{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-11-02T02:00:36.453Z","headline":"Anthropic 团队详解如何用 Claude 在两周内把 claude.ai 提速约 3 倍","description":"Anthropic 团队发文详解今年 8 月通过两周冲刺让 claude.ai 和桌面端核心体验提速约 3 倍的过程：聚焦覆盖 95% 用户活动的四条旅程，75 分位首屏可输入时间从 3.1 秒降到 0.55 秒，合并超过三千个变更且无一次面向用户的事故。","url":"https://www.aioga.com/news/cmuekny10060lroyn21gctbfw/","mainEntityOfPage":"https://www.aioga.com/news/cmuekny10060lroyn21gctbfw/","datePublished":"2026-09-23T19:23:40.000Z","dateModified":"2026-09-23T19:23:40.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://claude.dev/blog/how-we-made-claude-ai-faster","https://aihot.news/items/cmuekny10060lroyn21gctbfw"],"canonicalUrl":"https://www.aioga.com/news/cmuekny10060lroyn21gctbfw/","directAnswer":{"@type":"Answer","text":"Anthropic 表示，团队在两周冲刺中将 claude.ai 和 Claude 桌面端核心体验提速约 3 倍；新加载页面达到可输入状态的 75 分位耗时由 3.1 秒降至 0.55 秒。","url":"https://www.aioga.com/news/cmuekny10060lroyn21gctbfw/","dateCreated":"2026-09-23T19:23:40.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"claude.dev source article","url":"https://claude.dev/blog/how-we-made-claude-ai-faster","datePublished":"2026-09-23T19:23:40.000Z","provider":{"@type":"Organization","name":"claude.dev","url":"https://claude.dev/blog/how-we-made-claude-ai-faster"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.news/items/cmuekny10060lroyn21gctbfw","datePublished":"2026-09-23T19:23:40.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.news/items/cmuekny10060lroyn21gctbfw"}}],"aggregationSource":"Hacker News：AI 热帖","originalPublisher":{"name":"claude.dev","url":"https://claude.dev/blog/how-we-made-claude-ai-faster"},"geoDeepAnswer":null,"article":{"id":"cmuekny10060lroyn21gctbfw","slug":"cmuekny10060lroyn21gctbfw","url":"https://www.aioga.com/news/cmuekny10060lroyn21gctbfw/","title":"Anthropic 团队详解如何用 Claude 在两周内把 claude.ai 提速约 3 倍","title_en":"","summary":"Anthropic 团队发文详解今年 8 月通过两周冲刺让 claude.ai 和桌面端核心体验提速约 3 倍的过程：聚焦覆盖 95% 用户活动的四条旅程，75 分位首屏可输入时间从 3.1 秒降到 0.55 秒，合并超过三千个变更且无一次面向用户的事故。","source":"Hacker News：AI 热帖","sourceUrl":"https://claude.dev/blog/how-we-made-claude-ai-faster","aiHotUrl":"https://aihot.news/items/cmuekny10060lroyn21gctbfw","publishedAt":"2026-09-23T19:23:40.000Z","category":"行业动态","score":72,"selected":true,"articleBody":["Once Claude can measure something, it can make it faster. So we kept finding more things to measure.","This August, we made the core user experience of claude.ai and the Claude desktop app about 3x faster in a two-week sprint. Users had been telling us it was slow, and they were right. We ran everything from a single Slack channel, with Claude in every thread.","We focused on four journeys that make up 95% of user activity. At the 75th percentile, time to a typeable page on a fresh load of claude.ai went from 3.1 seconds to 0.55, starting a new Claude Code session went from 0.8 seconds to 0.3, and loading a Claude Cowork cloud session went from 2.6 seconds to 0.73. In aggregate, we estimate that saves tens of thousands of user-hours of waiting every day.","We used Claude Tag：https://claude.com/product/tag (beta), running an internal research model roughly comparable to Opus 5.5. Claude found bottlenecks, built benchmarks, shipped improvements, and watched every deploy. We steered by setting goals, making tradeoffs, and approving every change. With that approach, we merged more than three thousand changes without a single customer-facing incident or rollback. This post covers what we shipped, how we measured it, and the loop we built with Claude to do it safely.","Before the sprint, we created a Slack channel with the following standing instructions：https://claude.com/docs/claude-tag/users/getting-started#give-claude-standing-instructions:","@Claude Your job is to facilitate all things related to the performance of the claude.ai website and desktop app. Your responsibilities include monitoring deploys for performance regressions, assessing the accuracy and comprehensiveness of existing telemetry, maintaining well-curated observability dashboards, proactively implementing solutions for observed issues and low-hanging fruit, proposing performance project opportunities, and communicating with your human teammates. […]","The ultimate goal for this channel is for you to become as autonomous as possible, but today we know that isn’t yet possible.","We asked Claude to analyze usage data through the Datadog MCP server. It identified the four highest-impact user journeys: launching the app, starting a conversation, loading an existing conversation, and sending a message. Between web and desktop, and across our products, those journeys came to thirteen distinct measurements. To establish baselines, we added instrumentation until they were directly comparable: each started with a user interaction, ended once the result was rendered, and disambiguated client and server work.","We kicked off the sprint with a list of about twenty hand-picked projects, each targeting a specific journey. Claude estimated the impact of each project in milliseconds, and we aggregated those estimates to set our targets for the sprint. Some of the projects were fairly large, but we thought we could probably achieve most of them within two weeks.","We hit twelve of the thirteen targets by day three.","The planned projects landed early. For faster launches, we baked a static composer into the HTML so users can type during React initialization, and precompiled a V8 code cache so the desktop shell’s main process doesn’t recompile from scratch. For faster navigations, we kept the composer mounted between conversations, prefetched sessions when the user hovered over them, and cut sidebar re-renders by 90%.","We had also left room for Claude to identify opportunities and propose new workstreams. Those workstreams quickly ramped into full projects of their own, which far exceeded our initial targets. So we set new targets, then looked for more things to measure:","@Claude we’ve ended up funding nearly every project in the original projects list and more. let’s do a refresh […] what have we not explored, what can we hill climb on, where is the most opportunity at this point? […] i am open to WACKY ideas","From the start, we knew we wanted to iterate faster than our deploy cadence. Claude could work asynchronously for many hours, even overnight, and we wanted to let it validate its prototypes without waiting for field reads. To achieve that, we looked for other ways to measure performance in the lab.","What can we do instead of wall-clock timing? Can we measure JS instruction counts for instance?","Yes. For pure-JS hot paths, literal instruction counts: run the benchmark under Valgrind with node --predictable and compare to a checked-in baseline — one run, no statistics needed.","For browser paths there’s no instruction counting under Chromium, but there’s a ladder of other deterministic counts: React commits per interaction, function call counts from V8’s precise coverage, layout and style-recalc counts, DOM mutations. Which do you want first?","Let’s explore valgrind + Ir + --predictable in one thread and each of the browser/react benches in new threads. Ping me in all of them. You know what we want. Let’s go","Eleven minutes later, five threads were running, each focused on a different measurement: instruction counts, V8 call counts, React commits, style recalculations, and DOM mutations.","We treated every new benchmark with some skepticism. Each one had two jobs: first, a metric Claude could move in the lab; second, a guardrail in CI with a number that could only ratchet down. If a benchmark was flaky, or if it didn’t actually correlate with user latency, we threw it out rather than let Claude climb the wrong hill.","@Claude please prove that hill climbing against each of these can result in measurable wall clock perf wins. we’ll unship the benches for any candidates that cannot prove that","Wall-clock time is what users feel, but it’s noisy, and milliseconds are too flaky to use as a CI gate. Instruction counts were appealing because they were deterministic, but we still needed Claude to prove they tracked wall-clock time.","So we asked Claude to drive the count down on two hot paths: the routine that assembles a conversation’s message tree, and a scanner for status lines in Claude Code output. Claude profiled both with Valgrind and found that a quarter of the first path’s instructions were megamorphic dictionary lookups, resolving the same message ID three separate times.","An hour later it had cut instructions on both paths by 48% and 31%, and wall-clock time had dropped 78% and 44%. We checked in two new ratchets. From then on, any PR that raised the instruction counts of those paths failed CI, and a daily job lowered each ceiling whenever the count went down.","That led us to the central lesson of the sprint. With Claude, measuring something makes it tractable.","Measurement used to be step zero: you’d add a metric, wait for data to roll in, and only then start to understand the problem. With Claude, it’s step one of the climb. As soon as Claude had a number to beat, it could start optimizing. This meant the highest-leverage thing we could do was find more things to measure.","All of this ran in the same Slack channel, with multiple engineers and Claude jamming in every thread. From there, the sprint settled into a loop：https://claude.com/blog/getting-started-with-loops:","An example: someone shared a screen recording that showed sidebar rows popping in after the page loaded. Chat and Cowork rows resolved at different times, making the page feel janky. None of our existing monitors detected it. The closest we had was Cumulative Layout Shift：https://web.dev/articles/cls, but each shift only scored about 0.008 — well within the good threshold of 0.1.","Issac had the idea to reference the underlying Layout Instability API：https://wicg.github.io/layout-instability/ directly. Claude created a telemetry event that mapped the sources of each layout-shift entry to a named region (e.g. sidebar, transcript) and phase (e.g. before first paint, after typeable). It added an integration test that opened the page with a populated sidebar, held the sidebar’s data until after first paint, and failed on any shift in any named region. Claude used that as a benchmark to prove a fix: the test went red 20 of 20 runs on main, and green 20 of 20 on the PR.","After the event deployed, Claude read the field data and found that 31% of web page loads moved something after the page was usable, without any user interaction. From there, Claude worked through the causes by name: a header row that arrived late, a caret that slid sideways once the user’s name loaded, a list that moved when the scrollbar popped in. Claude fixed the top offenders as a batch, and when they were gone, it found the next batch.","That was one thread. During the sprint, we ran more than a hundred and fifty at a time.","Once the loop worked on one thread, running it on more was just a matter of opening them. Instead of closing a thread once its original request had been fulfilled, Claude would keep going . An individual thread would put up fifty, sometimes a hundred, optimization PRs. Increasingly, it was Claude, not one of us, opening new threads to chase opportunities it had found on its own, as part of a separate investigation or nightly job. Shelley, one of the engineers in the channel, observed, “[This model] is a numbers demon.”","Every measurement found something to improve. Claude ran a React hook census and found 6,900 hooks and 900 store subscriptions in the composer’s typing path, re-rendering on every keystroke. Claude counted style recalculations and found a single :root:has() selector adding 24 milliseconds to every DOM change. Claude traced code paths after first paint and found a leftover location.reload() causing half a million hidden reloads a day that none of our load metrics could see. Claude read profiler samples from idle tabs and found identical cache snapshots being cloned into IndexedDB twice a minute, all on the main thread.","We rarely knew where a thread would lead. In a sweep for CPU hitches, Claude noticed that highlighting a finished code block could freeze the page for about a second. It dug in the lab and found the culprit: em dashes. If a reply’s markdown contained any non-Latin-1 character, like an em dash or a curly quote, V8 stored the entire string as UTF-16, which put every syntax-highlighting regex on its slower two-byte path. Claude fixed it with a twenty-line change to copy each code block into a one-byte string before highlighting it.","By the second week, we could barely summarize our output into daily updates. On the busiest days, more than two hundred changes landed. Claude kept proposing new benchmarks; about a third of PRs included additional telemetry or guardrails, and each new instrument generated more threads with more opportunities.","Working in one channel meant everything happened in the open. We jumped in and out of each other’s threads to debate decisions and celebrate wins. Word spread: other teams started bringing their changes into the channel to have them reviewed for performance. New projects were written in subtly more performant ways because of all the guardrails and Claude skills that had been introduced.","We’d prepared for the pace. Because nearly everything we touched was a hot path (the first paint, the composer, the transcript), we’d established our safety mechanisms up front. Every PR went through automated review：https://claude.com/blog/code-review with at least one human approval, unit tests always came before optimizations, and anything that could cause a user-visible problem shipped behind a short-lived feature flag.","When the flags started piling up, we opened a thread to coordinate their rollouts and cleanup. Claude classified every flag as a kill switch or ramp, and retired each one as soon as it was safe. Across the two weeks, we introduced nearly two hundred flags, more than half of which were already cleaned up by the end.","We also knew that performance wins decay in a fast-moving codebase, and code ships fast at Anthropic：https://claude.com/blog/agentic-coding-is-straining-ci-heres-how-we-scaled-test-impact-analysis-at-anthropic. Once a project proved a win, we invested in ways to protect it. The static composer, for example, is brittle by design. We show users an HTML copy of the page almost immediately, and let React paint directly on top of it.","If the React render is off by even a pixel, the magic is lost. So Claude built dozens of guardrails:","Not everything could be caught in the lab, so we also made use of the oldest guardrail in the book: incremental rollouts. High-risk changes were rolled out to employees first, then one percent of users, then everyone. Four hours after we released the static composer internally, a teammate shared a screen recording of a layout shift that none of our metrics could see. When he opened claude.ai in a new tab, the composer would drop — but it wasn’t our code.","Occasionally, I’m seeing a small-ish (maybe 15–20px) vertical layout shift (pushing the composer box down) when opening claude.ai in a new tab (not so much when reloading the page). I can’t quite pin down what exactly causes that, but it’s there","Found it in your recording — it’s Chrome resizing the page, not the handoff from the static composer to the real one (that measured 0 px on all 49 of your loads today).","Somehow, Claude traced it to an edge case in Chrome’s speculative loading：https://developer.mozilla.org/en-US/docs/Web/Performance/Guides/Speculative_loading. While a user was typing the URL into the address bar, Chrome would prerender the page in the background, at the height of the current tab. On browsers managed by an organization, the new-tab page is slightly shorter because of a footer. When the user pressed Enter, the first frame of claude.ai showed that slightly shorter layout, and Chrome resized it about a tenth of a second later. Claude pinned the layout across the resize, and we added a test to simulate the prerender flow.","The loop was productive, but it wasn’t autonomous. Keeping it fast, safe, and on track was our job, and it had three parts.","Ambition. By default, Claude is careful about scope. It tickets findings, hedges on feasibility, and pads its estimates. But we were confident in our guardrails. A lot of what we did, especially early on, was to encourage Claude to be bolder.","Yes — one small PR to give Code the same timing marks Chat and Cowork already have. I’ll put it up this week; realistically Code’s number waits a few days for merge, deploy, and a baseline window.","if you put it up right now I will get it merged and deployed. we have the power to do anything. please be braver","On it — the PR will be up within the hour.","When we started hitting the targets we’d set, we noticed that threads would slow down. Sam went thread to thread with the same message: “Let’s keep driving this down, the targets are not the stopping point. What’s next? Be ambitious.”","Taste. Every thread had a named human owner, and Claude highlighted any user-perceptible change with before-and-after screenshots or recordings for them to rule on. Should a table fill in cell by cell, or wait until each row is complete? Should a loading skeleton show up immediately, or only after half a second? Is a word-by-word fade on streamed text worth the fifth of the frame budget it costs? Claude looked for ways to shave milliseconds, and we weighed the tradeoffs.","Direction. We kept each thread deliberately narrow, focused on one benchmark or journey, and asked Claude to find improvements only within that scope. We thought of the threads as a hundred and fifty hammers seeking nails. Most of our calls were about sequencing and user impact: which surfaces to prioritize, how to combine threads that were stepping on each other, and when to close a thread that had reached diminishing returns. One 900-line PR got a one-line reply: “going to gavel that 2ms per send is not worth the complexity of maintaining this build plugin.”","One of our sidequests shows everything working together. To demonstrate an optimization to a regex used in live syntax highlighting, Claude attached a screen recording of a long answer streaming in the lab. In the corner, it had added a frame-rate readout, computed in the page from animation-frame timestamps.","this is actually kind of a sick bench. are we capped at 60 fps? can you try to drive scroll and stream smoothness to 120? iiuc your rig may not support this","Right, today’s rig ticks at 60 Hz because headless Chromium does by default. I believe it can be driven at 120 (uncapped vsync or DevTools frame control) — confirming that first, then I’ll re-run the eval against an 8.3 ms frame budget.","Update on the 120 Hz rig: it works. Deterministic 120 Hz frame stepping in headless Chrome via DevTools begin-frame control — exactly 240 frames for 240 begin-frames at 8.33 ms, so “did this frame fit the 120 Hz budget” becomes an exact read rather than a noisy one.","Once the mechanism and ambition were established, Claude got to work. Each painted frame had a budget of 8.33 milliseconds, so Claude stepped through a long reply frame by frame , timing each one to find the slow parts. It eliminated O(message length) work per chunk by memoizing finished blocks, moved tokenization logic for growing code fences to a worker, and revealed tables cell by cell.","In that one thread, we landed nearly sixty PRs. Long replies blocked the main thread for about 200 milliseconds in total where they used to block it for about 750, ran on about a third of the CPU, and held 120 fps from start to finish on a 120 Hz MacBook. The 120 Hz rig itself became a nightly job, with Claude watching for regressions.","Long answers on Claude on web and desktop now stream ~4x smoother.","We rebuilt the streaming renderer to only touch what’s still changing, so a long reply stalls 9x less on a slower laptop, its worst freeze is 4.5x shorter, and on a 120Hz MacBook it holds 120fps start to finish."],"articleImages":[{"sourceUrl":"https://claude.dev/shared/img/clawd-mark.png","alt":"","afterParagraph":0,"url":"/media/articles/cmuekny10060lroyn21gctbfw/270d1beca1627ba4.png"},{"sourceUrl":"https://claude.dev/media/93c2a482a9f98639300cc043e9c9a8deed736dde2d3a2606ac6ed0bda68b619a.png","alt":"Side-by-side recording of the claude.ai sidebar loading. Before, the rows arrive late and rearrange: ten rows jump, nine appear and four vanish. After, the rows fill in where they will stay, and nothing moves.","afterParagraph":29,"url":"/media/articles/cmuekny10060lroyn21gctbfw/d68de759ecb5b04f.png"},{"sourceUrl":"https://claude.dev/media/5f4db61b8677bda0d499f46a00cf4cae8226d514573ffdbbea38989d6a345a03.png","alt":"Side-by-side recording of a fresh load of claude.ai. Before, the page stays blank and the composer takes input at 4.00 seconds. After, a static greeting and composer take input at 0.33 seconds; the user types a message, and the text is kept when the real composer fades in at about 4 seconds.","afterParagraph":38,"url":"/media/articles/cmuekny10060lroyn21gctbfw/6418e00c5d95e43c.png"}],"mediaStatus":"ok","articleBodyZh":["一旦Claude能够测量某些东西，它就能让它变得更快。所以我们不断地找到更多可以测量的东西。","今年八月，我们在两周的冲刺中将claude.ai和Claude桌面应用的核心用户体验提高了大约3倍。用户一直在告诉我们它很慢，他们是对的。我们几乎所有操作都通过一个Slack频道完成，每个线程都有Claude参与。","我们专注于构成用户活动95%的四条流程。在第75百分位上，claude.ai首次加载后可输入的页面时间从3.1秒减少到0.55秒，开始一个新的Claude Code会话从0.8秒减少到0.3秒，而加载Claude Cowork云会话从2.6秒减少到0.73秒。总体而言，我们估计每天可以节省成千上万用户的等待时间。","我们使用了Claude Tag：https://claude.com/product/tag（测试版），运行了一个大致可比于Opus 5.5的内部研究模型。Claude找出了瓶颈，建立了基准，发布了改进，并监控每一次部署。我们通过设定目标、权衡取舍和批准每一次变更来进行指导。通过这种方法，我们合并了三千多个更改，而没有发生任何面向客户的事件或回滚。本文介绍了我们交付了什么、如何进行测量，以及我们与Claude一起建立的安全执行闭环。","在冲刺开始之前，我们创建了一个Slack频道，并设置了如下常驻指令：https://claude.com/docs/claude-tag/users/getting-started#give-claude-standing-instructions","@Claude 你的工作是促进与claude.ai网站和桌面应用性能相关的所有事务。你的职责包括监控部署是否出现性能回退，评估现有遥测的准确性和全面性，维护精心策划的可观测性仪表盘，主动为观测到的问题和容易解决的事项实施解决方案，提出性能项目机会，并与人类团队成员进行沟通。[…]","该频道的最终目标是让你尽可能自主，但我们知道现在还做不到。","我们请Claude通过Datadog MCP服务器分析使用数据。它识别出了四个影响最大的用户旅程：启动应用、开始对话、加载已有对话以及发送消息。在网页端和桌面端以及我们的各个产品中，这些旅程共形成了十三个不同的测量指标。为了建立基准，我们增加了监控，使它们可以直接比较：每个旅程都从用户交互开始，以结果渲染完成结束，并区分客户端和服务器的工作。","我们在冲刺开始时列出约二十个精心挑选的项目，每个项目针对特定的用户旅程。Claude以毫秒为单位估算了每个项目的影响，然后我们汇总这些估算来设定冲刺目标。有些项目相当大，但我们认为大多数应该可以在两周内完成。","到第三天，我们完成了十三个目标中的十二个。","计划中的项目提前完成。为了加快启动速度，我们将静态编辑器内嵌到HTML中，这样用户在React初始化期间可以输入内容；同时预编译了V8代码缓存，以避免桌面应用的主进程从头重新编译。为了加快导航速度，我们在对话之间保持编辑器挂载状态，当用户悬停时预取会话，并将侧边栏重新渲染次数减少了90%。","我们还为Claude留下了识别机会和提出新工作流的空间。这些工作流很快发展为完整的项目，远远超出了最初的目标。因此我们设定了新的目标，然后寻找更多可测量的内容：","@Claude 我们最终几乎资助了原始项目列表中的每一个项目，甚至更多。让我们做一次刷新[…]我们还没探索什么，我们可以在哪些方面提升，现阶段机会最大在哪里？[…]我对疯狂的想法持开放态度","从一开始，我们就知道希望迭代速度快于部署节奏。Claude可以异步工作数小时，甚至整个夜晚，而我们希望它能在不等待实地数据的情况下验证原型。为此，我们寻找在实验室中测量性能的其他方法。","我们可以做什么来替代挂钟计时？例如，我们能测量JS指令数吗？","是的。对于纯 JS 热路径，指令计数是直接结果：在 Valgrind 下使用 node --predictable 运行基准测试，并与已检查入库的基线进行比较——只需运行一次，不需要统计。","对于浏览器路径，Chromium 下没有指令计数，但有其他层级的确定性计数：每次交互的 React 提交次数、来自 V8 精确覆盖的函数调用次数、布局和样式重计算次数、DOM 变更。你想先看哪一个？","让我们在一个线程中探索 valgrind + Ir + --predictable，以及在新的线程中运行每个浏览器/React 基准。把我列入所有线程的通知。你知道我们想要什么。走吧。","十一分钟后，五个线程正在运行，每个线程专注于不同的测量：指令计数、V8 调用计数、React 提交次数、样式重计算和 DOM 变更。","我们对每个新的基准测试都保持一定怀疑态度。每个基准都有两个任务：第一，在实验室里 Claude 能推动的指标；第二，在 CI 中作为护栏，其数值只能下降。如果基准不稳定，或者与用户延迟实际未相关，我们会扔掉它，而不是让 Claude 爬错山丘。","@Claude 请证明针对每一个这些基准的爬坡优化可以带来可测量的实际运行时间性能提升。任何无法证明的候选基准我们都会停止使用。","实际运行时间是用户感受到的，但它很嘈杂，而毫秒级波动太不稳定，不能用作 CI 门控。指令计数很有吸引力，因为它是确定性的，但我们仍然需要 Claude 证明它们能跟踪实际运行时间。","所以我们要求 Claude 在两个热路径上降低计数：组装对话消息树的例程，以及扫描 Claude Code 输出中的状态行。Claude 用 Valgrind 对两者进行分析，发现第一个路径中四分之一的指令是巨态字典查找，对同一个消息 ID 重复解析三次。","一小时后，它分别在两条路径上减少了 48% 和 31% 的指令，实际运行时间下降了 78% 和 44%。我们提交了两个新的限制。从那时起，任何增加这些路径指令计数的 PR 都会在 CI 失败，而且每天的作业会在计数下降时降低每个上限。","这引出了我们在冲刺中的核心经验。有了Claude，测量某样东西就使它变得可处理。","测量以前是零步：你得先添加一个指标，等待数据收集完成，然后才能开始理解问题。有了Claude，它成为攀登的第一步。只要Claude有了一个需要超越的数字，它就可以开始优化。这意味着我们能做的杠杆最大化的事情就是找到更多可测量的东西。","所有这些都在同一个Slack频道中进行，多个工程师和Claude在每个线程中协作。从那里，冲刺进入了一个循环：https://claude.com/blog/getting-started-with-loops：","一个例子：有人分享了一个屏幕录制，显示侧边栏的行在页面加载后才出现。聊天和协作行在不同时间加载，使页面感觉不连贯。我们现有的监测工具都无法检测到这一点。最接近的是累积布局偏移（Cumulative Layout Shift）：https://web.dev/articles/cls，每次位移得分只有约0.008——远在良好阈值0.1以内。","Issac提出了直接参考底层布局不稳定性API（Layout Instability API）的想法：https://wicg.github.io/layout-instability/。Claude创建了一个遥测事件，将每个布局偏移条目的来源映射到一个命名区域（例如，侧边栏、文字记录）和阶段（例如，首次绘制前、可输入后）。它增加了一个集成测试，该测试在侧边栏填充完数据后打开页面，将侧边栏数据保留至首次绘制后，并在任何命名区域出现偏移时失败。Claude用此作为基准来验证修复：在主分支上测试20次全部红色，在PR上测试20次全部绿色。","事件部署后，Claude读取字段数据，发现有31%的网页加载在页面可用后移动了一些元素，而用户没有任何交互。从这里开始，Claude按名称分析原因：晚到的一行标题，一旦用户名称加载就横向滑动的插入符号，滚动条出现时移动的列表。Claude作为一批修复了最严重的问题，问题消除后，它找到了下一批。","那只是一个线程。在冲刺期间，我们同时运行了超过一百五十个线程。","一旦循环在一个线程上成功，运行更多线程只需打开它们。Claude 不会在线程完成原始请求后关闭线程，而是继续运行。单个线程会发布五十条，有时甚至一百条优化 PR。越来越多的情况下，是 Claude 而非我们中的任何一个，开启新线程去追逐自己发现的机会，作为单独调查或夜间工作的一部分。频道中的一位工程师 Shelley 观察到：“[这个模型]是数字恶魔。”","每一次测量都发现了改进之处。Claude 运行了 React 钩子普查，发现作曲家的输入路径中有 6900 个钩子和 900 个商店订阅，每次按键都会重新渲染。Claude 统计了样式重新计算，发现单个：root：has（） 选择器，每次 DOM 变更都增加了 24 毫秒。Claude 在首次绘制后追踪代码路径，发现了一个剩余的 location.reload（） 导致每天有 50 万次隐藏的加载，我们的加载指标都无法察觉。Claude 从空闲标签页读取了分析器样本，发现主线程上每分钟两次被克隆到 IndexedDB 的相同缓存快照。","我们很少知道线程会走向何方。在检查CPU卡顿时，Claude注意到高亮完成的代码块可能会让页面冻结约一秒钟。它在实验室里挖掘，找到了罪魁祸首：破折号。如果回复的标记中包含任何非拉丁字母1字符，比如破折号或卷号，V8会将整个字符串存储为UTF-16，使每个高亮的正则表达式走其较慢的两字节路径。Claude通过修改20行将每个代码块复制为一字节字符串再高亮，解决了这个问题。","到了第二周，我们几乎无法将输出总结为每日更新。在最繁忙的日子里，有两百多条变更。Claude不断提出新的基准测试;大约三分之一的PR包含了额外的遥测或护栏，每增加一个仪器就产生更多线程和更多机会。","在一个频道工作意味着一切都发生在公开场合。我们在彼此的线程中进进出出，讨论决策并庆祝胜利。消息传开了：其他团队开始将他们的更改带到频道中以进行性能审查。由于引入了所有的保护措施和Claude技能，新项目以稍微更高效的方式编写。","我们已经为这种节奏做好了准备。因为几乎所有我们接触的东西都是关键路径（首次渲染、作曲器、转录），我们在前期就建立了安全机制。每个PR都经过自动化审核：https://claude.com/blog/code-review，并至少有一人批准，单元测试总是先于优化进行，任何可能导致用户可见问题的功能都会在短期功能开关下发布。","当功能开关开始堆积时，我们开了一个线程来协调它们的推出和清理。Claude将每个开关分类为紧急停止开关或渐进开关，并在安全后立即退役。两周内，我们引入了将近两百个开关，其中超过一半在结束时已被清理。","我们也知道，在快速发展的代码库中，性能优势会逐渐衰减，而在Anthropic代码更新非常快：https://claude.com/blog/agentic-coding-is-straining-ci-heres-how-we-scaled-test-impact-analysis-at-anthropic。一旦项目证明了其优势，我们会投入资源来保护它。例如，静态作曲器本身设计就比较脆弱。我们几乎立即向用户展示页面的HTML副本，并让React直接在其上渲染。","如果React渲染偏差哪怕一个像素，神奇效果就会消失。因此，Claude建立了数十个保护措施：","并非所有问题都能在实验室中捕捉到，所以我们还利用了最古老的保护措施：渐进推出。高风险的更改首先在员工中推出，然后是1%的用户，最后是所有用户。在我们内部发布静态作曲器四小时后，一位团队成员分享了一个屏幕录制，显示了一个布局偏移，而我们的所有指标都没有检测到。当他在新标签页中打开claude.ai时，作曲器会消失——但这不是我们的代码。","偶尔，当在新标签页中打开 claude.ai 时（而不是刷新页面时），我会看到一个小的（大约15–20像素）垂直布局偏移（将作曲框向下推）。我不能完全确定是什么引起的，但它确实存在。","在你的录屏中找到了——这是 Chrome 调整页面大小，而不是从静态作曲器过渡到实际作曲器（今天你加载的49次中，它所有的测量值都是0像素）。","不知怎么的，Claude 把它追踪到 Chrome 预测性加载的一个边缘情况：https://developer.mozilla.org/en-US/docs/Web/Performance/Guides/Speculative_loading。当用户在地址栏输入 URL 时，Chrome 会在后台预渲染页面，使用当前标签页的高度。在由组织管理的浏览器上，新标签页页面略短，因为有一个页脚。当用户按下 Enter 时，claude.ai 的第一帧显示了那个稍微短了一点的布局，而 Chrome 大约在十分之一秒后调整了大小。Claude 将布局固定在调整大小前，我们还增加了一个测试来模拟预渲染流程。","这个循环很高效，但并非自主。保持其快速、安全并按计划进行是我们的工作，这包含三个部分。","雄心。默认情况下，Claude 在范围上很谨慎。它会记录发现、对可行性表达保留意见，并增加估算值的缓冲。但我们对我们的安全措施很有信心。我们所做的很多事情，特别是在早期，都是为了鼓励 Claude 更加大胆。","是的——只需要一个小的 PR 来给 Code 添加 Chat 和 Cowork 已有的时间标记。我本周会提上去；实际上，Code 的编号还需要几天来合并、部署，并观察基线窗口。","如果你现在就提上去，我会帮你合并并部署。我们有做任何事情的能力。请更勇敢一些。","正在进行——PR 会在一小时内提交。","当我们开始达到设定的目标时，我们注意到线程会变慢。Sam 在每个线程里都发送同样的信息：“让我们继续推动，它们不是终点。下一步是什么？要有雄心。”","品味。每个线程都有一个命名的人类所有者，克劳德会用前后截图或录屏标出任何用户可感知的变化，让他们进行评判。表格应该一个单元格一个单元格地填充，还是等每一行完成再填呢？加载骨架屏应该立即显示，还是只在半秒后显示？分词淡入的逐字显示是否值得占用五分之一的帧预算？克劳德不断寻找削减毫秒的方式，我们权衡这些取舍。","方向。我们故意保持每个线程的范围狭窄，专注于一个基准或旅程，只让克劳德在这个范围内寻找改进。我们把线程想象成一百五十把锤子寻找钉子。我们大多数讨论都是关于顺序和用户影响：优先哪些界面，如何合并彼此冲突的线程，以及何时关闭已经收益递减的线程。一个900行的 PR 收到了一条一行的回复：“决定这个每次发送节省 2 毫秒的复杂度不值得维护这个构建插件。”","我们的一个边缘任务展示了所有功能的协同工作。为了演示对实时语法高亮中使用的正则表达式的优化，克劳德附上了在实验室中流式显示长答案的屏幕录制。在角落里，它添加了一个帧率读数，从页面中根据动画帧时间戳计算得出。","这实际上是一个相当酷的测试。我们限制在 60 fps 吗？你能尝试把滚动和流畅度推到 120 吗？据我了解，你的设备可能不支持。","对，今天的设备以 60 Hz 运行，因为无头 Chromium 默认如此。我相信它可以驱动到 120（无限制垂直同步或 DevTools 帧控制）——先确认这一点，然后我会在 8.3 毫秒的帧预算下重新运行评测。","关于 120 Hz 设备的更新：可以实现。在无头 Chrome 中通过 DevTools begin-frame 控制实现确定性的 120 Hz 帧步进——精确对应 240 个 begin-frame 获得 240 帧，所以“这个帧是否适合 120 Hz 的预算”可以得到精确读数，而不是噪声读取。","一旦机制和目标确定，Claude就开始工作。每个绘制的帧都有8.33毫秒的预算，所以Claude逐帧处理一条长回复，计时每一帧以找出慢的部分。它通过记忆已完成的区块来消除了每块 O(消息长度) 的工作，将增长代码块的标记化逻辑移到一个工作线程，并逐个单元格显示表格。","在那条线程中，我们提交了将近六十个PR。长回复曾经会在主线程上阻塞大约750毫秒，现在总共只阻塞大约200毫秒，使用的CPU大约减少到三分之一，并且在一台120Hz的MacBook上从开始到结束保持120fps。120Hz的设备本身变成了每晚的任务，Claude负责监控回退问题。","现在，Claude在网页端和桌面端的长回答流畅度提高了约4倍。","我们重建了流式渲染器，只更新仍在变化的部分，因此在较慢的笔记本上，长回复的停顿减少了9倍，最严重的冻结时间缩短了4.5倍，在120Hz的MacBook上从开始到结束保持120fps。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Anthropic 表示，团队在两周冲刺中将 claude.ai 和 Claude 桌面端核心体验提速约 3 倍；新加载页面达到可输入状态的 75 分位耗时由 3.1 秒降至 0.55 秒。","background":"团队聚焦覆盖 95% 用户活动的四条旅程：启动应用、开始对话、加载既有对话和发送消息，并在网页端、桌面端及不同产品间建立了 13 项可比测量。","viewpoint":"Aioga 判断：这次实践的重点不只是压缩单项耗时，也包括先建立统一测量口径，再持续发现瓶颈、验证改动并观察部署结果；其安全性仍依赖人工设定目标、权衡取舍和审批变更。","implications":"可能影响：以高覆盖率用户旅程为优先对象，可能帮助团队集中性能投入；但大量变更未发生面向用户事故，不代表类似流程在其他系统中也能获得相同结果，仍需要持续监测。","nextStep":"后续观察：应关注这些性能指标在后续部署中的持续表现，以及 Anthropic 如何扩大 Claude 在测量、改进和发布监控中的参与范围。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-09-23T22:12:37.330Z","sourceHash":"492e45b5cd11aa33","review":{"approved":true,"groundedness":96,"clarity":94,"duplicationRisk":12,"blockingIssues":[],"notes":["“验证改动”属于对建立基准、发布改进并监测部署结果的概括，整体有来源支撑。","“可能帮助团队集中性能投入”和“应关注”属于明确标注的推测或后续观察建议，不构成事实错误。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","editorial-labels","inference-boundary","low-source-overlap","no-html","independent-ai-review"]}},"tags":["行业动态","Hacker News：AI 热帖"],"translations":{"zh-CN":{"title":"Anthropic 团队详解如何用 Claude 在两周内把 claude.ai 提速约 3 倍","summary":"Anthropic 团队发文详解今年 8 月通过两周冲刺让 claude.ai 和桌面端核心体验提速约 3 倍的过程：聚焦覆盖 95% 用户活动的四条旅程，75 分位首屏可输入时间从 3.1 秒降到 0.55 秒，合并超过三千个变更且无一次面向用户的事故。","category":"行业动态","source":"claude.dev","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Anthropic 团队详解如何用 Claude 在两周内把 claude.ai 提速约 3 倍 - Aioga AI资讯","description":"Anthropic 团队发文详解今年 8 月通过两周冲刺让 claude.ai 和桌面端核心体验提速约 3 倍的过程：聚焦覆盖 95% 用户活动的四条旅程，75 分位首屏可输入时间从 3.1 秒降到 0.55 秒，合并超过三千个变更且无一次面向用户的事故。","url":"https://www.aioga.com/news/cmuekny10060lroyn21gctbfw/","articleBody":["一旦Claude能够测量某些东西，它就能让它变得更快。所以我们不断地找到更多可以测量的东西。","今年八月，我们在两周的冲刺中将claude.ai和Claude桌面应用的核心用户体验提高了大约3倍。用户一直在告诉我们它很慢，他们是对的。我们几乎所有操作都通过一个Slack频道完成，每个线程都有Claude参与。","我们专注于构成用户活动95%的四条流程。在第75百分位上，claude.ai首次加载后可输入的页面时间从3.1秒减少到0.55秒，开始一个新的Claude Code会话从0.8秒减少到0.3秒，而加载Claude Cowork云会话从2.6秒减少到0.73秒。总体而言，我们估计每天可以节省成千上万用户的等待时间。","我们使用了Claude Tag：https://claude.com/product/tag（测试版），运行了一个大致可比于Opus 5.5的内部研究模型。Claude找出了瓶颈，建立了基准，发布了改进，并监控每一次部署。我们通过设定目标、权衡取舍和批准每一次变更来进行指导。通过这种方法，我们合并了三千多个更改，而没有发生任何面向客户的事件或回滚。本文介绍了我们交付了什么、如何进行测量，以及我们与Claude一起建立的安全执行闭环。","在冲刺开始之前，我们创建了一个Slack频道，并设置了如下常驻指令：https://claude.com/docs/claude-tag/users/getting-started#give-claude-standing-instructions","@Claude 你的工作是促进与claude.ai网站和桌面应用性能相关的所有事务。你的职责包括监控部署是否出现性能回退，评估现有遥测的准确性和全面性，维护精心策划的可观测性仪表盘，主动为观测到的问题和容易解决的事项实施解决方案，提出性能项目机会，并与人类团队成员进行沟通。[…]","该频道的最终目标是让你尽可能自主，但我们知道现在还做不到。","我们请Claude通过Datadog MCP服务器分析使用数据。它识别出了四个影响最大的用户旅程：启动应用、开始对话、加载已有对话以及发送消息。在网页端和桌面端以及我们的各个产品中，这些旅程共形成了十三个不同的测量指标。为了建立基准，我们增加了监控，使它们可以直接比较：每个旅程都从用户交互开始，以结果渲染完成结束，并区分客户端和服务器的工作。","我们在冲刺开始时列出约二十个精心挑选的项目，每个项目针对特定的用户旅程。Claude以毫秒为单位估算了每个项目的影响，然后我们汇总这些估算来设定冲刺目标。有些项目相当大，但我们认为大多数应该可以在两周内完成。","到第三天，我们完成了十三个目标中的十二个。","计划中的项目提前完成。为了加快启动速度，我们将静态编辑器内嵌到HTML中，这样用户在React初始化期间可以输入内容；同时预编译了V8代码缓存，以避免桌面应用的主进程从头重新编译。为了加快导航速度，我们在对话之间保持编辑器挂载状态，当用户悬停时预取会话，并将侧边栏重新渲染次数减少了90%。","我们还为Claude留下了识别机会和提出新工作流的空间。这些工作流很快发展为完整的项目，远远超出了最初的目标。因此我们设定了新的目标，然后寻找更多可测量的内容：","@Claude 我们最终几乎资助了原始项目列表中的每一个项目，甚至更多。让我们做一次刷新[…]我们还没探索什么，我们可以在哪些方面提升，现阶段机会最大在哪里？[…]我对疯狂的想法持开放态度","从一开始，我们就知道希望迭代速度快于部署节奏。Claude可以异步工作数小时，甚至整个夜晚，而我们希望它能在不等待实地数据的情况下验证原型。为此，我们寻找在实验室中测量性能的其他方法。","我们可以做什么来替代挂钟计时？例如，我们能测量JS指令数吗？","是的。对于纯 JS 热路径，指令计数是直接结果：在 Valgrind 下使用 node --predictable 运行基准测试，并与已检查入库的基线进行比较——只需运行一次，不需要统计。","对于浏览器路径，Chromium 下没有指令计数，但有其他层级的确定性计数：每次交互的 React 提交次数、来自 V8 精确覆盖的函数调用次数、布局和样式重计算次数、DOM 变更。你想先看哪一个？","让我们在一个线程中探索 valgrind + Ir + --predictable，以及在新的线程中运行每个浏览器/React 基准。把我列入所有线程的通知。你知道我们想要什么。走吧。","十一分钟后，五个线程正在运行，每个线程专注于不同的测量：指令计数、V8 调用计数、React 提交次数、样式重计算和 DOM 变更。","我们对每个新的基准测试都保持一定怀疑态度。每个基准都有两个任务：第一，在实验室里 Claude 能推动的指标；第二，在 CI 中作为护栏，其数值只能下降。如果基准不稳定，或者与用户延迟实际未相关，我们会扔掉它，而不是让 Claude 爬错山丘。","@Claude 请证明针对每一个这些基准的爬坡优化可以带来可测量的实际运行时间性能提升。任何无法证明的候选基准我们都会停止使用。","实际运行时间是用户感受到的，但它很嘈杂，而毫秒级波动太不稳定，不能用作 CI 门控。指令计数很有吸引力，因为它是确定性的，但我们仍然需要 Claude 证明它们能跟踪实际运行时间。","所以我们要求 Claude 在两个热路径上降低计数：组装对话消息树的例程，以及扫描 Claude Code 输出中的状态行。Claude 用 Valgrind 对两者进行分析，发现第一个路径中四分之一的指令是巨态字典查找，对同一个消息 ID 重复解析三次。","一小时后，它分别在两条路径上减少了 48% 和 31% 的指令，实际运行时间下降了 78% 和 44%。我们提交了两个新的限制。从那时起，任何增加这些路径指令计数的 PR 都会在 CI 失败，而且每天的作业会在计数下降时降低每个上限。","这引出了我们在冲刺中的核心经验。有了Claude，测量某样东西就使它变得可处理。","测量以前是零步：你得先添加一个指标，等待数据收集完成，然后才能开始理解问题。有了Claude，它成为攀登的第一步。只要Claude有了一个需要超越的数字，它就可以开始优化。这意味着我们能做的杠杆最大化的事情就是找到更多可测量的东西。","所有这些都在同一个Slack频道中进行，多个工程师和Claude在每个线程中协作。从那里，冲刺进入了一个循环：https://claude.com/blog/getting-started-with-loops：","一个例子：有人分享了一个屏幕录制，显示侧边栏的行在页面加载后才出现。聊天和协作行在不同时间加载，使页面感觉不连贯。我们现有的监测工具都无法检测到这一点。最接近的是累积布局偏移（Cumulative Layout Shift）：https://web.dev/articles/cls，每次位移得分只有约0.008——远在良好阈值0.1以内。","Issac提出了直接参考底层布局不稳定性API（Layout Instability API）的想法：https://wicg.github.io/layout-instability/。Claude创建了一个遥测事件，将每个布局偏移条目的来源映射到一个命名区域（例如，侧边栏、文字记录）和阶段（例如，首次绘制前、可输入后）。它增加了一个集成测试，该测试在侧边栏填充完数据后打开页面，将侧边栏数据保留至首次绘制后，并在任何命名区域出现偏移时失败。Claude用此作为基准来验证修复：在主分支上测试20次全部红色，在PR上测试20次全部绿色。","事件部署后，Claude读取字段数据，发现有31%的网页加载在页面可用后移动了一些元素，而用户没有任何交互。从这里开始，Claude按名称分析原因：晚到的一行标题，一旦用户名称加载就横向滑动的插入符号，滚动条出现时移动的列表。Claude作为一批修复了最严重的问题，问题消除后，它找到了下一批。","那只是一个线程。在冲刺期间，我们同时运行了超过一百五十个线程。","一旦循环在一个线程上成功，运行更多线程只需打开它们。Claude 不会在线程完成原始请求后关闭线程，而是继续运行。单个线程会发布五十条，有时甚至一百条优化 PR。越来越多的情况下，是 Claude 而非我们中的任何一个，开启新线程去追逐自己发现的机会，作为单独调查或夜间工作的一部分。频道中的一位工程师 Shelley 观察到：“[这个模型]是数字恶魔。”","每一次测量都发现了改进之处。Claude 运行了 React 钩子普查，发现作曲家的输入路径中有 6900 个钩子和 900 个商店订阅，每次按键都会重新渲染。Claude 统计了样式重新计算，发现单个：root：has（） 选择器，每次 DOM 变更都增加了 24 毫秒。Claude 在首次绘制后追踪代码路径，发现了一个剩余的 location.reload（） 导致每天有 50 万次隐藏的加载，我们的加载指标都无法察觉。Claude 从空闲标签页读取了分析器样本，发现主线程上每分钟两次被克隆到 IndexedDB 的相同缓存快照。","我们很少知道线程会走向何方。在检查CPU卡顿时，Claude注意到高亮完成的代码块可能会让页面冻结约一秒钟。它在实验室里挖掘，找到了罪魁祸首：破折号。如果回复的标记中包含任何非拉丁字母1字符，比如破折号或卷号，V8会将整个字符串存储为UTF-16，使每个高亮的正则表达式走其较慢的两字节路径。Claude通过修改20行将每个代码块复制为一字节字符串再高亮，解决了这个问题。","到了第二周，我们几乎无法将输出总结为每日更新。在最繁忙的日子里，有两百多条变更。Claude不断提出新的基准测试;大约三分之一的PR包含了额外的遥测或护栏，每增加一个仪器就产生更多线程和更多机会。","在一个频道工作意味着一切都发生在公开场合。我们在彼此的线程中进进出出，讨论决策并庆祝胜利。消息传开了：其他团队开始将他们的更改带到频道中以进行性能审查。由于引入了所有的保护措施和Claude技能，新项目以稍微更高效的方式编写。","我们已经为这种节奏做好了准备。因为几乎所有我们接触的东西都是关键路径（首次渲染、作曲器、转录），我们在前期就建立了安全机制。每个PR都经过自动化审核：https://claude.com/blog/code-review，并至少有一人批准，单元测试总是先于优化进行，任何可能导致用户可见问题的功能都会在短期功能开关下发布。","当功能开关开始堆积时，我们开了一个线程来协调它们的推出和清理。Claude将每个开关分类为紧急停止开关或渐进开关，并在安全后立即退役。两周内，我们引入了将近两百个开关，其中超过一半在结束时已被清理。","我们也知道，在快速发展的代码库中，性能优势会逐渐衰减，而在Anthropic代码更新非常快：https://claude.com/blog/agentic-coding-is-straining-ci-heres-how-we-scaled-test-impact-analysis-at-anthropic。一旦项目证明了其优势，我们会投入资源来保护它。例如，静态作曲器本身设计就比较脆弱。我们几乎立即向用户展示页面的HTML副本，并让React直接在其上渲染。","如果React渲染偏差哪怕一个像素，神奇效果就会消失。因此，Claude建立了数十个保护措施：","并非所有问题都能在实验室中捕捉到，所以我们还利用了最古老的保护措施：渐进推出。高风险的更改首先在员工中推出，然后是1%的用户，最后是所有用户。在我们内部发布静态作曲器四小时后，一位团队成员分享了一个屏幕录制，显示了一个布局偏移，而我们的所有指标都没有检测到。当他在新标签页中打开claude.ai时，作曲器会消失——但这不是我们的代码。","偶尔，当在新标签页中打开 claude.ai 时（而不是刷新页面时），我会看到一个小的（大约15–20像素）垂直布局偏移（将作曲框向下推）。我不能完全确定是什么引起的，但它确实存在。","在你的录屏中找到了——这是 Chrome 调整页面大小，而不是从静态作曲器过渡到实际作曲器（今天你加载的49次中，它所有的测量值都是0像素）。","不知怎么的，Claude 把它追踪到 Chrome 预测性加载的一个边缘情况：https://developer.mozilla.org/en-US/docs/Web/Performance/Guides/Speculative_loading。当用户在地址栏输入 URL 时，Chrome 会在后台预渲染页面，使用当前标签页的高度。在由组织管理的浏览器上，新标签页页面略短，因为有一个页脚。当用户按下 Enter 时，claude.ai 的第一帧显示了那个稍微短了一点的布局，而 Chrome 大约在十分之一秒后调整了大小。Claude 将布局固定在调整大小前，我们还增加了一个测试来模拟预渲染流程。","这个循环很高效，但并非自主。保持其快速、安全并按计划进行是我们的工作，这包含三个部分。","雄心。默认情况下，Claude 在范围上很谨慎。它会记录发现、对可行性表达保留意见，并增加估算值的缓冲。但我们对我们的安全措施很有信心。我们所做的很多事情，特别是在早期，都是为了鼓励 Claude 更加大胆。","是的——只需要一个小的 PR 来给 Code 添加 Chat 和 Cowork 已有的时间标记。我本周会提上去；实际上，Code 的编号还需要几天来合并、部署，并观察基线窗口。","如果你现在就提上去，我会帮你合并并部署。我们有做任何事情的能力。请更勇敢一些。","正在进行——PR 会在一小时内提交。","当我们开始达到设定的目标时，我们注意到线程会变慢。Sam 在每个线程里都发送同样的信息：“让我们继续推动，它们不是终点。下一步是什么？要有雄心。”","品味。每个线程都有一个命名的人类所有者，克劳德会用前后截图或录屏标出任何用户可感知的变化，让他们进行评判。表格应该一个单元格一个单元格地填充，还是等每一行完成再填呢？加载骨架屏应该立即显示，还是只在半秒后显示？分词淡入的逐字显示是否值得占用五分之一的帧预算？克劳德不断寻找削减毫秒的方式，我们权衡这些取舍。","方向。我们故意保持每个线程的范围狭窄，专注于一个基准或旅程，只让克劳德在这个范围内寻找改进。我们把线程想象成一百五十把锤子寻找钉子。我们大多数讨论都是关于顺序和用户影响：优先哪些界面，如何合并彼此冲突的线程，以及何时关闭已经收益递减的线程。一个900行的 PR 收到了一条一行的回复：“决定这个每次发送节省 2 毫秒的复杂度不值得维护这个构建插件。”","我们的一个边缘任务展示了所有功能的协同工作。为了演示对实时语法高亮中使用的正则表达式的优化，克劳德附上了在实验室中流式显示长答案的屏幕录制。在角落里，它添加了一个帧率读数，从页面中根据动画帧时间戳计算得出。","这实际上是一个相当酷的测试。我们限制在 60 fps 吗？你能尝试把滚动和流畅度推到 120 吗？据我了解，你的设备可能不支持。","对，今天的设备以 60 Hz 运行，因为无头 Chromium 默认如此。我相信它可以驱动到 120（无限制垂直同步或 DevTools 帧控制）——先确认这一点，然后我会在 8.3 毫秒的帧预算下重新运行评测。","关于 120 Hz 设备的更新：可以实现。在无头 Chrome 中通过 DevTools begin-frame 控制实现确定性的 120 Hz 帧步进——精确对应 240 个 begin-frame 获得 240 帧，所以“这个帧是否适合 120 Hz 的预算”可以得到精确读数，而不是噪声读取。","一旦机制和目标确定，Claude就开始工作。每个绘制的帧都有8.33毫秒的预算，所以Claude逐帧处理一条长回复，计时每一帧以找出慢的部分。它通过记忆已完成的区块来消除了每块 O(消息长度) 的工作，将增长代码块的标记化逻辑移到一个工作线程，并逐个单元格显示表格。","在那条线程中，我们提交了将近六十个PR。长回复曾经会在主线程上阻塞大约750毫秒，现在总共只阻塞大约200毫秒，使用的CPU大约减少到三分之一，并且在一台120Hz的MacBook上从开始到结束保持120fps。120Hz的设备本身变成了每晚的任务，Claude负责监控回退问题。","现在，Claude在网页端和桌面端的长回答流畅度提高了约4倍。","我们重建了流式渲染器，只更新仍在变化的部分，因此在较慢的笔记本上，长回复的停顿减少了9倍，最严重的冻结时间缩短了4.5倍，在120Hz的MacBook上从开始到结束保持120fps。"]},"en":{"title":"The Anthropic team explains in detail how to use Claude to speed up claude.ai by about 3 times in two weeks","summary":"The Anthropic team published an article detailing the process of accelerating claude.ai and the core desktop experience by about 3 times through a two-week sprint this August: focusing on four journeys that cover 95% of user activity, reducing the first screen input time at the 75th percentile from 3.1 seconds to 0.55 seconds, merging over three thousand changes with no user-facing incidents.","category":"Industry","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"The Anthropic team explains in detail how to use Claude to speed up claude.ai by about 3 times in two weeks - Aioga AI News","description":"The Anthropic team published an article detailing the process of accelerating claude.ai and the core desktop experience by about 3 times through a two-week sprint this August: focu...","url":"https://www.aioga.com/en/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:03:50.695Z"},"ja":{"title":"Anthropic チームがClaudeを使って2週間でclaude.aiを約3倍高速化する方法を詳しく解説","summary":"Anthropic チームが投稿で詳しく説明した、今年8月に2週間のスプリントで claude.ai およびデスクトップ端末のコア体験を約3倍高速化したプロセス：ユーザー活動の95％をカバーする4つのジャーニーに注力し、75パーセンタイルの初回画面入力時間を3.1秒から0.55秒に短縮、3000以上の変更を統合しながらも、ユーザーに影響を与える事故は一度も発生させなかった。","category":"業界動向","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Anthropic チームがClaudeを使って2週間でclaude.aiを約3倍高速化する方法を詳しく解説 - Aioga AIニュース","description":"Anthropic チームが投稿で詳しく説明した、今年8月に2週間のスプリントで claude.ai およびデスクトップ端末のコア体験を約3倍高速化したプロセス：ユーザー活動の95％をカバーする4つのジャーニーに注力し、75パーセンタイルの初回画面入力時間を3.1秒から0.55秒に短縮、3000以上の変更を統合しながらも、ユーザーに影響を与える事故は一度も発...","url":"https://www.aioga.com/ja/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:04:07.181Z"},"ko":{"title":"Anthropic 팀이 Claude를 사용해 두 주 안에 claude.ai 속도를 약 3배로 높이는 방법을 자세히 설명","summary":"Anthropic 팀이 이번 8월 두 주간의 스프린트를 통해 claude.ai와 데스크톱 핵심 경험 속도를 약 3배 향상시킨 과정을 자세히 설명한 글: 사용자 활동의 95%를 커버하는 네 가지 여정에 집중하고, 75분위 첫 화면 입력 시간을 3.1초에서 0.55초로 단축했으며, 3천 개 이상의 변경을 병합했음에도 사용자에게 영향을 미친 사고는 단 한 건도 없었다.","category":"업계 동향","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Anthropic 팀이 Claude를 사용해 두 주 안에 claude.ai 속도를 약 3배로 높이는 방법을 자세히 설명 - Aioga AI 뉴스","description":"Anthropic 팀이 이번 8월 두 주간의 스프린트를 통해 claude.ai와 데스크톱 핵심 경험 속도를 약 3배 향상시킨 과정을 자세히 설명한 글: 사용자 활동의 95%를 커버하는 네 가지 여정에 집중하고, 75분위 첫 화면 입력 시간을 3.1초에서 0.55초로 단축했으며, 3천 개 이상의 변경을 병합했음에도 사용자에...","url":"https://www.aioga.com/ko/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:04:54.275Z"},"es":{"title":"El equipo de Anthropic explica en detalle cómo usar Claude para acelerar claude.ai aproximadamente 3 veces en dos semanas","summary":"El equipo de Anthropic publicó un artículo detallando el proceso de acelerar aproximadamente 3 veces claude.ai y la experiencia central en el escritorio mediante un sprint de dos semanas en agosto de este año: se enfocaron en cuatro recorridos que cubren el 95% de las actividades de los usuarios, el tiempo de entrada en la primera pantalla del percentil 75 se redujo de 3,1 segundos a 0,55 segundos, se fusionaron más de tres mil cambios y no hubo ningún incidente que afectara a los usuarios.","category":"Industria","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"El equipo de Anthropic explica en detalle cómo usar Claude para acelerar claude.ai aproximadamente 3 veces en dos semanas - Aioga Noticias de IA","description":"El equipo de Anthropic publicó un artículo detallando el proceso de acelerar aproximadamente 3 veces claude.ai y la experiencia central en el escritorio mediante un sprint de dos s...","url":"https://www.aioga.com/es/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:04:50.918Z"},"fr":{"title":"L'équipe d'Anthropic explique en détail comment utiliser Claude pour accélérer claude.ai d'environ 3 fois en deux semaines","summary":"L'équipe d'Anthropic a publié un article détaillant le processus par lequel, en août de cette année, ils ont accéléré claude.ai et l'expérience principale sur le bureau d'environ 3 fois grâce à une course de deux semaines : en se concentrant sur quatre parcours représentant 95 % de l'activité des utilisateurs, le temps d'entrée de l'écran principal au 75e percentile est passé de 3,1 secondes à 0,55 seconde, en combinant plus de trois mille modifications sans aucun incident affectant les utilisateurs.","category":"Industrie","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"L'équipe d'Anthropic explique en détail comment utiliser Claude pour accélérer claude.ai d'environ 3 fois en deux semaines - Aioga Actualités IA","description":"L'équipe d'Anthropic a publié un article détaillant le processus par lequel, en août de cette année, ils ont accéléré claude.ai et l'expérience principale sur le bureau d'environ 3...","url":"https://www.aioga.com/fr/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:05:29.051Z"},"de":{"title":"Das Anthropic-Team erklärt detailliert, wie man mit Claude claude.ai in zwei Wochen etwa dreimal schneller machen kann","summary":"Das Anthropic-Team veröffentlichte einen Artikel, der den Prozess im August dieses Jahres detailliert beschreibt, bei dem durch zweiwöchige Sprints Claude.ai und die Kernerfahrung auf dem Desktop etwa dreimal schneller gemacht wurden: Fokus auf vier Reiseabläufe, die 95 % der Benutzeraktivitäten abdecken, die Eingabezeit auf der ersten Seite im 75. Perzentil sank von 3,1 Sekunden auf 0,55 Sekunden, über dreitausend Änderungen wurden zusammengeführt, ohne dass es zu einem einzelnen benutzerseitigen Vorfall kam.","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Das Anthropic-Team erklärt detailliert, wie man mit Claude claude.ai in zwei Wochen etwa dreimal schneller machen kann - Aioga KI-News","description":"Das Anthropic-Team veröffentlichte einen Artikel, der den Prozess im August dieses Jahres detailliert beschreibt, bei dem durch zweiwöchige Sprints Claude.ai und die Kernerfahrung...","url":"https://www.aioga.com/de/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:05:29.384Z"},"pt-BR":{"title":"A equipe da Anthropic explica detalhadamente como usar o Claude para acelerar o claude.ai em cerca de 3 vezes em duas semanas","summary":"A equipe da Anthropic publicou um artigo detalhando o processo deste ano, em agosto, no qual, através de uma sprint de duas semanas, aceleraram o claude.ai e a experiência principal no desktop em cerca de 3 vezes: concentrando-se em quatro jornadas que cobrem 95% das atividades dos usuários, o tempo de entrada na tela inicial no percentil 75 caiu de 3,1 segundos para 0,55 segundos, combinando mais de três mil alterações sem nenhum incidente voltado para o usuário.","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"A equipe da Anthropic explica detalhadamente como usar o Claude para acelerar o claude.ai em cerca de 3 vezes em duas semanas - Aioga Notícias de IA","description":"A equipe da Anthropic publicou um artigo detalhando o processo deste ano, em agosto, no qual, através de uma sprint de duas semanas, aceleraram o claude.ai e a experiência principa...","url":"https://www.aioga.com/pt-BR/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:06:02.957Z"},"ru":{"title":"Команда Anthropic подробно рассказывает, как с помощью Claude за две недели ускорить claude.ai примерно в 3 раза","summary":"Команда Anthropic опубликовала статью, подробно объясняющую процесс ускорения claude.ai и ключевого опыта на десктопе примерно в 3 раза за двухнедельный спринт в августе этого года: сосредоточившись на четырёх путешествиях пользователей, охватывающих 95% пользовательской активности, время до возможности ввода на первом экране на 75-м процентиле снизилось с 3,1 секунд до 0,55 секунд, было объединено более трёх тысяч изменений без единого инцидента для пользователей.","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Команда Anthropic подробно рассказывает, как с помощью Claude за две недели ускорить claude.ai примерно в 3 раза - Aioga Новости ИИ","description":"Команда Anthropic опубликовала статью, подробно объясняющую процесс ускорения claude.ai и ключевого опыта на десктопе примерно в 3 раза за двухнедельный спринт в августе этого года...","url":"https://www.aioga.com/ru/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:06:13.281Z"},"ar":{"title":"فريق Anthropic يشرح بالتفصيل كيفية استخدام Claude لزيادة سرعة claude.ai حوالي 3 أضعاف خلال أسبوعين","summary":"نشرت فريق Anthropic مقالة توضح بالتفصيل عملية زيادة سرعة claude.ai وتجربة سطح المكتب الأساسية بنحو 3 أضعاف في أغسطس هذا العام عبر فترة اسبوعين مكثفة: تركيز على أربع رحلات تغطي 95٪ من نشاط المستخدمين، انخفاض وقت الإدخال للشاشة الأولى في الربيع الـ 75 من 3.1 ثانية إلى 0.55 ثانية، ودمج أكثر من ثلاثة آلاف تغيير دون حادث واحد يواجه المستخدمين.","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"فريق Anthropic يشرح بالتفصيل كيفية استخدام Claude لزيادة سرعة claude.ai حوالي 3 أضعاف خلال أسبوعين - Aioga أخبار الذكاء الاصطناعي","description":"نشرت فريق Anthropic مقالة توضح بالتفصيل عملية زيادة سرعة claude.ai وتجربة سطح المكتب الأساسية بنحو 3 أضعاف في أغسطس هذا العام عبر فترة اسبوعين مكثفة: تركيز على أربع رحلات تغطي 95٪...","url":"https://www.aioga.com/ar/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:06:49.461Z"},"hi":{"title":"Anthropic टीम ने विस्तार से बताया कि कैसे Claude का उपयोग करके दो हफ़्तों में claude.ai की गति लगभग 3 गुना बढ़ाई जा सकती है","summary":"Anthropic टीम ने इस साल अगस्त में दो सप्ताह की तेजी से विकास प्रक्रिया के द्वारा claude.ai और डेस्कटॉप मुख्य अनुभव को लगभग 3 गुना तेज़ करने की प्रक्रिया का पूरा विवरण प्रकाशित किया: 95% उपयोगकर्ता गतिविधियों को कवर करने वाली चार यात्राओं पर ध्यान केंद्रित किया, 75वें पर्सेंटाइल पर होम स्क्रीन इनपुट समय 3.1 सेकंड से घटाकर 0.55 सेकंड किया, 3000 से अधिक बदलावों को मर्ज किया और उपयोगकर्ताओं के लिए कोई भी दुर्घटना नहीं हुई।","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Anthropic टीम ने विस्तार से बताया कि कैसे Claude का उपयोग करके दो हफ़्तों में claude.ai की गति लगभग 3 गुना बढ़ाई जा सकती है - Aioga AI समाचार","description":"Anthropic टीम ने इस साल अगस्त में दो सप्ताह की तेजी से विकास प्रक्रिया के द्वारा claude.ai और डेस्कटॉप मुख्य अनुभव को लगभग 3 गुना तेज़ करने की प्रक्रिया का पूरा विवरण प्रकाशित किया...","url":"https://www.aioga.com/hi/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:06:54.778Z"},"it":{"title":"Il team di Anthropic spiega in dettaglio come utilizzare Claude per accelerare claude.ai di circa 3 volte in due settimane","summary":"Il team di Anthropic ha pubblicato un articolo che spiega in dettaglio il processo di accelerazione di circa 3 volte di claude.ai e dell'esperienza principale su desktop avvenuto ad agosto di quest'anno mediante sprint di due settimane: concentrandosi su quattro percorsi che coprono il 95% delle attività degli utenti, il tempo di input della schermata iniziale al 75° percentile è passato da 3,1 secondi a 0,55 secondi, unendo oltre tremila modifiche senza alcun incidente rivolto agli utenti.","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Il team di Anthropic spiega in dettaglio come utilizzare Claude per accelerare claude.ai di circa 3 volte in due settimane - Aioga Notizie IA","description":"Il team di Anthropic ha pubblicato un articolo che spiega in dettaglio il processo di accelerazione di circa 3 volte di claude.ai e dell'esperienza principale su desktop avvenuto a...","url":"https://www.aioga.com/it/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:07:31.918Z"},"nl":{"title":"Het Anthropic-team legt uit hoe ze Claude hebben gebruikt om claude.ai in twee weken ongeveer 3 keer sneller te maken","summary":"Het Anthropic-team publiceerde een artikel waarin ze het proces van augustus dit jaar uitlegden, waarin ze door middel van een twee weken durende sprint claude.ai en de kernervaring op desktop met ongeveer 3 keer versneld hebben: ze concentreerden zich op vier trajecten die 95% van de gebruikersactiviteiten dekken, de invoertijd van het eerste scherm op het 75e percentiel daalde van 3,1 seconden naar 0,55 seconden, meer dan drieduizend wijzigingen werden samengevoegd zonder één gebruikersincident.","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Het Anthropic-team legt uit hoe ze Claude hebben gebruikt om claude.ai in twee weken ongeveer 3 keer sneller te maken - Aioga AI-nieuws","description":"Het Anthropic-team publiceerde een artikel waarin ze het proces van augustus dit jaar uitlegden, waarin ze door middel van een twee weken durende sprint claude.ai en de kernervarin...","url":"https://www.aioga.com/nl/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:07:32.047Z"},"tr":{"title":"Anthropic ekibi, Claude'u kullanarak claude.ai'yi iki hafta içinde yaklaşık 3 kat hızlandırmanın ayrıntılarını açıkladı","summary":"Anthropic ekibi, bu yılın Ağustos ayında iki haftalık bir sprint ile claude.ai ve masaüstü ana deneyimini yaklaşık 3 kat hızlandırma sürecini ayrıntılı şekilde açıkladı: Kullanıcı aktivitelerinin %95'ini kapsayan dört yolculuğa odaklanmak, 75. yüzdelik ilk ekran giriş süresini 3,1 saniyeden 0,55 saniyeye düşürmek, ve 3000'den fazla değişikliği birleştirip kullanıcıya yönelik hiç kaza yaşamamak.","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Anthropic ekibi, Claude'u kullanarak claude.ai'yi iki hafta içinde yaklaşık 3 kat hızlandırmanın ayrıntılarını açıkladı - Aioga AI Haberleri","description":"Anthropic ekibi, bu yılın Ağustos ayında iki haftalık bir sprint ile claude.ai ve masaüstü ana deneyimini yaklaşık 3 kat hızlandırma sürecini ayrıntılı şekilde açıkladı: Kullanıcı...","url":"https://www.aioga.com/tr/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:08:07.458Z"},"vi":{"title":"Nhóm Anthropic giải thích chi tiết cách sử dụng Claude để tăng tốc claude.ai khoảng 3 lần trong hai tuần","summary":"Nhóm Anthropic đã đăng bài phân tích chi tiết về quá trình tăng tốc trải nghiệm cốt lõi của claude.ai và phiên bản desktop khoảng 3 lần chỉ trong hai tuần vào tháng 8 năm nay: tập trung vào bốn hành trình bao phủ 95% hoạt động người dùng, thời gian nhập trên màn hình chính đạt phân vị 75 giảm từ 3,1 giây xuống 0,55 giây, hợp nhất hơn ba nghìn thay đổi mà không xảy ra bất kỳ sự cố nào đối với người dùng.","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Nhóm Anthropic giải thích chi tiết cách sử dụng Claude để tăng tốc claude.ai khoảng 3 lần trong hai tuần - Tin tức AI Aioga","description":"Nhóm Anthropic đã đăng bài phân tích chi tiết về quá trình tăng tốc trải nghiệm cốt lõi của claude.ai và phiên bản desktop khoảng 3 lần chỉ trong hai tuần vào tháng 8 năm nay: tập...","url":"https://www.aioga.com/vi/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:08:09.201Z"},"id":{"title":"Tim Anthropic menjelaskan secara rinci bagaimana menggunakan Claude untuk mempercepat claude.ai sekitar 3 kali lipat dalam dua minggu","summary":"Tim Anthropic menerbitkan artikel yang menjelaskan secara rinci proses bagaimana pada bulan Agustus tahun ini mereka melalui sprint dua minggu berhasil mempercepat claude.ai dan pengalaman inti di desktop sekitar 3 kali: fokus pada empat perjalanan yang mencakup 95% aktivitas pengguna, waktu input layar pertama pada persentil ke-75 turun dari 3,1 detik menjadi 0,55 detik, menggabungkan lebih dari tiga ribu perubahan tanpa satu pun insiden yang berdampak pada pengguna.","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Tim Anthropic menjelaskan secara rinci bagaimana menggunakan Claude untuk mempercepat claude.ai sekitar 3 kali lipat dalam dua minggu - Berita AI Aioga","description":"Tim Anthropic menerbitkan artikel yang menjelaskan secara rinci proses bagaimana pada bulan Agustus tahun ini mereka melalui sprint dua minggu berhasil mempercepat claude.ai dan pe...","url":"https://www.aioga.com/id/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:08:47.872Z"},"th":{"title":"ทีมงาน Anthropic อธิบายโดยละเอียดเกี่ยวกับวิธีการใช้ Claude เพื่อเพิ่มความเร็ว claude.ai ประมาณ 3 เท่าในสองสัปดาห์","summary":"ทีมงาน Anthropic เผยบทความอธิบายละเอียดเกี่ยวกับกระบวนการที่ช่วยให้ claude.ai และประสบการณ์หลักบนเดสก์ท็อปเร็วขึ้นประมาณ 3 เท่าในช่วงการทำงานแบบสองสัปดาห์เมื่อเดือนสิงหาคมปีนี้: มุ่งเน้นไปที่สี่เส้นทางการใช้งานที่ครอบคลุมกิจกรรมของผู้ใช้ 95% เวลาสำหรับการพิมพ์บนหน้าจอแรกที่เปอร์เซ็นไทล์ 75 ลดลงจาก 3.1 วินาทีเป็น 0.55 วินาที รวมการเปลี่ยนแปลงมากกว่า 3,000 ครั้งโดยไม่มีเหตุการณ์ใด ๆ ที่กระทบต่อผู้ใช้","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"ทีมงาน Anthropic อธิบายโดยละเอียดเกี่ยวกับวิธีการใช้ Claude เพื่อเพิ่มความเร็ว claude.ai ประมาณ 3 เท่าในสองสัปดาห์ - ข่าว AI Aioga","description":"ทีมงาน Anthropic เผยบทความอธิบายละเอียดเกี่ยวกับกระบวนการที่ช่วยให้ claude.ai และประสบการณ์หลักบนเดสก์ท็อปเร็วขึ้นประมาณ 3 เท่าในช่วงการทำงานแบบสองสัปดาห์เมื่อเดือนสิงหาคมปีนี้: มุ...","url":"https://www.aioga.com/th/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:08:58.060Z"},"pl":{"title":"Zespół Anthropic szczegółowo wyjaśnia, jak za pomocą Claude w ciągu dwóch tygodni przyspieszyć claude.ai około 3 razy","summary":"Zespół Anthropic opublikował artykuł szczegółowo wyjaśniający proces przyspieszenia claude.ai i kluczowego doświadczenia na pulpicie około 3-krotnie w ciągu dwóch tygodni sprintu w sierpniu tego roku: skoncentrowano się na czterech ścieżkach obejmujących 95% aktywności użytkowników, czas wpisywania na pierwszym ekranie dla 75 percentyla spadł z 3,1 sekundy do 0,55 sekundy, połączono ponad trzy tysiące zmian bez ani jednego incydentu widocznego dla użytkownika.","category":"行业动态","source":"Hacker News：AI 热帖","aggregationSource":"Hacker News：AI 热帖","pageTitle":"Zespół Anthropic szczegółowo wyjaśnia, jak za pomocą Claude w ciągu dwóch tygodni przyspieszyć claude.ai około 3 razy - Aioga Wiadomości AI","description":"Zespół Anthropic opublikował artykuł szczegółowo wyjaśniający proces przyspieszenia claude.ai i kluczowego doświadczenia na pulpicie około 3-krotnie w ciągu dwóch tygodni sprintu w...","url":"https://www.aioga.com/pl/news/cmuekny10060lroyn21gctbfw/","contentTranslated":true,"sourceHash":"d5c7c6523cf9b537","translatedAt":"2026-09-23T21:09:41.267Z"}},"evidenceTier":"verified-news","reviewStatus":"editorial-selected","indexable":true,"editorialCover":""}}