--- title: Anthropic used Claude to make Claude 3x faster, and... description: The Anthropic team behind claude.ai shared a workflow where they used Claude itself to profile, debug, and prioritize fixes, cutting load times by 3x in two... canonical: https://daily.dev/posts/anthropic-used-claude-to-make-claude-3x-faster-and-published-the-playbook-mdawigk2l twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Anthropic used Claude to make Claude 3x faster, and published the playbook | daily.dev og:description: The Anthropic team behind claude.ai shared a workflow where they used Claude itself to profile, debug, and prioritize fixes, cutting load times by 3x in two... og:url: https://daily.dev/posts/anthropic-used-claude-to-make-claude-3x-faster-and-published-the-playbook-mdawigk2l og:image: https://api.daily.dev/og/posts/MDaWIgK2l.png og:image:alt: Anthropic used Claude to make Claude 3x faster, and published the playbook og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Anthropic used Claude to make Claude 3x faster, and published the playbook **[Trends](https://daily.dev/sources/trends)** · 2 min read · 2 upvotes · 0 comments ## Summary The Anthropic team behind claude.ai shared a workflow where they used Claude itself to profile, debug, and prioritize fixes, cutting load times by 3x in two weeks. The thread includes actual prompts and methods rather than vague claims, which drew attention from developers like Addy Osmani. The core loop: feed Claude performance data, let it flag bottlenecks, act on the output, repeat. Anthropic has an obvious incentive to make Claude look good in this story, and it remains unclear whether the workflow generalizes beyond a team that already deeply knows its own codebase. ## Content The Anthropic team behind claude.ai dropped a thread this week claiming they cut load times by 3x in two weeks, and the method is the part worth paying attention to: they used Claude itself to measure, debug, and drive the performance work. The thread includes the actual prompts and methods, which is what got Addy Osmani and others sharing it. It's not a vague "AI helped us ship faster" post. It's a concrete workflow where Claude is doing the grunt work of profiling and diagnosing, not just generating boilerplate. The basic loop: feed Claude performance data, let it identify bottlenecks, use its output to prioritize fixes, repeat. The team says the speed of iteration was the point. Two weeks is fast for a 3x improvement on a production web app, and the argument is that the AI-in-the-loop debugging cycle is what made that pace possible. There's an obvious caveat here: Anthropic has every incentive to publish a story where Claude looks good. Using your own product to improve your own product and then writing about it is a marketing move as much as an engineering one. That doesn't make the methods wrong, but it's worth keeping in mind when evaluating the claims. What's actually useful is the prompts. If the workflow is reproducible outside of Anthropic's internal setup, developers can test it on their own apps. That's the real question the thread leaves open: does this generalize, or does it work because the team already knows their codebase inside out and Claude is just a fast rubber duck? Either way, the playbook is public now. Worth a read if you're doing any frontend performance work. ## Questions this post answers ### How did Anthropic use Claude to make claude.ai load 3x faster? The team fed Claude performance data, had it identify bottlenecks, then used its output to prioritize and drive fixes in an iterative loop. This AI-in-the-loop debugging cycle let them achieve a 3x load time improvement on the production claude.ai web app within two weeks, with Claude doing profiling and diagnosis rather than just generating code. _Developers testing AI-driven performance workflows can compare notes and results on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 3 discussions and 198 comments across x (as of 2026-09-24). **TL;DR:** Reaction is a mix of genuine appreciation for the transparency of sharing real prompts/methods, jokes about the recursive irony of using Claude to fix Claude, and skepticism that the underlying app/codebase quality (and cost/usage limits) justify the celebration. **Sentiment:** 35% positive · 40% mixed · 25% skeptical **The case for** - Several developers praised that the post included actual prompts and methodology instead of just marketing claims. - Some found the 'measure, then let Claude propose fixes' loop genuinely reproducible and plan to try it on their own projects. - A few called it the best example of dogfooding they've seen, using the model to optimize its own product. **The pushback** - Multiple commenters joked that being 3x faster just means users hit usage/token limits sooner. - Several questioned whether the baseline app was simply badly engineered (e.g. thousands of React hooks firing on keypress, regex-based syntax highlighting) and this was just basic cleanup rather than an AI breakthrough. - Some pointed out the win applies to the website, not inference speed, causing confusion/arguments in replies. - Doubts were raised about whether p50 vs p95/tail latency was actually improved, and whether the gains reflect real UX benefit versus a nicer dashboard metric. **By community** - x (mixed): Reactions swing between praise for the transparent, prompt-level methodology and jokes/skepticism about token limits, costs, and whether the underlying code was simply poorly built to begin with. **Hottest debate:** Whether the reported speedup reflects genuine engineering skill and a poorly-optimized baseline being fixed, or is an inflated, self-serving story given Anthropic's incentive to promote Claude. **Open questions** - How much of the 3x gain came from the model's suggestions versus basic code cleanup/deleting bloat? - Did the improvement hold at tail latency (p95) or only at median (p50)? - Does this workflow generalize to other codebases, or does it depend on already deeply understanding your own system? **Highlights** > @ClaudeDevs How is this considered impressive? You took a horrible baseline (3s++ cold load for a chat app) and made marginal improvements (e.g. static composers) to code that sounds mediocre at best and is likely horrendous under the hood (6900 hooks in the typing path is one clue, among > — [alynchfc on x · 4 points](https://x.com/alynchfc/status/2102907547183177737) > @ClaudeDevs This is such a clusterfuck of bad decisions and nobody calling this out 6900 react hooks trigger on key press? freezing the thread on stream? syntax highlighting via regex?!? How is this acceptable? > — [g2sans on x · 1 points, 3 comments](https://x.com/g2sans/status/2102899052866109485) > @ClaudeDevs The em dash bug is my favourite: one character sent the markdown string down V8’s slower two-byte regex path. A 20-line fix took first code-block highlighting from about a second to 350ms. Profilers find the best plot twists. > — [tensorquay on x · 1 points](https://x.com/tensorquay/status/2102847635547959367) > @ClaudeDevs The lasting win from a performance sprint should be a regression check. Make the next slowdown easy to catch before users have to report it. > — [jeffbruchado on x](https://x.com/jeffbruchado/status/2102840032352768077) > @ClaudeDevs the 3x is nice but the method is the headline. using claude to debug claude, measuring performance with the thing being measured. recursion used to be a sci-fi plot, now its a sprint task. who benchmarks the benchmarker? > — [DataDiscovered on x](https://x.com/DataDiscovered/status/2102850826083385446) **Source threads** - [x](https://x.com/trq212/status/2102841583158878500) · 0 points · 0 comments - [x](https://x.com/addyosmani/status/2102840978428305513) · 0 points · 0 comments - [x](https://x.com/ClaudeDevs/status/2102839691154427983) · 0 points · 198 comments ## Similar posts on daily.dev - [How Anthropic uses Claude in Marketing](https://daily.dev/posts/how-anthropic-uses-claude-in-marketing-mnichyg2h) · Claude · 91 upvotes · 4 comments - [Anthropic doubles Claude usage outside peak hours — but it won’t last forever](https://daily.dev/posts/anthropic-doubles-claude-usage-outside-peak-hours-but-it-won-t-last-forever-ylf6ychkw) · The New Stack · 2 upvotes · 0 comments --- Tags: [#performance](https://daily.dev/tags/performance), [#ai-coding](https://daily.dev/tags/ai-coding), [#claude](https://daily.dev/tags/claude), [#anthropic](https://daily.dev/tags/anthropic) [View this post on daily.dev](https://daily.dev/posts/anthropic-used-claude-to-make-claude-3x-faster-and-published-the-playbook-mdawigk2l) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Anthropic used Claude to make Claude 3x faster, and published the playbook","url":"https://daily.dev/posts/anthropic-used-claude-to-make-claude-3x-faster-and-published-the-playbook-mdawigk2l","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/anthropic-used-claude-to-make-claude-3x-faster-and-published-the-playbook-mdawigk2l"},"datePublished":"2026-09-23T19:25:04.112Z","dateModified":"2026-09-24T05:56:17.128Z","description":"The Anthropic team behind claude.ai shared a workflow where they used Claude itself to profile, debug, and prioritize fixes, cutting load times by 3x in two...","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/anthropic-used-claude-to-make-claude-3x-faster-and-published-the-playbook-mdawigk2l","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"performance,ai-coding,claude,anthropic","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"Anthropic used Claude to make Claude 3x faster, and published the playbook"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/anthropic-used-claude-to-make-claude-3x-faster-and-published-the-playbook-mdawigk2l#faq","mainEntity":[{"@type":"Question","name":"How did Anthropic use Claude to make claude.ai load 3x faster?","acceptedAnswer":{"@type":"Answer","text":"The team fed Claude performance data, had it identify bottlenecks, then used its output to prioritize and drive fixes in an iterative loop. This AI-in-the-loop debugging cycle let them achieve a 3x load time improvement on the production claude.ai web app within two weeks, with Claude doing profiling and diagnosis rather than just generating code. Developers testing AI-driven performance workflows can compare notes and results on daily.dev."}}]} ```