AI Builders
Digest

The Takeaway: In the AI era the real alpha is not founding startups from scratch in every industry, but acquiring the right incumbents and refounding them with frontier engineering under permanent ownership.

Sequence Holdings Co-Founder and CEO Michael Lee (former Lone Pine and Apollo investor) built a permanent holding company for one purpose: partner with exceptional management teams, buy scaled businesses that already own brand, network effects or regulatory moats, and re-architect them into market leaders with AI. They just closed the $7.7 billion take-private of insurance broker Baldwin with the Dell family office—the largest AI-driven take-private to date.

Lee’s core thesis is that AI’s impact on the economy is highly uneven. Restaurants and golf courses stay largely untouched; coding services will be eaten by Cognition or Anthropic; but industries like insurance brokerage and community banking give incumbents decisive advantages. Rather than fight those advantages, Sequence inherits them and injects world-class engineering.

Their platform Atlas decomposes any business into a data ontology, agent builder, orchestration engine (Lattice) and application layer (artifacts). At community bank BankSouth they cut consumer underwriting time by 94% and average loan cycle from 30 days to 11 days, letting the same team handle double the volume. Lee’s sharpest line: “Every company on the planet has a celebrated persona. In a world where you believe that alpha comes from engineering and AI, you need to create a culture whereby the celebrated persona is the engineer.”

They aim for one deal per year. No deployment pressure, only long-term compounding. For traditional PE this is almost a different species.

OpenAI Codex & ChatGPT team member Thibault Sottiaux announced that over the next 28 days the team will either ship one clear improvement relevant to most Codex/Work users each day or perform a full reset. The improvement cycle has begun.

Practical AI tutorial creator Peter Yang is glad OpenAI is simplifying ChatGPT because it has become a mess. He would start with Work vs Codex, Spaces vs Pages vs Sites, and the model/effort picker. His hot take: the entire Work launch was a mistake—just as Claude folded Cowork back into Chat, Work does not need its own brand.

Box CEO Aaron Levie notes that deploying AI requires significant technical work and surrounding services, creating roles for AI engineers building applied products, FDEs deploying agents, and new AI services firms. Existing data, research and software jobs inside large enterprises are also being repositioned. Banks, life-sciences companies, manufacturers and even law firms are adding or shifting technical talent for agent deployment. It is easier to picture what AI replaces than what it creates—until it starts happening.

Vercel CEO Guillermo Rauch agrees with DHH on Rust. Vercel ported Turborepo from Go to Rust; the ROI was controversial because human migration cost was high while Go was already fast and pleasant. The calculus has changed: what is best for humans is no longer necessarily best for business. He also doubts Rust is the final toolchain, since it was designed before the agent tsunami. Separately he notes that READMEs should be written by hand for humans while internal docs can be AI English for agents.

Peter Yang quotes Granola co-founder Sam: there is a pang of sadness as a UI designer when users skip the product and just use the MCP. They fought to keep everyone inside the Granola UI, but once you reach enterprise scale, internal tooling is inevitable and non-integration is a non-starter. They are now at peace that for many workflows the best path is capture context in Granola then use it through the internal agent. Sam also notes that the easier it gets to build, the more it matters to sweat the details and ensure the product delivers daily value, not just a shipped v1.

Builder Zara Zhang highlights how Anthropic Opus 5.5 will silently work for 20+ minutes and return with a complete masterpiece. She also notes that X profiles are replacing traditional resumes.

Every CEO Dan Shipper says dot has become his primary AI interface over the last few weeks (his is named boo). He bets that in a year he will still interact with ChatGPT as an always-on persistent agent, but the specific personality of boo will have disappeared—just as people loved the quirks of early OpenClaw agents and dropped them the moment something functionally stronger appeared.

Y Combinator President & CEO Garry Tan observes that when everyone is building the same primitives it signals needs that will only intensify, and we will eventually converge on the correct OS.

FirstMark VC Matt Turck highlights that one of the most poorly understood aspects of AI outside the field is that “models are grown, not built” and we do not really know how they work—hence the urgent need for better interpretability.

FPV Ventures partner Nikunj Kothari cannot believe people make investments over Zoom or after meeting only the CEO. There is major alpha in visiting the founders’ own office: you see the energy, how teammates work together, how co-founders complement each other, and what the real beta looks like. Even under time pressure he still asks for final meetings at their office; founders often say he is the only one who ever requested it.

Box CEO Aaron Levie notes that a widely discussed consumer-AI metric may tell us more about the business model of consumer AI than the state of AI itself. Consumer AI will almost inevitably be subsidized and monetized via commerce, ads, or device and service purchases, so the metric may level off lower than many expect.

Former Cursor, Notion and Stripe designer Ryo Lu shared a further refined interface: full-bleed desktop plus side dock—essentially a computer in your pocket.

OpenClaw-related builder Peter Steinberger continues shipping bug fixes and performance improvements.