- Rundown
- Cosmic Rundown: Outages, Rust 1.100, Local Models

Cosmic AI
November 13, 2026
This article is part of our ongoing series exploring the latest developments in technology, designed to educate and inform developers, content teams, and technical leaders about trends shaping our industry.
Today's links fall into a few groups. Several hosted services had trouble. Some people wrote about taking their devices back from the cloud. Rust shipped a new release. There was also a lot of interest in running models on your own hardware. Below, the outages come first as a table, followed by short sections and a list of quick hits.
Status Page Check
Three stories were about services being down or about who controls the software you depend on:
| Story | What it is | Discussion |
|---|---|---|
| Service degradation on Microsoft 365 | Microsoft's status page during a broad 365 incident | HN |
| Elevated errors on claude.ai, Claude Code, and platform.claude.com | An incident covering Anthropic's chat app, coding tool, and API console | HN |
| "LibreOffice works 365 days a year. Just saying." | The LibreOffice project's reply to the Microsoft outage | HN |
If you depend on an AI coding tool every day, keep a fallback ready before the next incident. That could be a second provider, a local model, or a plain editor.
Taking Devices Back From the Cloud
Two hands-on projects made the same argument through code:
- Reverse engineering my Coway air purifier to ditch the cloud (discussion) walks through cutting a consumer device off its vendor's servers.
- Show HN: My $300 embroidery machine needed a subscription, so I built my own app (discussion) replaces a subscription-gated companion app with an open source one.
Both projects come from the same frustration as the outage table. When the vendor's server is the only path to a feature, the feature stops working when the server does, or when the vendor changes its business model.
Systems Languages Had a Good Day
- Rust 1.100 (discussion) is out. Read the official release notes before upgrading your toolchain in CI.
- Relax: a JIT compiler for Rust to reduce build times (discussion) goes after Rust's best-known pain point, slow compiles.
- Why Zig over C for systems programming (discussion) makes the case for Zig in the work C usually does.
- A technical vision for the Haskell ecosystem (discussion) lays out where Haskell tooling could go next.
- Lonewolf 0.3 (discussion) is a standalone XMPP server written in Rust, and it is a practical example of the language in use.
Security: Breaches, Incidents, and Supply Chains
This section is less a set of summaries and more a set of questions to ask your own team:
- Who holds your customer data, and how much of it? Brazil's Caixa insurer breach exposes names and tax IDs of all 7.6M customers (discussion).
- Do your scraping and automation vendors have access to secrets? Apify posted a security incident report (discussion).
- Would your team spot a fake video call? The FBI says North Korean hackers used fake video calls in a $400M crypto hack (discussion).
- How old is your database code? Turning a 22-year-old Postgres type-confusion primitive into RCE (discussion) is a detailed exploit writeup.
- Are your CI dependencies pinned? Lock your GitHub Actions (discussion) explains why third-party actions should not float.
Local Models and MCP Tooling
Running models on your own machine is getting cheaper:
- Edge0-145B: frontier AI capabilities on your own computer (discussion).
- Apple's Foundation Models framework (discussion) gives app developers on-device model access.
- LM Studio vs. Ollama on Linux (discussion) runs the same model files on an RTX 3090 and compares the results.
- Frontier AI takes only months to get 10 times cheaper (discussion) puts numbers on how fast costs are falling.
New MCP and agent tooling from the newest queue:
- CtxWise targets large MCP payloads, schema bloat, and cache misses (discussion).
- Lore lets you ask questions of your architecture decision records using local hybrid search and MCP (discussion).
- AI Juggler orchestrates the desktop Claude and Codex apps (discussion).
AI and Trust in Writing
- AI is destroying the trust between writers and readers (discussion) looks at AI-generated literary scams.
- Is it OK to write blog posts with AI yet? (discussion) is one writer thinking through the question.
- Jonathan Blow: If AI solved coding, why is my software still the same? (discussion) is a skeptical talk on what AI has actually changed.
Quick Hits
- IceWM: the lightweight window manager is back on the front page. (HN)
- Switcher (2004): an early Mac multitasking story. (HN)
- AI unlocks Gödel's notebooks: the notebooks are said to contain hundreds of proved theorems. (HN)
- EU Digital Identity Wallet rollout tracker: every member state's progress on one board. (HN)
- OpenTelemetry for Balatro: observability tooling applied to a card game. (HN)
- Isometric SF: a pixel-art map of the Bay Area. (HN)
- CERN built the LHC's next-gen magnets. (HN)
- Doomsday: 13 November, A.D. 2026 (1960): a 1960 paper whose predicted date is today. (HN)
- Arm boss Rene Haas approached in Nvidia CEO search. (HN)
- Anthropic sued for trademark infringement over 'Mythos' model. (HN)
What This Means for Content Teams
Plan for outages and keep your content portable. The Microsoft 365 and Claude incidents are a reminder that any hosted tool can go down. A headless CMS keeps content separate from the front end, and the Cosmic REST API serves it from an edge-cached API with full import and export. Your content is not locked inside any one app.
Pin and scope your integrations. The GitHub Actions and Apify stories both come down to limiting what third parties can touch. In Cosmic, each Bucket has separate read and write keys, so a front end that only reads content never holds a write key.
Give agents a defined interface. CtxWise, Lore, and AI Juggler show how much work now goes into connecting agents to tools. The hosted Cosmic MCP Server gives Claude, Cursor, and other MCP clients a scoped way to read and manage content. Cosmic AI Agents draft and publish inside your content model.
Keep a person in the loop. The Atlantic piece and the rmoff post are both about trust. Workflows add approval gates, so an editor reviews AI-assisted content before readers see it. You can see the rest of the platform on the features page.
To get started, read the documentation or create a free account.






