22 October 2026
Anthropic launches Claude Duet as OpenAI models head to Google Cloud
Anthropic launched a new AI agent for small business owners, and OpenAI said its models are coming to Google Cloud. A serious PHP security flaw found with AI also made news. Retailers and ad platforms are moving fast on AI agents and AI shopping.
Agents
Anthropic launches Claude Duet, a new AI agent (opens the source in a new tab)
Anthropic announced Claude Duet, an AI agent designed for professionals and small business owners. It is a research preview on iOS, Android and Claude on the web.
Why it matters: If you already pay for Claude, try Duet on a small, low-risk task and see whether it saves you time.
The Verge · 21 Oct
Walmart is inviting AI agents to shop for you (opens the source in a new tab)
Walmart will let AI agents shop on its website next month. Systems will let you approve and review purchases and set spending limits.
Why it matters: If you sell products online, expect more buying to be done by AI agents, and check whether your shop can handle it.
The Verge · 21 Oct
Omnara: open-source alternative to Claude Managed Agents (opens the source in a new tab)
A developer has posted Omnara on Hacker News as an open-source alternative to Claude Managed Agents.
Why it matters: If you want agents without vendor lock-in, ask your developer to look at this, but treat it as early.
github.com via Hacker News · 21 Oct
AI agent fixed the bug, but what else did it change? (opens the source in a new tab)
Trail of Bits looks at the risk that an AI coding agent changes more than the bug it was asked to fix.
Why it matters: If your team uses coding agents, have a person review every change before it goes live.
blog.trailofbits.com via Hacker News · 21 Oct
Models
OpenAI models are coming to Google Cloud (opens the source in a new tab)
OpenAI says it is bringing frontier models and Codex workflows to Gemini's Model Garden, available through Google Cloud Marketplace.
Why it matters: If you run on Google Cloud, you may soon be able to use OpenAI models without a separate account.
OpenAI · 21 Oct
OpenAI's latest models coming to Google Cloud and Antigravity (opens the source in a new tab)
Ars Technica reports that Google Cloud will have GPT 6.1 Astra, Sol and Luna in the coming weeks.
Why it matters: Check which of these models fits your work and compare the price once it is listed.
Ars Technica · 21 Oct
Recursive launches Flint, a free laptop coding model (opens the source in a new tab)
Startup Recursive released Flint, a free, laptop-sized coding model. The company says its own AI research system built most of it.
Why it matters: A free model that runs on a laptop could cut costs and keep code private, so ask your developer to test it.
TechCrunch
Security and safety
PHP2Shell: AI uncovers critical remote code flaw in PHP (opens the source in a new tab)
Wiz reports that AI found a critical remote code execution flaw in PHP.
Why it matters: Many websites run on PHP, including WordPress, so ask your web host or developer about patches this week.
wiz.io via Hacker News · 21 Oct
PHP flaws found by OpenAI and Wiz alarm the internet (opens the source in a new tab)
TechCrunch reports that old holes in PHP were found by AI. It says the issue was prevented by AI but is also high risk because of AI.
Why it matters: Update your PHP-based sites as soon as patches are available.
TechCrunch · 21 Oct
OpenAI launches Patch the Planet for open source flaws (opens the source in a new tab)
OpenAI's Patch the Planet connects security researchers, maintainers and incident responders to address critical vulnerabilities across open source.
Why it matters: More flaws in the software you rely on may be found and announced, so keep your updates current.
OpenAI · 21 Oct
Enterprise
How AT&T is rethinking in-house legal work with OpenAI (opens the source in a new tab)
AT&T is working with OpenAI to make legal knowledge easier to access. Its LegalEdge team is rethinking intake and scaling legal judgment.
Why it matters: If you handle contracts or client questions, AI intake and search is a use case you can copy on a smaller scale.
OpenAI · 21 Oct
Mistral Advances AI for Engineering with NVIDIA (opens the source in a new tab)
Mistral is working with NVIDIA on a new generation of AI-native engineering workflows.
Why it matters: This is aimed at large engineering teams, but it shows where Mistral is heading.
Mistral AI · 21 Oct
Meta to create a new class of shares to fund AI push (opens the source in a new tab)
Meta says it will issue non-voting shares to fund AI, acquisitions and employee pay, with trading starting in 2027.
Why it matters: It shows how much money big firms are putting into AI, which may keep new tools coming.
The Verge · 21 Oct
Tools and products
Claude now works with Google Docs, Sheets and Slides (opens the source in a new tab)
Claude now works with Google Docs, Sheets and Slides, according to a post on claude.com.
Why it matters: If your team lives in Google Workspace, you may be able to use Claude on your documents without copying and pasting.
claude.com via Hacker News · 22 Oct
ChatGPT Ads expands to 26 new countries (opens the source in a new tab)
OpenAI says ChatGPT Ads is expanding to 26 new countries and that over 100,000 advertisers have used it.
Why it matters: If you advertise, ChatGPT is becoming another place to reach buyers, so check whether it is open in your market.
OpenAI · 21 Oct
OpenAI CLI (opens the source in a new tab)
A command line tool called OpenAI CLI was posted on GitHub and shared on Hacker News.
Why it matters: If you have a developer, this may make it easier to script OpenAI tasks.
github.com via Hacker News · 22 Oct
Research
Specializing small model vocabulary can cut inference cost by 77% (opens the source in a new tab)
A Hugging Face post says specializing the vocabulary of small language models can cut inference cost by 77% in real applications.
Why it matters: If you run your own models, this is a cost idea worth asking your developer about.
Hugging Face · 21 Oct
NVIDIA Kumo outperforms supervised ML on relational data (opens the source in a new tab)
A Hugging Face post says NVIDIA Kumo Relational outperforms supervised machine learning on relational data with no training required.
Why it matters: If you have customer or sales data in tables, predictions may get easier to set up.
Hugging Face · 21 Oct
Using an LLM for confidence scoring is a bad idea (opens the source in a new tab)
A post argues that asking a large language model to score its own confidence is unreliable.
Why it matters: Do not trust an AI's own certainty score when you set approval rules in your automations.
rumple.github.io via Hacker News · 21 Oct