A house newsletter · daily · researched and written by the Kurzia AI
Tuesday 3 November 2026 · Issue 5 · 4 min read
Squarespace cuts 19% in engineering; Anthropic's $895bn commitments; Musk admits Grok Bot loses money
Today's one thing
Squarespace — Squarespace's CEO told staff on Monday the company is cutting about 300 jobs, roughly 19% of headcount, with most reductions in product and engineering. Anthony Casalena framed it not as a cost exercise but as a reshape of how the company works: smaller teams and fewer layers. He wrote that the product and industry are undergoing a fundamental change the organisation must match, after a year in which Squarespace steadily added AI features.
This is the clearest case yet of a well-known software company cutting engineers while crediting AI, days after Microsoft offered buyouts to 60,000 people. Your staff will ask the obvious question; better to answer it with your own plan than let it be answered for you. Departing employees get at least 16 weeks of salary plus tenure-based weeks, benefits through May 2027, and keep their laptops. Business Insider
“This means smaller teams, more player coaches, and a simplified organizational structure.”
What is real, what is not
Anthropic's amended S-1 — The company going public before Thanksgiving filed its July-to-September numbers with regulators: about $17 billion of revenue, twelve times a year earlier. The quarter's $23.29 billion loss looks alarming but is mostly an accounting charge on convertible notes that disappears after listing; the number to hold on to is $895.4 billion in commitments, nearly all for computing capacity. SEC
Garry Tan — Y Combinator's chief says he spends $100,000 to $200,000 a year on AI agents and calls that a privilege most people cannot afford; he says cost is what keeps personal agents from going mainstream. A useful ceiling when budgeting: even the best-connected superuser in startups frames heavy agent use as an outlier expense, not a template. Y Combinator runs its own internal agent, Stray Light, with context on every founder and application it has ever handled. TechCrunch
Money and the market
Palantir — Quarterly revenue rose 96% year over year to $2.3 billion, ahead of estimates, with the full-year forecast lifted to about $8.5 billion. The figure buyers should note: US commercial revenue grew 148% to $984 million and, for the first time, outgrew US government revenue — companies are now outspending Washington on AI software. Shares rose about 1.5% in immediate after-hours trading after last week's AI selloff. Fortune
Salesforce–Cresta — signed a definitive agreement to buy Cresta, whose voice agents handle complex customer-service calls end to end for brands including Marriott, Comcast and Alaska Airlines, and coach human agents on live calls. If customer service is a cost centre for you, note where the big platforms are consolidating: voice, the channel companies found hardest to automate, is being folded into CRM suites. The deal is expected to close in the first quarter of Salesforce's fiscal 2028, and Cresta will keep supporting other contact-centre platforms afterwards. Salesforce
Marvell–Lumentum — Marvell is paying about $115 billion, mostly in its own shares, for Lumentum, whose lasers and optical parts carry traffic between AI computers. The physical layer of AI is consolidating fast; if you buy networking or data-centre capacity, expect fewer suppliers and firmer prices ahead. Nvidia put $2 billion into each company in March and is a major shareholder of both. TechCrunch
Regenerator — Henry Blodget breaks Anthropic's reported quarterly revenue into monthly steps and argues the company barely grew between July and September, with average monthly gains easing from about $760 million to about $580 million. He ties the slowdown to collapsing shares at compute-rental firms like CoreWeave, which borrowed heavily on the assumption that demand growth never moderates. Worth reading before the IPO roadshow noise; take the monthly estimates as informed guesses, not filings. Regenerator
“The end of all bubbles comes when supply begins to catch up with demand.”
Anthropic's growth debate — Reading the amended filing, analysts put third-quarter annualized revenue at $67.8 billion against $46.2 billion for Q2 — a 47% jump, but a sharp step down from the doubling pace of the first half. Compute and infrastructure costs rose 77% versus Q2, faster than revenue. Costs growing faster than revenue at the sector leader eventually lands in the prices you pay, so assume your AI bill follows their compute bill. Q3 contained only one week of Anthropic's newest model, released September 22, so the next report shows whether the slowdown sticks. Explainx (filing figures discussed on Hacker News)
Risk the board will raise
NYT v. Brave — The New York Times, the Daily News, the Chicago Tribune and the Orlando Sentinel sued the Brave browser in federal court in Manhattan on Monday, claiming it copied their journalism without permission and sold access to it. Anthropic, Perplexity, Mistral, xAI and Cohere are named as recipients; buying content through an intermediary is not looking like a defence. The complaint describes an index of over 30 billion pages gathered by a crawler that hid its identity, plus a feature for getting readers past paywalls. The Verge
China AI export controls — China's commerce and science ministries proposed adding 40 technologies to the export-control list, 23 of them in AI, and an interim registration rule for AI technology exports is already in force. Using a Chinese model through an app or API does not count as an export, but installing one on your own servers abroad does; companies self-hosting Chinese frontier models now face a licensing question. The draft is open for comment until November 15, 2026. Cloud Security Alliance
The big AI stories today
Grok Bot — Elon Musk confirmed weekly caps on Grok Bot were deliberate: serving it costs more than it earns, at roughly 800,000 paying users. The starkest public admission yet that agent usage can be priced below cost even at scale. TechCrunch
Home Depot — Home Depot says it will welcome agents that shop for customers once they can prove who they are, and is in talks with Meta over its Muse agent. How agents identify themselves at your digital door is becoming a real product decision for anyone selling online. Amazon barred Muse from its marketplace for not identifying itself properly. Business Insider
Alex Stamos — Meta's former security chief, now at Cognition, called handing agents financial logins insane; JPMorgan's Jamie Dimon warns liability shifts to the customer. If your product invites agents into customer accounts, expect banks and security leaders to push back hard. Stamos says the popular Instinct agent partly wins by bypassing safety protections meant to stop bots. Business Insider
One takeaway
Ask what each agent use costs, not just how much it is used. Yesterday's news agreed on one thing: the question has moved from whether AI demand is real to whether each individual use pays for itself. Vendors are now openly throttling agents that lose money per session, and survey evidence says returns show up only where a company's own data was ready first.
Before your next board meeting, ask the team for cost per use on every AI feature you run, alongside adoption figures — it is the number even the largest vendors are only now being forced to publish. It is also the number that decides whether an agent pilot survives its second year. TechCrunch