---
title: 'In the News: November 18, 2026 (Morning)'
description: 'Anthropic says a two-week Claude loop on staging readied Claude Docs for launch; Jesse Vincent gives agents a dedicated toolsmith.'
canonical_url: 'https://darkfactory.dev/news/2026-11-18-morning'
markdown_url: 'https://darkfactory.dev/news/2026-11-18-morning.md'
collection: news
date_published: '2026-11-18T07:05:00-05:00'
date_modified: '2026-11-18T07:05:00-05:00'
---

# In the News: November 18, 2026 (Morning)


Anthropic has published how it load-tested Claude Docs before launch: one engineer left a Claude Code session running against staging for two weeks, breaking the service and fixing it, with a person approving only the changes that were hard to undo. The figures are the team's own. Below it, Jesse Vincent has turned the helper scripts agents write in passing into a separate job with its own subagent.

## 1. Anthropic ran a two-week Claude loop against staging before the Claude Docs launch

**[Using a "chaos loop" to harden the Claude Docs launch](https://claude.dev/blog/chaos-loop-claude-docs-launch/)** · Vivek Nair, Anthropic · claude.dev, November 17, 2026

From September 4, about two weeks before launch, Nair had a Claude Code session drive a staging cluster at over a million synthetic docs a day. The loop picked a failure scenario, ran the harness, read telemetry through the Datadog MCP, proposed a fix, shipped it to staging and repeated. A coordinator session handed work to sub-agents, usually about 11 at a time in the busiest week and 30 at the peak. Before the run, Nair told Claude which decisions needed his approval: "basically anything hard to undo", such as schema or key-layout changes.

The team also picked its stack for the agent's benefit. It wrote the service in Rust and vendored its few dependencies: "We wanted any number Claude measured to trace back to code we wrote." In one case, a large doc was using about 3 GB on its pod; profiling traced it to every keystroke copying the doc into several data structures. After a copy-on-write fix the doc topped out around 180 MB, and the pod estimate for launch fell from more than 500 to about 50.

Anthropic reports that users created over 4.6 million docs by September 22, with 60 server errors across 2 billion requests, 629 production rolls in September and no rollback. Nair puts the inference cost at roughly $125,000 at Claude Fable 5.2 list price. He estimates a similar push on a small stack at $3,000 to $10,000, and says he has not run it. The post does not include the underlying telemetry, and we have not seen the figures reported independently.

**Why it matters:** This is a worked recipe for an agent-run load test with a stated price: give the agent telemetry, write down the invariants, and reserve hard-to-undo changes for a person. The argument that runtime choice should follow what the agent can measure is a new criterion for picking a stack, and the post says the team did not compare Rust against Go directly.

## 2. Jesse Vincent gives coding agents a dedicated toolsmith

**[The Toolsmith](https://blog.fsck.com/2026/11/17/toolsmith/)** · Jesse Vincent · Massively Parallel Procrastination, November 17, 2026

Vincent, who wrote the Superpowers plugin, says his agents already followed an AGENTS.md rule to script anything they would do twice. They built those scripts on the way to other work and left them in ephemeral scratch directories. His conclusion: "'write a script to automate this task' is a separable mission and deserves an agent's exclusive focus."

The [toolsmith plugin](https://github.com/prime-radiant-inc/toolsmith) is published under Prime Radiant's GitHub organization. It hands that work to a subagent that keeps a git repository of tools at `~/.local/share/toolsmith/tools`, with a general catalog and one per project. An agent that needs a helper searches the catalog first, and the subagent builds or extends a tool when nothing fits. The README says it is built for Claude Code and generated for other harnesses, and that the path outside Claude Code is untested. Vincent reports dozens of tools built in a couple of weeks of his own use; no other measurements are given.

**Why it matters:** Treating tool-building as its own mission, with its own context and a persistent catalog, is a decomposition choice any harness can copy. The plugin costs one install to try, and the catalog it leaves behind can be inspected.
