$2B seed, $12B valuation: Thinking Machines Lab's new record

Mira Murati’s new AI startup, Thinking Machines Lab, closed a $2 billion seed round led by Andreessen Horowitz, valuing the company at $12 billion despite having no product or revenue. Investors include Nvidia, Accel and Cisco, and two‑thirds of the team are ex‑OpenAI staff. The cash will fund a first product launch and open‑source tools for researchers.

PE‑backed SaaS firms are hitting a hidden collapse point when payment‑in‑kind (PIK) debt matures. Medallia’s $6.4 B buy turned into a $5.1 B equity wipeout, and a dozen similar deals sit on the same fuse. The analysis flags which companies are already down, which are ticking, and who’s next to implode.

Hybrid “dequity” loans, part debt, part equity, have surged to $30 billion since 2023 as private‑equity firms grapple with the deepest exit drought since 2009. Direct‑lending arms at Ares, Neuberger Berman, KKR and others are offering these costly lifelines, a response to $3.2 trillion of unsold assets that threaten fund liquidity.

A 23‑year‑old founder thought a Starbucks pitch to Sequoia would seal a deal, only to see a private chat message accidentally broadcast onstage, nearly blowing the round. The mishap underscores that Sequoia values substance over swagger, and missing this cue led them to back WhatsApp instead of GroupMe.

Cognition, the maker of the Devin AI coding agent, bought AI‑IDE startup Windsurf, inheriting its 250‑person team and $82 million ARR after Google only poached the leadership. The deal consolidates two AI coding platforms under one roof, positioning Cognition to compete with Cursor and expand its enterprise offering.

Brandur explains that the most viable product is a single, self‑contained piece of software that delivers clear value, not a sprawling suite. By pricing and packaging around this minimal unit, founders can launch faster, reduce overhead, and avoid the endless build‑vs‑buy trap that AI tools only partially solve.

Vertical AI founders who simply ask executives for advice can turn those conversations into design partners. A sweet spot of five to ten partners, mixing cold outreach with warm introductions and targeting marquee names, creates credibility and early‑stage feedback without a hard sell.

Fiona Fung, who runs Anthropic’s Claude Code and Cowork teams, reveals how AI tools let her engineers ship eight times more code per quarter. She explains which roles AI will reshape next and how to keep culture intact as agents blur job boundaries.

Paul Graham shows that in startups, science, and art, performance rewards grow faster than linearly: the top 1% create vastly more value than the next 99%. This means talent selection and focusing on exponential growth rates are decisive for founders, turning modest effort into massive payoff or total failure.

Fiona Fung, who runs Anthropic’s Claude Code and Cowork teams, reveals how AI tools let her engineers ship eight times more code per quarter. She explains which roles AI will reshape next and how to keep culture intact as agents blur job boundaries.

Ethan Thornton dropped out of MIT, raised $300 M Series C and now runs six defense projects, from a vertical‑takeoff strike aircraft to a cheap drone interceptor, simultaneously. He argues that spreading effort is essential to out‑create China, targeting operational deployment of three systems and rate‑manufacturing by year‑end.

Shopify CEO’s AI memo turned into an open‑access policy: every employee can run any model without budget caps. The move pushed non‑engineers, especially support teams, to spend tens of thousands on advanced tools, and leadership celebrated the unexpected ROI. The case shows how removing gate‑keeping fuels rapid, company‑wide AI adoption.

AI agents are redefining the old "toothbrush test" for software by proving that occasional, high‑impact interactions can make a product indispensable. This "magic minimum" means tools that solve rare but painful problems deliver lasting value, opening a new class of viable, low‑frequency AI businesses.