The AI Productivity Weekly: Saturday 24 October 2026
Claude Duet, Notion agents that run for a day, Perplexity Computer on your Mac, and OpenAI and Claude side by side on Google Cloud.
This Week in One Minute
Welcome to Edition 2 of The AI Productivity Weekly, a short Saturday roundup of the AI news that changes how you work.
The theme: agents that keep working after you step away, and more choice over where they run. This edition covers news through Saturday, Oct 24. All dates are in UTC.
1. Anthropic opens a research preview of Claude Duet, a proactive agent app
Claude Duet is a standalone app (iOS, Android and claude.ai) that sends a morning brief across your connected apps, flags what needs you during the day, and works in its own virtual machine while you’re away. It runs on Claude Opus 5.5 and asks before it sends a message, makes a payment, signs a document or deletes anything. The waitlist preview starts with Pro and Max subscribers in the US and shares their plan’s usage limits; Team and Enterprise come later.
Why it matters for your productivity: the assistant comes to you with what needs a decision, instead of you checking every inbox.
2. Notion rebuilds its agents: day-long tasks, ChatGPT plans and a new credit system
Notion Agent and Custom Agents now have their own computer and browser and can work on one task for more than 24 hours, up from five minutes. Notion says its upgraded Auto router gets results similar to frontier models at about 90% lower cost. You can connect a ChatGPT Plus, Pro or Go account to your personal Notion Agent, and Business and Enterprise seats now get 1,500 Standard and 500 Select credits a month (details).
Why it matters for your productivity: long jobs like a weekly research digest can run unattended. Check your usage before existing workspaces switch credits on Nov 2.
3. Perplexity Computer can now run fully on your Mac
The Perplexity Mac app can now keep Computer’s model, planning and file processing on the device, using Qwen 3.6 or Perplexity’s PPLX Pro. With permission it works across local files and apps like Notes, Mail and Numbers, and you decide when a task may use the cloud. It’s open to Pro, Max and Enterprise subscribers on supported Apple silicon Macs; Perplexity recommends 36 GB of memory or more.
Why it matters for your productivity: confidential files can get agent help without leaving your machine.
4. OpenAI and Claude models both arrive on Google Cloud
On Oct 21, OpenAI’s GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna began an invite-only preview in Google Cloud’s Model Garden, with Codex support and general availability “planned in the coming weeks.” Usage can count against existing Google Cloud commitments. On Oct 22, Anthropic made the full Claude Platform available on Google Cloud (Anthropic managed), with Google Cloud sign-in and billing.
Why it matters for your productivity: teams on one cloud can pick the best model for each job without a new vendor contract.
5. Microsoft 365 Copilot adds Projects to Cowork
Projects give related Cowork tasks one home for shared files, instructions and outputs, so you add context once and every task builds on it. Files stay in OneDrive with existing permissions and labels. Projects rolled out worldwide in the Cowork tab on Windows, Mac and web for Microsoft 365 Copilot (Premium) users; a Chat and Cowork version reaches Frontier program users by mid-November.
Why it matters for your productivity: no more pasting the same brief into every new task.
6. Google Meet translates speech across 63 languages and detects who speaks what
Gemini-powered speech translation in Meet now covers 63 languages and detects the spoken language automatically, so each participant hears the call in their preferred language, even when speakers switch. Guests can use it when the organizer has an eligible Workspace or Google AI plan; base Workspace and Google AI Pro plans get up to 90 minutes per meeting.
Why it matters for your productivity: a multilingual team call no longer needs one shared language.
7. Anthropic hands Agent Skills to the Agentic AI Foundation
Agent Skills, the standard where a folder with a SKILL.md file teaches an agent a specific task, now lives at the Linux Foundation’s Agentic AI Foundation, the same home as MCP. Anthropic says more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, support it. Engineers from Anthropic, Google, OpenAI and others will maintain it.
Why it matters for your productivity: a skill you write for your team’s process should keep working as you switch or mix agents.
8. OpenAI’s API adds 12-hour prompt caching and a generally available Decisions API
Developers can now choose 12-hour prompt caching, so shared instructions and context get the lower cached-input price even when requests are hours apart; 12-hour cache writes cost 2.5x the input price. On Oct 22 the Decisions API, which turns text and images into typed answers, became generally available with gpt-6-luna.
Why it matters for your productivity: agents that check in a few times a day get cheaper to run.
Also This Week
- Oct 19: Microsoft outlined how Project Perception, its agentic security system, can contain a company’s AI agents by giving each an identity and permissions, scoring the risk of what they attempt, and assigning Guardian Agents to supervise them.
- Oct 19: Meta released Muse Glimmer 30B, an Apache 2.0 open-weight agent model that runs locally on a single consumer GPU at about 4-bit precision.
- Oct 19: With Connected Sheets, Gemini in Google Sheets can now analyze BigQuery data in plain language, running queries against the full dataset; Gemini in Sheets also reached iOS.
- Oct 21: Docs Live lets eligible Workspace users draft structured documents by voice in the Google Docs mobile app.
- Oct 22: SpaceXSI launched the SpaceX Super Intelligence Partner Network for firms deploying or integrating Grok.
- Oct 23: Gemini Notebook now works as a Connected App in the Gemini app: type @Gemini Notebook in a chat to create notebooks, add sources or ask questions about them.
Tool to Try This Week: Write One Agent Skill
Pick a task you explain to AI more than once a week, like a status update in your team’s format. Create a folder with a SKILL.md file: YAML frontmatter with a name and a description of when to use it, then plain Markdown steps, an example and your house rules (spec at agentskills.io). Run it on last week’s notes in any agent that supports skills and fix what it gets wrong. Keep passwords and client data out of the file.
What to Watch Next Week
- Week of Oct 26: auto mode becomes the default for new sessions in most Claude apps on Pro, Max and Team plans. An automated check replaces step-by-step approvals; you can switch back to manual.
- Oct 29: GitHub Universe opens in San Francisco, with an option to join virtually.
- Around Oct 29: the one-month window in which ChatGPT dots usage doesn’t count toward plan allowances ends; OpenAI still owes per-plan terms.
- Nov 2: Notion’s new credit system reaches existing Business and Enterprise workspaces, and Meet speech translation starts reaching Scheduled Release domains.
That’s Edition 2. If one item here saves you an hour next week, forward it to the teammate who’d use it too, and come back next Saturday for Edition 3.