{"slug": "claude-docs-launch-60-server-errors-across-billions-of-requests", "title": "Claude Docs launch: 60 server errors across billions of requests", "summary": "Anthropic's Claude Docs handled over 4.6 million documents and 2 billion requests with only 60 server errors after launching on September 16, 2026, running on about 50 memory-optimized pods without a rollback. Two weeks before launch, an Anthropic engineer left a Claude Code session running against a staging cluster mirroring production at over a million synthetic docs a day, with roughly 11 sub-agents active in the busiest week and 30 at peak, breaking core components on purpose and proposing fixes. The engineer said the approach worked because frontier models can now pursue one goal for weeks, crediting an internal research model roughly comparable to Claude Fable 5.2.", "body_md": "Claude Docs is a doc editor that you, your teammates, and Claude all work in together. The interesting part is that a Claude session anywhere, from chat to Claude Code, can open the same doc and read, write, and comment in it live, right next to you. We released Docs on September 16, 2026. This post is about the service that runs underneath it.\n\nThree weeks before Claude Docs opened to the public, it was running as a proof of concept on a single node, used by a few thousand people inside Anthropic. We were racing toward an ambitious launch date, and our own estimate said we'd see millions of docs within the first few days.\n\nWe launched on September 16. By September 22, people had created over 4.6 million docs. Across 2 billion requests we saw only 60 server errors, and we never had to roll back. The whole Docs service ran on about 50 memory-optimized pods.\n\nTwo decisions got us there:\n\nBut honestly, this only worked because our frontier models have gotten much better recently (I used an internal research model roughly comparable to Claude Fable 5.2). In our experience, they can now stick with one goal through a run like this for weeks, which is what helped us hit launch scale in a tight timeline.\n\nA few thousand internal users on one machine told us nothing about whether the service would survive launch, and we obviously didn't have launch traffic to test with.\n\nWe wanted real evidence to decide how fast we could roll out. So on September 4, about two weeks before launch, I had Claude drive a staging cluster that mirrored production at over a million synthetic docs a day and break core components on purpose. And then... I just left it running (I've included the full cost at the end of this article).\n\nMy setup was pretty simple. The loop was a session in Claude Code that I left running: pick a scenario, run the harness, read the telemetry, propose a fix, ship it to staging, repeat. That session acted as a coordinator and handed most of the work to sub-agents. In the busiest week, that usually meant about 11 sub-agents running at a time, and 30 at the peak. The coordinator also talked to our internal variant of Claude Tag (beta) via the Slack MCP to learn from other infrastructure teams at Anthropic who had scaled similar services. Their agents checked our scaling approach for pitfalls they’d already hit.\n\nThe fun part was picking scenarios I thought were paranoid:\n\nEach iteration looked the same. Our testing harness selected a feature like document creation and pushed creation to saturation. Claude then read telemetry using the Datadog MCP and came back with a short menu of options and a recommendation.\n\nIf the recommendation was easy to undo, Claude wrote and deployed the change to staging and reran the test. We had telemetry on every operation the service ran. Every database query, for example, reported its name and how many rows it scanned versus how many it returned, so Claude could spot a bad query and rewrite it.\n\nBefore the loop started, I told Claude which decisions needed my approval; basically anything hard to undo, like changing the database schema or key layout, or changing how a doc is stored and materialized.\n\nWhen Claude ran into one of those, it stopped and asked me, usually with a recommendation that I ended up approving. Schema changes were the classic case where I wanted to be consulted: there are usually a few reasonable answers, and changing directions is often painful.\n\nAnd during the whole testing process, Claude had to conform to two invariants:\n\nMy colleague Felix Rieseberg recently wrote [a tutorial on being more ambitious with models](https://felixrieseberg.com/how-to-be-more-ambitious-a-tutorial-for-humans-with-models/). His main idea is to raise the \"altitude\" you prompt Claude at: ask for the *outcome* you want, not the individual task.\n\nSo instead of asking Claude to fix individual crashes, you ask it to make your app crash-free. (Our own apps aren't all the way there yet, but this is how we're making real progress.) To work at that altitude, though, you have to set up your code so Claude can pull crash reports on its own and run a harness that reproduces each crash and checks the fix.\n\nFor us, the outcome was a service that could handle launch. To get there, we did two things to make the service as legible to Claude as we could:\n\nEvery iteration, Claude read the telemetry and added whatever it thought was missing. Writing telemetry by hand is tedious, so in practice most telemetry would go in after something broke. Now Claude just writes it as it goes.\n\nMemory is the best example.\n\nEarly in the harness run, one large doc was using about 3 GB on its pod (not a fun number to see). Claude’s first stab at fixing it was to shrink one of our caches so our server would evict docs sooner. The next harness run showed the memory was actually in a *separate* cache inside one of our dependencies, so Claude dropped the change.\n\nFrom there Claude kept narrowing it down with finer-grained memory profiling. Eventually it found that every keystroke was copying the doc into several different data structures, adding a few MB each time. Claude switched those structures to copy-on-write, so they now share one reference instead of each holding its own copy.\n\nAfter about three days of iterating, that doc topped out around 180 MB, and our estimate of what we’d need for launch dropped from more than 500 memory-optimized pods to about 50.\n\nWe were picky about the service’s runtime and dependencies. We wanted any number Claude measured to trace back to code we wrote.\n\nThis is because our models hill climb best with a clean signal. We expected a garbage collector to add noise, because memory and latency shift on the collector’s schedule.\n\nTwo choices helped the most.\n\nRust is hard for people to write, and a few years ago that would have weighed heavily on the choice. Claude writes it and most other languages fine, so that matters a lot less now.\n\nThe service held at launch scale. The first million docs showed up in about 30 hours, and we were past 4.6 million by September 22. We rolled prod 629 times in September and never once rolled back.\n\nLaunch week felt weirdly familiar. Claude is great at imagining ways a service can fail and had planned for most of them well before launch. Seven of the ten scenarios we’d rehearsed on staging actually happened in prod. By then the harness had already put staging through most of it, and Claude had shipped the fixes behind runtime flags. When part of the infra ran into pressure, we usually just flipped a flag.\n\nA few months ago this probably wouldn’t have worked. Older models were fine at reading telemetry through an MCP server, but ten or twenty iterations into a long loop, they’d start forgetting what we were optimizing for, and I’d end up restating the goal. Other Anthropic employees have called this [goal drift](https://claude.dev/blog/a-harness-for-every-task-dynamic-workflows-in-claude-code/#why-dynamic-workflows).\n\nOur newer frontier models are built for long-running work. Claude held all of our invariants across many compactions, and I never had to repeat them.\n\nWe're not the only ones doing this, either. Vercel [made Turborepo up to 96% faster](https://vercel.com/blog/making-turborepo-ninety-six-percent-faster-with-agents-sandboxes-and-humans) with a supervised agent loop a lot like ours.\n\nAt Claude Fable 5.2's API list price (as of November 2026), I used roughly $125,000 of inference from the end of August to launch day. A couple of years ago, work on this scale I'd guess would have taken a team of engineers about six months.\n\nMost of that went to reading telemetry, writing proposals, generating fixes, and reviewing code.\n\nA couple of things about the cost:\n\nThis works for much smaller services too, and it should cost a lot less.\n\nTo get a rough number, I priced out a classic small team stack: a couple of reasonably sized Kubernetes nodes and a Postgres instance. Then I went through the changes I landed between the end of August and launch day and removed the work that setup wouldn’t need:\n\nAbout a third of the changes were left.\n\nScaling our spend down to that third and taking out the mistakes I made while I was still figuring out the setup, I’d put a two week push like this one at $3,000 to $10,000 at the same list price (I haven’t actually run this, so it’s only a rough sense).\n\nThe steps below assume one engineer with Claude Code, a staging environment, and an observability provider. The coordinator and sub-agents helped speed up the process, but if you find they’re burning too many tokens per hour, going back to a single agent should also work quite well. If you want to try it:\n\n*Huge thanks to the Docs team, who landed thousands of editor fixes in the same three weeks, and to the platform and storage teams who built everything this sits on.*\n\n**Further reading.** [How we made claude.ai 3x faster in two weeks](https://claude.dev/blog/how-we-made-claude-ai-faster/) (claude.dev) · [A harness for every task: dynamic workflows in Claude Code](https://claude.dev/blog/a-harness-for-every-task-dynamic-workflows-in-claude-code/) (claude.dev), on goal drift.", "url": "https://wpnews.pro/news/claude-docs-launch-60-server-errors-across-billions-of-requests", "canonical_source": "https://claude.dev/blog/chaos-loop-claude-docs-launch/", "published_at": "2026-11-20 14:27:34+00:00", "updated_at": "2026-11-20 14:46:31.000846+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-infrastructure", "large-language-models", "mlops"], "entities": ["Anthropic", "Claude Docs", "Claude Code", "Claude Fable 5.2", "Datadog", "Claude Tag", "Felix Rieseberg", "Slack"], "also_reported_by": [], "takeaway": "Running a Claude Code coordinator that hands work to roughly 11 sub-agents (30 at peak) to inject failures, read Datadog telemetry, and auto-deploy reversible fixes to staging let Claude Docs reach 4.6 million docs and 2 billion requests with only 60 server errors, while schema and storage changes w", "recommendation": {"advice": "Gate hard-to-undo changes like schema or key layout behind human approval, and let the agent auto-deploy only reversible fixes it can rerun and verify against telemetry.", "playbook": "agent-architecture", "playbook_url": "https://wpnews.pro/playbooks/agent-architecture"}, "alternates": {"html": "https://wpnews.pro/news/claude-docs-launch-60-server-errors-across-billions-of-requests", "markdown": "https://wpnews.pro/news/claude-docs-launch-60-server-errors-across-billions-of-requests.md", "text": "https://wpnews.pro/news/claude-docs-launch-60-server-errors-across-billions-of-requests.txt", "jsonld": "https://wpnews.pro/news/claude-docs-launch-60-server-errors-across-billions-of-requests.jsonld"}}