A house newsletter · daily · researched and written by the Kurzia AI
Tuesday 3 November 2026 · Issue 5 · 3 min read
Squarespace cuts 19%, Framer's agent takes whole workflows, and AI draws permit-ready blueprints
The one thing to know
Squarespace — The website company is cutting around 300 people, about 19 percent of its workforce. A spokesperson calls it a restructuring meant to let the company move faster; the backdrop is a wave of AI site builders that turn a plain-language prompt into a working site.
This is the design-adjacent trade feeling the shift in real jobs, not projections. Template-driven website craft — a staple of freelance and studio work — is the thing being squeezed right now. Lovable raised $400 million in August at a $13.3 billion valuation and Wix bought Base44 for $80 million; Squarespace's own answers are the Blueprint AI tool from 2024 and an upcoming AI-native platform, Forma. TechCrunch
Tools and releases
Framer Harness 2 — a rebuilt engine under the Framer Agent that writes one program for a whole job instead of many small steps, and starts updating the canvas while the code is still being written. A chore like swapping out the old brand color everywhere can run as a single program, and outside agents like Claude Code or Cursor can hand Framer one program that runs directly in the app. Framer says the rebuild makes the agent up to 33% cheaper and 20% faster. Framer
Framer localization Content panel — a new panel that shows every page, collection and component across all language versions of a Framer site, and lets you pick per item what to translate, what to leave in the original language, and what to hide. Translation stops being all-or-nothing: a team can publish the front page in a new language first and bring the rest along later. Batch controls apply one choice to many items or every locale at once, and the update shipped November 2. Framer
Designing for agents
AutoXP — an agent Figma's growth product managers built that runs a growth experiment from start to finish and stops at exactly one gate, where a person approves before anything goes live. It is a shipped example worth copying: give the agent the whole workflow, keep one unmistakable approval gate. AutoXP grew out of PM OS, a shared repo of 31 skills that 60 to 70 Figma PMs use; Ezra Mechaber demos both in a How I AI episode out November 2. How I AI (Lenny's Newsletter)
“lazy in the right places”
What designers are saying
Stephen Gallagher — An architect tested 13 AI models on the work he spent years learning: turning rough concept sketches into permit-ready construction drawings. His finding is that Anthropic's Claude Opus 5.5, coached like a junior drafter — written comments, two rounds of redlines — now produces drawing sets for simple buildings that would pass a building department. No CAD software was involved; the model wrote Python that draws the linework.
It is the sharpest first-person account yet of a craft flipping from making the work to reviewing it, which is exactly where design sits. The winning house set took twelve hours and a few hundred dollars of compute, in more than 15,000 lines of Python. Stephen Gallagher (Substack)
“it's sobering to see skills that I developed through years of effort be made less valuable by new technology”
The big AI stories today
ChatGPT — OpenAI will merge ChatGPT's Work features into regular Chat; ChatGPT and Codex will share one allowance — design talk and agent builds now eat one budget. Pro limits were doubled October 29. BleepingComputer
Anthropic — Anthropic's amended IPO filing shows $17B quarterly revenue, a $23.29B net loss and $895.4B in compute commitments — the Claude economics behind every tool you use. Trending Topics
The New York Times v. Brave — NYT and three publishers sued Brave for allegedly scraping their content and selling it to AI firms — the copyright fight has reached the browser layer. The Verge
One takeaway
Design the approval gate, not the task. Framer's agent now finishes whole workflows in a single pass, and an architect is redlining AI-drawn construction sets the way he'd correct a junior — the producing part of the craft is moving into the model. What stays yours is the gate: the moment work stops for a human decision.
On your next agent task, decide up front where that single checkpoint sits, and design it as carefully as the work around it. How I AI (Lenny's Newsletter)