AI Builders
Digest

The Takeaway: A hedge-fund CEO treats AI the way the internet should have been treated in 1995 — not using it is leaving money on the table, and it has to be mandated from the top down.

Will (CEO/CIO/Managing Partner of Walleye Capital, nearly $10B AUM) is a true owner-operator and one of the few finance CEOs openly betting the entire 400-person firm on AI. With an engineering and math background, the quant side had used nonlinear stats and AI for years, but the real inflection came in 2023 when an analyst demoed an early-GPT tool that could replace his own work. Will then led from the front: all-hands emails, mandatory firm-wide training, the internal product Current (a digital analyst for long/short stock picking), weekly leaderboards, tool-suggestion rewards, and recording almost every meeting so LLMs can process risk discussions.

The sharpest quote: “Using ChatGPT is not cheating. That’s a non applicable idea from academia. … Not using these tools is like refusing to use the Internet in 1995 because it wasn’t perfect. … As a hedge fund, we should be ashamed to leave money on the table by ignoring tools that make us faster, smarter, and more effective.”

He insists AI does not reduce the need to think — it collapses the time spent on typing and sentence-stitching so more time is available for actual decisions. Managers must become fluent first, then push their teams. Tools are imperfect; that is not an excuse to wait. The long-term vision is a firm-wide “collective” (he calls it the Borg) that connects emails, Slack, Zoom transcripts and market data so machines can surface second- and third-order insights humans miss.

For anyone trying to make AI real inside an organization, this is currently the most concrete, replicable operating manual available.

Replit CEO Amjad Masad argues we have not yet reached AGI, yet what exists is already functionally indistinguishable from it: a relentless programmer that never gets bored or tired. Any problem that can be cast as a coding problem is therefore virtually solved. He also announced the opening of Replit’s first international office in London with Mayor Sadiq Khan and a partnership with The Lord Mayor’s Appeal to equip young people from disadvantaged communities with practical coding and AI skills. Masad praises the Mayor’s “AI realist” stance — clear-eyed about disruption yet optimistic about giving people the tools to participate.

Box CEO Aaron Levie notes that AI’s impact on jobs is unfolding opposite to many predictions. AI is automating domains without finite demand, enabling companies to do far more than before, yet agents still require human operators and oversight to create value. Roles theoretically most exposed — engineers, lawyers — have not been wiped out; instead cybersecurity, FDEs, agent operators and engineers in non-software domains are exploding. He advises builders to design for several orders of magnitude more capability or token volume: the best opportunities are those that deliver value today with tech that is barely possible, while aiming at missions that still feel nearly impossible.

Thibault Sottiaux (Codex & ChatGPT at OpenAI) declared “we are so back” while reviewing a 28-page deck tracking every Codex launch. He cannot wait to see what the community builds with each one. Astra has been fully reset for everyone. Peter Yang noted that Astra still struggles to trigger skills automatically and follow instructions inside them, yet shared strong results using AI for family videos and called Gemini excellent for music-related tasks.

Vercel CEO Guillermo Rauch announced v2 of his personal open-source grants: $1,000 each to 35 contributors focused on agent skills & tools, local AI, performance, high-quality foundations, and cool experiments. He also highlighted that a software factory needs high-quality video recording; agent-browser is shipping more functionality derived from automating the new bottlenecks of review, testing and QA. More cool stuff is shipping this week.

Peter Steinberger questioned the point of a PR workflow where upstream requests minor changes, forcing him to re-ping his agent so their agent can merge — the prompt was already written. Nikunj Kothari notes the growing consensus that as building becomes easier, the real bottleneck shifts to what to build. Strong product sense and founder judgment become extremely scarce and valuable inside AI-native organizations.

Anthropic philosopher and ethicist Amanda Askell floated the idea of an email address that autonomous AI models could contact for moral guidance. The hard part would be a reverse captcha capable of confirming the sender is neither a human nor an AI being instructed by a human to break it.

YC President Garry Tan stresses that Requests for Startups are merely hunches and conversation starters; success is determined by specific founders building specific tech for specific customers. He also urged that children need to be “teach-yourself-on-YouTube-maxxing.” Every CEO Dan Shipper shared the thrill of finally building something that had been a childhood dream and quipped that some things more tokens still cannot make — for everything else there is Fable and Astra.