News archive
Recruiting, HR & AI news: November 9, 2026
Monday, November 9, 2026. 54 stories, each linked to its source.
Ranked by importance within each day. Pick a topic to see it newest first.
Monday, November 9
Meta Superintelligence Labs has released Muse Ember, the smallest and most cost-efficient model in its Muse family, built for the high-volume work agents repeat all day: tool calls, coding subtasks, routing, extraction and long-document parsing. It costs $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens, with a 1M-token context window and up to 512K output tokens; requests over 200K input tokens are billed at higher long-context rates ($0.30 input, $1.50 output). Meta pitches Ember as a cheaper alternative to Anthropic's Claude Haiku 5.5 and OpenAI's GPT-6 Luna, meant to run subtasks alongside a larger planner model such as Muse Wonder. It's available now in Muse Code and on the Meta Model API as `muse-ember-1.0`, with weights kept private.
Get the news by email
Free newsletter
Get the Work Insiders weekly
One email a week: the recruiting and HR tech news that mattered, the tools we just added to the directory, and one practical guide. No spam, and you can unsubscribe in one click.
Meta brought Muse, its personal AI agent, to smartwatches, with version 15.0 of the Muse iOS app adding an Apple Watch companion and availability also confirmed on Wear OS watches.
ChatGPT Enterprise is now available through AWS Marketplace, helping organizations simplify procurement through existing AWS purchasing and billing relationships.
Nominal, which makes software for testing and engineering complex hardware, has raised a $240 million Series C at a $3 billion valuation, led by Lightspeed Venture Partners and co-led by Insight Partners, with Lux Capital, Sequoia, General Catalyst, Founders Fund, Salesforce Ventures and Lockheed Martin Ventures also participating. SiliconANGLE reports the round triples the $1 billion valuation from its $80 million Series B-2 in March. Nominal's Hardware Intelligence Platform connects test-stand and sensor data with engineering context so AI agents can work from it; its first agent, Nominal Analyst, investigates engineering data and runs analyses with answers traceable to the evidence, while engineers keep control of reviews and approvals. The company serves more than 80 commercial and government organizations, including five of the seven largest defense contractors, and plans more agents for requirements verification, planning, scheduling and fleet health, plus expansion through acquisitions.
Automattic has launched Spacefast, a new company that lets you build a website or app inside the AI you already use, including ChatGPT, Claude, Grok Bot, Cursor, Codex and Claude Code, then ask it to publish and get a live private link back in the chat. Spacefast includes hosting, your own domain and a database, supports stacks like Next.js, Astro, TanStack Start and Vite, and lets teammates comment directly on the page so the agent can turn feedback into fixes; built-in app authentication, transactional email and shared team knowledge are due in the coming weeks. It runs on WordPress.com's infrastructure and is available as an official plugin in ChatGPT and Claude, plus an API, CLI and MCP server for other agents. It's free to start, Personal plans include $1 a month of usage, and the Go plan is $5 a month with unlimited spaces and more usage.
ClickHouse has made its Managed Postgres service generally available on AWS (GCP stays in private preview) and acquired Xata, the agent-native Postgres company, in what it calls its largest platform expansion yet. Xata's copy-on-write storage will bring instant database branches and clones, automatic idling and scale-to-zero environments, so developers and AI agents can build and test in isolated Postgres copies, and ClickHouse plans a dedicated tier for agent sandboxing and agent memory. The company also launched ClickHouse Observability in public beta, ClickHouse Security (an agentic SIEM built on its RunReveal acquisition) in private preview, and the open-source Langfuse AI Gateway in public beta for tracking AI usage and costs. ClickHouse says it now serves more than 5,000 customers, and with GA the Managed Postgres beta discount ends and backup storage and egress become billable.
Expo published a pre alpha preview of Exact, an MIT licensed cross platform app runtime written in Rust on the assumption that much of the code will be written by AI agents, running natively on the web, macOS, iOS and Linux.
Meta is expanding the AI age-estimation technology it uses on Instagram to every market, so accounts it suspects belong to teens are placed in Teen Account protections even when they signed up with an adult birthday. Teen Accounts automatically limit who can contact a teen, what content they see and how much time they spend in the app, and Meta says the expansion will cover millions more teens. Meta is continuing the rollout on Facebook, plans to launch similar technology globally on Threads later this year, and has started using Apple's age-range signal as an extra check in Australia. Parents in the US will start seeing an in-feed Instagram notice this month explaining how teens are placed in these protections and how to make sure a teen's age is correct.
Independent benchmarking firm Artificial Analysis has released version 2 of its Decision Index, which ranks AI models on making decisions and extracting facts, and 45% of it now runs on private datasets, up from 25%, to guard against benchmark gaming. The biggest addition is AA-Paperclip, a private set of 1,500 documents split across 500 resumes, 500 receipts and invoices, and 500 AI research papers, where a model must make five decisions and pull ten facts from each one and only passes the document if all 15 answers are right. Claude Haiku 5.5 (Max) passed 92% of documents and GPT-6 Luna (Max) 86%, while the top specialized decision model, Drex Max v1.1 from Nace AI, got 97% of individual answers right but passed only 63% of documents. Drex Max v1.1 leads the decision-model leaderboard at 73, within a point of Fastino Labs' GLiDE v1.2 and Baseten's Mammoth 27B, and the new headline score also penalizes models that are confidently wrong.
Google has expanded its eighth-generation TPU lineup with two new systems, bringing it to four TPU systems announced in a single year. TPU 8tx is built for secure and confidential inference and small-scale training: it uses the same silicon as TPU 8t, runs on 5th Gen AMD EPYC host CPUs, and keeps data and model weights protected even while in use, which Google and AMD call the first cloud TPU with hardware-enforced confidential computing, aimed at governments and regulated companies. TPU 8i Max puts two TPU 8i chips on one package for double the compute, SRAM and HBM bandwidth, targeting long-context work such as many agents operating across a whole codebase or hours of video. Bloomberg reports both are reconfigurations of existing designs rather than new chips and will be available in 2027.
Learn how OpenAI’s Control Room tracks model workloads across our research infrastructure and can rapidly stop them when safety or security concerns arise.
No stories on this topic yet.






