Stripe pays $7.5B for the model router; Anthropic's 6-to-1 agent data; what Notion, Cursor and Linear shipped

Gmail Live — voice search in Gmail's mobile apps: ask a question out loud and Gemini answers from your inbox, showing an email source card at every conversational turn so users can see where each answer came from. If you build productivity features, voice is now bundled into Google's Workspace tier rather than being anyone's differentiator. It is on by default for organizations with Gemini in Workspace access, rolling out gradually from October 15 over up to 15 days, but only to US users with EN-US as their language. Google Workspace Updates

Gmail AI Inbox — automatically extracts action items, deliverables and deadlines from email threads as suggested to-dos, and summarizes important topics across multi-message threads. Google is expanding it from consumers to paid Workspace users. Bundled into Gmail, it resets expectations for any product whose workflow starts in email. There is no admin control to turn it off, it is US EN-US only for now, and Scheduled Release domains get it from October 26. Google Workspace Updates

Linear — stacked pull requests now display in Linear Diffs, diffs update in realtime while a coding agent works, and a new AI credit usage explorer breaks spend down by team, feature, user or loop. Per-user cost control is becoming a first-class admin feature — worth copying if you meter or price agent usage internally. Admins can set spend limits per workspace, user or loop; when a limit is hit, AI credit usage pauses until it resets or an admin raises it. Linear Changelog

Cursor — two new bots: Rollouts writes a monitoring plan as a PR comment when a pull request opens, then verifies the change's health per environment after deploy; Security Review reports exploitable bugs on every PR. Projects, in beta, holds context over months of work and delegates to thousands of subagents on its own cloud computer. Deploy verification moves agent output from 'did it code' to 'did it work in production' — the gap most agent features still have. Both bots are on Teams and Enterprise plans, with trial credits covering roughly 50 and 500 changes respectively. Cursor Changelog

Notion 3.8 — MCP connections let outside agents read workspace context and update docs and task trackers, Notion agents reach more than 150 tools, and a Skills API syncs skills to GitHub, ChatGPT and Claude. New admin MCP controls show who connected each tool and let you revoke access. A clear signal of where agent context is settling: one durable store per company, with any agent allowed to read it under admin oversight. On Business and Enterprise plans, an existing ChatGPT Plus or Pro subscription can cover AI usage in Notion with no markup. Notion Release Notes

Muse (Meta) — a podcast episode, not a launch: Meta design director Alex Cornell explains how Muse is designed — one persistent conversation instead of separate threads, and progressive disclosure that shows the same activity as an emoji for casual users, a task list for anyone who wants detail, and raw logs for engineers. The most concrete public answer yet to what an agent interface should feel like when it works. Cornell argues visible chain-of-thought is usually a workaround for latency — Muse is fast enough to just deliver results quietly. How I AI (Lenny's Podcast Network)

“How would this work if you were talking to another human?”

How employment is changing at Anthropic — By August, agent hours outnumbered employee hours 6 to 1, and junior technical hires fell from 27% to 9% as agents absorbed implementation work. Direct input for your headcount and review-capacity planning: the bottleneck Anthropic hit is human attention, not code output. Meta-work — building dashboards and monitoring agent pipelines — grew from 10% to 22% of interactive Claude sessions between fall 2025 and spring 2026. Anthropic

“an army of Claudes, with every one asking for my attention”

AI prototyping at Uber — two Uber product managers describe a year of opening AI prototyping — Lovable, Figma Make, Claude Code, Cursor — beyond product teams to engineering, platform, sales and ops. Prototypes moved earlier in the lifecycle; PRDs got sharper rather than replaced, in a pairing of problem framing, early prototype, then PRD, then final design. A tested playbook for discovery, including explicit rules for when to go prototype-first versus PRD-first. A Business Platform PM explored six concepts for one problem in about 20 minutes, and nearly 40% of hackathon submissions used a prototyping tool. Uber Engineering Blog

Rillet — an AI-native ERP whose two-person frontend team tripled its shipping rate in three months building agents on Vercel's open-source eve framework. Customer success highlights what a customer sees in a Chrome extension; an agent captures the context, writes the spec, makes the change and opens a pull request. A working answer to how small a team can be when agents handle the loop end to end — customer-facing fixes no longer need an engineer in the middle. The in-house agent Nebula ships customer requests to production within two hours, and the team closed more than 800 Linear tickets in its first four months. Vercel Customers

The new UX of AI Agents — a paid deep dive mapping more than 40 real UI examples from 26 companies into seven patterns: memory you can see and edit, risk-graded approvals (Perplexity's Allow, Always ask or Deny per tool), canvas collaboration, hand-editable output, background agents, audit trails, and agents embedded where people already work. A pattern library to steal from before designing your own agent surface, with named examples rather than abstraction. Intercom found positive experiences with AI agents go from 49% to 74% once someone watches one resolve a real query; WorkOS reports attackers writing prompts designed to make people approve agent actions unread. Department of Product

Gemini 4 Argon — Google's new frontier model, globally available since October 13 in the API, Antigravity, the Gemini app and Search. The number that matters for product budgets: an introductory $2 per million input tokens and $10 per million output tokens, doubling after November 30, 2026. If you default to Gemini anywhere in your stack, lock in your cost model or plan the switch before the intro price ends. Google says Chrome teams used it to separate a roughly 6-million-line legacy folder from Chromium's 40-million-line codebase, with agentic code submissions up more than 35% and rollbacks down nearly a third. Google Blog

Gemini skills — skills in the Gemini app began rolling out October 13 and should finish by mid-November, but the Workspace rollout — previously due to complete October 12 — is delayed with no new date announced. If enablement or workflow plans depend on Workspace skills, they just slipped; note the Gemini app version ships with no admin controls at all. Sharing and a Google Drive integration for Gemini app skills are still to come, and Workspace skills are restricted to users over 18. Google Workspace Updates

App Store guidelines — Apple now explicitly bans apps that use AI to create offensive content, including ones that pass review as ordinary apps and reveal nudify features afterward — the bait-and-switch tactic — and will remove apps that add undisclosed features after approval. If you ship a consumer AI app, moderation tooling and an honest review build are now platform requirements, not nice-to-haves. Apps hosting user-generated content must filter objectionable material and offer reporting and blocking; Apple says it blocked over 150,000 apps and removed 193,000 developer accounts last year. TechCrunch

Digital Omnibus — the European Commission's proposal to amend the AI Act: rules for high-risk uses in sensitive areas like employment apply at most 16 months later than originally planned, tied to the availability of standards, with product-embedded high-risk rules delayed at most 12 months. EU compliance work on your roadmap moves — but nothing is cancelled, so keep the work scoped and re-sequenced rather than dropped. SME-style simplified obligations extend to small mid-cap companies, covering an additional 8,250 companies in Europe, and an EU-level regulatory sandbox for real-world testing is proposed from 2028. European Commission AI Act Service Desk