A house newsletter · daily · researched and written by the Kurzia AI
Wednesday 11 November 2026 · Issue 13 · 4 min read
Claude gets whiteboards and meetings, Cursor Projects goes multiplayer, and design hiring hits a record low
The one thing to know
TechCrunch · yesterday
Anthropic ships whiteboards, sheets and a meeting-room Claude
Takeaway: beta on paid plans
Anthropic turned Claude into a shared workspace yesterday. Claude Whiteboards, in beta on paid plans, lets a team and Claude brainstorm on one board, which Claude organizes into a timeline with owners and dates and can then turn into a doc or slide deck. Claude Sheets builds and edits spreadsheets with the team, outlining each change it makes. Claude Meetings, in beta on Team and Enterprise plans, joins a Google Meet call as a visible participant, drafts the plan being discussed while people talk, and leaves on its own if an outside guest joins.
This is the agent moving onto your surfaces instead of you exporting out of its chat: boards go to Miro, Lucid or Vani by Zoho, and full apps like Adobe, Atlassian and HubSpot now open inside Claude so the agent works in the same tool as you. If you design agent products, steal the trust moves — Claude names itself in the meeting chat and walks out when a stranger arrives. Anthropic's pitch is that people already brainstorm with Claude and then move the result into another tool by hand; the board exists to remove that hand-off.
Tools and releases
Firefly Builder — Adobe's new prototyping tool, in private beta inside Firefly, for building designs that show an idea — its most direct step onto Figma's territory. Alongside it, Adobe released an MCP, a standard that lets ChatGPT, Claude or Gemini drive features in Photoshop, Illustrator, After Effects and Premiere, and made its Firefly agent proactive and always on. Photoshop also gains relighting and prompt-based style sliders; Premiere gets a new pro color-grading system. Adobe says it will watch how people use the exposed features and keep opening more of them up. TechCrunch
“We are still figuring out where and when and how to draw the line.”
Cursor Projects — the multi-agent workspace Cursor introduced in September is now multiplayer: teammates join alongside their agents, with one task board showing who and which agent is doing what, plus a shared context layer of requirements, diagrams, specs and mockups sitting on top of each person's own chats. Agents now talk across owners, spotting overlaps like a designer redesigning a page while an engineer adds reviews to it. Cursor
Lovable's security scanners — check your app each time you publish — the code, the logic, the database schema and the rules deciding who can see your data — and a Fix button hands the exact finding to an agent to resolve in a few clicks. Turn on auto-fix and many findings clear with your next chat message; publishing controls can block anything with critical findings from going live. The most common finding in Lovable apps is data-access rules left too open. Scheduled deep scans come to all paid plans this month. Lovable
Modern Web Guidance — a free set of skills, stewarded by the Chrome, Edge, Firefox and MDN teams, that steers coding agents away from dated web patterns and onto current platform features — the training data those agents learned from is loaded with old code. One npx command installs it for Claude, Cursor, Codex and other agents, and a legacy-scan command flags outdated patterns in an existing codebase. It covers 250 modern web features and 203 real-world use cases, from container queries to native dialogs, kept lean by pruning what models already know. GitHub
Designing for agents
ChatGPT Business Agents — verified agents that speak for a brand inside ChatGPT, launching in beta for US e-commerce and online-service businesses. When a business comes up in an answer, you can ask its agent about offerings or the difference between two products, and the brand's own agent answers. The trust pattern is worth studying: chats are clearly labeled, kept separate from your main conversation, and the company behind the agent had to claim and verify itself through Business Hub — which also shows brands where they surface in ChatGPT and on which topics. OpenAI is not publishing its verification methods, since they differ by business type, says VP of monetization Benji Shomair. The Verge
Grok Build's share links — Grok Build, the vibe-coding agent from SpaceXSI, leaked confidential data from thousands of apps — including 24 built inside SpaceXSI itself — because its share links put the random secret in the subdomain, where passive DNS services can list every link. Nearly 95 percent of the 43,601 links found opened the app to anyone holding them. The researchers say the same exposure sits in Lovable, Bolt and Replit apps, while Claude, ChatGPT and Gemini put the secret in the URL path, where DNS cannot see it. On agent-built products, the share-link default is a design decision someone has to own. Ars Technica
What designers are saying
Lenny's Newsletter · yesterday
Design hiring hits a record low; agent fluency rises
Lenny Rachitsky's late-2026 job-market report counts about 1,900 open product design roles at tech companies worldwide — the lowest since tracking began, and down 28 percent in a year. User research has fallen harder still: US hires are down two-thirds from 2022, and teams increasingly learn post-launch instead.
The counterweight for you: AI-tool fluency is now a named designer skill. 'Claude' has become the second most-mentioned term in designer job descriptions, passing CSS, with Cursor close behind — and demand for Figma experience keeps growing. Designers aren't dropping their main tool; they're layering agents onto it, and the JDs now say so out loud. PM demand points the other way: product-manager openings have roughly doubled in number relative to design openings since early 2023, now 3.8 to 1.
The big AI stories today
Anthropic's IPO — Anthropic opened its IPO roadshow: 825 million shares at $70–$78, near a $1.4 trillion valuation — Claude's maker hits Nasdaq as 'ANT', possibly trading next week. Anthropic
AlphaProof Code — Google DeepMind's AlphaProof Code had agents write six million lines of machine-checked proofs verifying Signal's real code in four weeks — deep verification is now agent work. Google DeepMind
One takeaway
Move the agent onto the shared surface — and check its defaults. Today's launches share one shape: agents are leaving the private chat for shared surfaces, from Claude's whiteboards to Cursor's team task board. Run your next kickoff on one and watch where it breaks — that's where the design work now sits. And before anything vibe-coded ships, check the share-link setting: Grok Build shows what a default nobody designed can cost. TechCrunch