo, but it’s one of several. Commercial options include Resolve AI, ’s Investigator, and Datadog’s Bits AI SRE, and many teams build their own in-house loop. The verification pattern in this post is tool-agnostic: whichever AI-SRE you run, you can wrap its suggested fix in the same mirrord exec verification loop before it reaches a human. What can an AI-SRE do, and what are its limits? An AI-SRE takes the first investigative pass when an alert fires: it gathers recent deploys, related logs, metric anomalies, runbooks, and similar past incidents, then proposes a remediation, sometimes auto-executing low-risk actions like a pod restart or config flag flip. What it generally can’t do is tell you whether its proposed fix actually works against real cluster state; most are single-shot and don’t verify. That’s the gap mirrord closes: running mirrord exec in the loop tests each candidate patch against the live cluster, so only fixes that clear the alert’s SLO reach a human. How is verifying a fix with mirrord different from using a staging environment? A staged copy runs as its own principal with its own environment and needs a build, push, and deploy cycle on every iteration. mirrord exec runs

the patched code with the target pod’s real network identity, environment, secrets, and mounts, hitting the same live downstreams, in seconds and without a deploy. Each run is its own process, so you can verify many candidate patches in parallel, which matters in a noisy incident where the AI proposes several fixes. If you haven’t read Part I yet, it has the recipe: the pattern, the verification code, and how to wire it into the AI-SRE you’re already running. Have questions? Reach out at [email protected] or on our community Slack

Helmholtz AI: Democratising AI for a data

AI consulting Leverage AI methods & expertise and power up your research.

Flow 升级为 AI 工作代理平台，发布自研 Repattern AI 引擎

在首尔举行的“AX Festa 2026”上，Madras Check 宣布将其协作工具 Flow 升级为 AI 工作代理平台，并发布自研引擎 Repattern AI。CEO Lee Hak-jun 指出，传统协作工具存在数据与价值、目标与执行、以人为中心三方面脱节，导致 SaaS 难以直接提升效率。Flow 通过接入过去10年约5000家客户的协作数据，推出四类 AI 代理（Mate、Smart、Consulting、Automation），支持将代理设为项目负责人，可自动处理 Bug、更新进度并反馈结果。此外，Flow 支持通过模型上下文协议（MCP）对接 ChatGPT、Claude 等外部 AI 工具，并开放 API 与开发者中心。内部案例显示，团队借助 Flow MCP 与 Claude 在3天内搭建出 CRM 仪表盘。新版本还整合 OKR 功能，打通目标、执行与复盘，并联动外部业绩数据。Lee Hak-jun 强调，核心在于将协作数据接入 AI，放大数据价值并加速目标到执行的连接。 #Flow #AI工作代理 #MadrasCheck #RepatternAI #协作工具 #SaaS #企业AI #OKR #MCP #科技新闻

Telegram必备的搜索引擎，极搜JISOU帮你精准找到，想要的群组、频道、视频、音乐

👉 t.me/jisou2?start=a_8247614025

小米发布AI代理自动演进框架HarnessX，无需调整模型平均提升14.5%

据VentureBeat报道，小米研究团队推出AI代理自动演进框架HarnessX，旨在不调整底层模型的前提下持续优化代理系统表现。该框架采用模块化架构，其核心AEGIS引擎通过分析运行日志、规划改进方向、生成代码修改方案及评审门控机制自动优化系统。在15组模型-基准测试中，14组实现性能提升，平均增幅达14.5%，其中Qwen3.5-9B在规划类测试中提升44%。HarnessX还支持“协同演进”机制，可将优化数据用于模型强化学习，额外带来平均4.7%的性能提升，但目前仅适用于开源模型。研究团队使用Claude Opus 4.6作为元代理，当前元代理仍需依赖较强前沿模型，这是该方案现阶段限制之一。 #小米 #HarnessX #AI代理 #自动演进 #大模型 #机器学习 #科技新闻

谷歌将Computer Use能力原生集成到Gemini 3.5 Flash

谷歌宣布，在其最新发布的大语言模型Gemini 3.5 Flash中，原生集成了“Computer Use”（计算机使用）能力。这意味着该模型不再需要额外插件或第三方工具，即可直接操控计算机界面，执行点击、输入、滚动等操作，甚至能够自主打开浏览器、操作桌面应用，完成复杂的自动化任务。这一集成将极大简化AI代理（Agent）的开发流程，使Gemini能够像人类一样与软件交互，从而显著提升在办公自动化、数据录入、网页测试等场景中的实用性。谷歌表示，此举旨在推动AI从“对话助手”向“行动助手”进化，赋予模型真正的“动手能力”。开发者社区对此反应积极，认为这标志着大模型在实际应用中的一次关键突破。目前该功能已面向部分开发者开放测试，未来将逐步扩大可用范围。 #谷歌 #Gemini #AI #大模型 #ComputerUse #智能代理 #自动化 #技术突破

人工智能代理引发问题，组织却仍在大量部署

伦敦，2026年6月25日／美通社／——调查显示，当前大多数大型组织内部，至少有一个AI代理在自主做出决策、访问数据并采取行动，而这些行为并未被任何人全面追踪。尽管各组织领导层深知AI代理可能带来安全隐患、数据泄露和管理失控等问题，但他们仍在不断部署更多此类代理。专家指出，这种“明知故犯”的做法反映出组织在追求效率与风险管控间的失衡，可能引发更严重的系统性风险。 #AI代理 #人工智能 #企业风险 #数据安全 #科技行业 #美通社

谷歌发布Gemini 3.5 Flash，原生集成“计算机使用”工具

谷歌宣布推出Gemini 3.5 Flash模型，首次将“计算机使用”工具直接原生集成至模型本身，取代此前依赖外部测试框架的2.5版本。这一变革使AI模型能够直接调用计算机操作能力（如点击、输入、导航界面），无需额外搭建中间层，从而大幅降低开发门槛。谷歌表示，该方案旨在推动AI从单纯的对话助手向可执行实际任务的“数字同事”演进。开发者可通过API快速构建具备自主操作能力的智能代理，实现代理从概念验证到实际落地的跨越。这一更新有望加速AI在自动化办公、软件测试、数据采集等场景中的应用，标志着大模型向具身化、工具化方向迈出关键一步。 #谷歌 #Gemini #AI #大模型 #计算机使用 #智能代理 #数字同事

Tecan 集成 Agentic AI 加速实验室智能化，携手 NVIDIA 推动数据驱动实验室

瑞士实验室自动化企业 Tecan 于 2026 年 6 月 25 日宣布，将 Agentic AI（自主智能体）能力集成至其实验室分析平台 Introspect 中，并采用 NVIDIA BioNeMo Agent Toolkit，以推动数据驱动型实验室发展。该集成使 AI 智能体能够在 Introspect 内直接调用科学分析能力，实现对实验室数据、工作流与系统性能的持续监测与主动干预，改变传统的被动故障排查模式。增强版 Introspect 已开放早期访问，重点面向制药、生物技术与临床检验等领域。这是双方于 2026 年 3 月达成合作后的首个里程碑，旨在通过 AI 赋能提升科学发现效率与实验室生产力。Tecan 同时注重部署中的透明性、可靠性与可控自动化，确保 Agentic AI 成为关键研究流程的可信技术，并计划继续开发包括 Physical AI 在内的下一代仪器平台。 #Tecan #NVIDIA #AgenticAI #实验室自动化 #数据驱动 #BioNeMo #AI #科技新闻 #制药 #生物技术

亚马逊云科技数据库副总裁G2：AI Agent倒逼数据库进化，重构并非唯一路径

在2026年亚马逊云科技中国峰会上，亚马逊云科技数据库服务副总裁Ganapathy“G2”指出，随着AI Agent从聊天机器人进化为能独立规划与交付业务结果的企业级“数字员工”，底层数据库逻辑正被彻底改写。G2强调，数据才是客户利用先进技术创造新价值的最终差异化来源。他认为，面对AI Agent的大规模部署，数据库无需全面重构，而是需要针对智能体需求进行灵活进化，例如增强实时数据处理、支持复杂工作流。这一观点为企业应对AI时代的数据基础设施变革提供了新的思路。 #亚马逊云科技 #数据库 #AI Agent #数据分析 #科技新闻

Tecan携手NVIDIA推出Agentic AI平台，加速数据驱动实验室部署

2026年6月25日，全球实验室自动化解决方案提供商Tecan宣布，将在其分析平台Introspect中集成Agentic AI能力，采用NVIDIA BioNeMo Agent Toolkit，推动数据驱动实验室建设。该技术使AI智能体能够持续分析实验室数据、工作流与系统表现，主动识别影响效率与质量的潜在问题，并提出可执行建议，实现从传统被动监控到主动优化模式的转变。首批应用聚焦制药、生物技术与临床实验室环境。Tecan生命科学业务部负责人表示，此举旨在打造主动支持科学家、提升生产力的智能实验室解决方案。双方还着重部署了agentic guardrails，确保AI在实验室中的透明性与可靠性，并计划继续共同开发基于Physical AI的下一代实验室仪器。 #Tecan #NVIDIA #AgenticAI #BioNeMo #数据驱动实验室 #实验室自动化 #AI #科技新闻

荣耀定义Agentic OS，下一代终端操作系统7月亮相

在2026年上海世界移动通信大会（MWC上海）开幕首日，荣耀首次系统定义了下一代移动终端操作系统Agentic OS。该操作系统以人为中心，具备意图驱动、自然交互、主动智能及天生跨端四大核心特征，旨在推动终端从“工具”向“伙伴”演进。荣耀产品线总裁方飞表示，AI时代终端将不再是“应用的容器”，而是“智能体的舞台”，终端连接用户与云端生态，并整合算力、传感器和交互能力。荣耀通过AI Agent的感知力、规划力和行动力重构产品形态与人机交互：感知力方面，平台级AI赋能“YOYO建议”实现“服务找人”；规划力方面，部署自研端云大模型矩阵并发布YOYO Claw；行动力方面，推出Magic GUI、Magic Agent大模型及Robot Phone实现具身智能交互。据悉，Agentic OS完整技术框架将于2026年7月发布，阶段性成果通过MagicOS 11亮相，其还将带来安卓首个动态液态玻璃设计。此外，荣耀在GSMA牵头下发布了全球首个《6G终端研究报告》。 #荣耀 #AgenticOS #操作系统 #MWC上海 #AI智能体 #6G #MagicOS #科技新闻