Research Period: 19 October to 26 October 2026
1. Google Meet now translates 63 languages in real time, with automatic language detection
What changed
Google Workspace has expanded real-time speech translation in Google Meet to 63 languages and added automatic language detection, so participants no longer need to select a source language before a meeting starts. The update is available on Business Standard, Business Plus, Enterprise Standard, and Enterprise Plus plans. Rapid Release domains have it from 22 October; Scheduled Release domains follow on 2 November.
Why it matters
If your team works with international suppliers, clients, or colleagues, this removes one of the more friction-filled moments in global meetings: figuring out in advance who speaks what and configuring translation before the call. Automatic detection handles the setup step for you. With 63 languages, coverage extends well beyond Western Europe to include Hindi, Arabic, Swahili, and many others. You do not need a new subscription; if your organisation is already on one of the eligible Workspace plans, this is available in your admin console now.
Use cases
- Enable translation on your next supplier call with a manufacturer in Germany, France, or Japan, without preparing a bilingual summary afterwards
- Use it in a client meeting where an international partner is more comfortable speaking in their first language
- Run internal briefings for a team spread across the UK, India, and West Africa without a shared working language
- Test it on a lower-stakes call first to check accuracy before relying on it with an important client
- Check your Workspace Admin Console to confirm whether your domain is on Rapid or Scheduled Release and plan accordingly
Sources: Google Workspace Updates
2. OpenAI Decisions API opens to all developers: cheaper, faster structured classification and routing
What changed
OpenAI’s Decisions API opened to all developers in public beta on 6 October, running on GPT-6 Luna. We have not covered it before. It returns structured outputs in one of three forms: a binary probability (predicate), a labelled choice from a list you define, or a score against levels you set. It accepts text and images. Pricing is $0.10 per million input tokens, with output tokens at no charge. European Economic Area and Switzerland data residency is available from launch.
Why it matters
In many agentic workflows, the most expensive step is not the complex reasoning: it is the thousands of small classification calls that route work, filter noise, or decide whether to escalate. At $0.10 per million input tokens against roughly $1–3 for a full model (our estimate), and about ten times the speed (OpenAI’s own figure), ticket triage, email sorting and similar routing-heavy systems become much cheaper to run. If you already build with the OpenAI API, this is a single endpoint change. It is an alternative to TypeSafe’s Jev, which launched last month: Jev costs $0.042 per million input tokens, with output also free.
Use cases
- Route inbound customer enquiries by type (billing, technical, sales) using a choices output rather than a full-text response, before they reach your team
- Score incoming reviews or feedback for urgency and sentiment as a low-cost first pass before further processing
- Add an intent gate to your agent pipeline to check whether a tool call is actually needed, saving tokens on simpler requests
- Build a document classifier that assigns categories before sending each document to a more expensive summarisation step
- Run a simple priority filter for your inbox using the predicate output, which returns a probability rather than a full explanation
Sources: OpenAI Blog | AI Weekly | TypeSafe
3. Anthropic donates the Agent Skills open standard to the Agentic AI Foundation, alongside MCP
What changed
Anthropic announced on 22 October that it is donating its Agent Skills open standard to the Agentic AI Foundation (AAIF), a Linux Foundation project it co-founded with Block and OpenAI. Agent Skills lets AI agents carry specialised knowledge and instructions via a plain-text SKILL.md file. It already works in ChatGPT, Cursor, GitHub Copilot, and VS Code, with 70-plus platforms integrated, more than one million skills published, and around 280 million installations.
Why it matters
The AAIF now governs both Model Context Protocol, which lets AI tools connect to your data, and Agent Skills, which carries your way of working. If you have set up MCP for Claude or used Agent Skills in Cursor or VS Code, this matters because both standards are now maintained by a vendor-neutral body under the Linux Foundation, reducing the risk that either gets tied to one company’s roadmap. For members building AI workflows: a SKILL.md file you write today will work across any platform that adopts the AAIF standard, not just Anthropic’s tools.
Use cases
- Search the public Agent Skills registry for published skills that match your existing workflows, such as coding conventions, a brand voice guide, or a document structure template
- Write a SKILL.md file for your company’s writing style or technical standards and use it consistently across Claude Code, Cursor, and GitHub Copilot
- Explore how MCP (connecting to your tools and data) and Agent Skills (what your agent knows how to do) work as complementary building blocks in the same foundation
- Install and test one community-published skill in VS Code this week to understand what the ecosystem already offers before writing your own
- Keep an eye on the AAIF working group for governance updates as enterprise policies around agent capabilities develop
Sources: Anthropic Blog
Additional Noteworthy
Claude Duet (Anthropic): Anthropic launched Claude Duet, a proactive AI assistant built on Opus 5.5 that sends a morning brief across connected apps, executes tasks by text or voice, and checks before taking consequential actions. Currently in research preview for US Claude Pro and Max subscribers only; UK and international availability not yet announced. Watch for a rollout date. | iPhone in Canada
OpenAI models on Google Cloud Model Garden: GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are now in private preview on Google Cloud Model Garden, alongside OpenAI’s Codex agentic workflows. Usage draws down existing Google Cloud committed spend and appears on a single invoice. GA expected in coming weeks. | OpenAI Blog | The Next Web
ChatGPT advertising reaches 89 countries: OpenAI added 26 new markets on 21 October, including South Africa, Nigeria, Argentina, Chile, Colombia, and several APAC countries. UK and EU markets were added in August and remain covered. Ads appear on the standard and Go subscription tiers only; paid plans remain ad-free. | PPC.land
Strategic Awareness
A Nielsen survey of 1,700 senior managers across seven countries, commissioned by Monday.com, highlights a widening performance evaluation gap in AI-adopting organisations. More than half of managers now find it harder to assess individual employee contribution because AI-polished output makes it less clear who did the work. Around 78% say they have changed how they evaluate staff, weighting skills like prompt writing and AI delegation more heavily. Fifty-eight per cent expect to manage more AI agents than human employees within a year; 46% say they worry about losing control of those agents. The underlying point is one worth sitting with: how you run your team may matter as much as which tools you adopt.
Sources: TechCrunch
Looking Ahead
Claude Duet is US-only for now, but it signals a shift towards AI assistants that act proactively rather than waiting to be asked; the UK rollout is worth tracking. The AAIF Agent Skills ecosystem is growing quickly, and the combination of MCP and Agent Skills as open, vendor-neutral standards makes this a good moment to start building reusable skills for your own workflows before the ecosystem matures further.
This briefing focuses on AI developments relevant to SME operations, emphasising practical applications in productivity, cost efficiency, and business capability. For questions or to suggest sources reach out on the group WhatsApp.
