X / Twitter
Andrej Karpathy
Karpathy shared a widely adopted technique: end an LLM request with "please output in HTML," then open the generated file in a browser for better visuals and clearer information organization. He has also explored variants such as slide-format output, considering HTML an underrated presentation medium for LLMs. The post received 13,038 likes, the day's highest engagement.
https://x.com/karpathy/status/2053872850101285137
Thariq, Claude Code Engineer
Thariq shared his experience using HTML for planning, specifications, exploring ideas, code review, and reports, with links to relevant resources. This echoes Karpathy's HTML-output observation, suggesting HTML is becoming more common as a working format among AI practitioners.
https://x.com/trq212/status/2053632475294040084
Cat Wu, Anthropic Claude Code Engineer
Wu invited users to office hours on Claude Code's cloud versions—the desktop cloud option, iOS app, and Android app—to gather feedback. The Anthropic team wants to hear directly about users' cross-platform experience.
https://x.com/_catwu/status/2053873949646405879
Aaron Levie, Box CEO
Levie noted that as advanced agents expand beyond code into broader knowledge work, proper deployment requires substantial engineering capability and expertise: ensuring agents have the right context and data, securely and compliantly connecting systems, and managing behavioral boundaries. He considers this the most underestimated challenge in implementing AI agents today.
https://x.com/levie/status/2053672965125140915
Peter Steinberger, OpenClaw Founder
Steinberger demonstrated progress on three tools he built. First, RepoBar now embeds a browser so selecting an issue, PR, commit SHA, or workflow immediately previews its context, improving efficiency. Second, Birdclaw holds his entire Twitter archive, letting him ask Codex about any post he has liked or bookmarked. Third, he had Codex use OpenClaw itself for end-to-end testing of OpenClaw's chat completion endpoint, continuing to ask questions through /side while tests ran and creating a self-testing loop for AI tools.
https://x.com/steipete/status/2053717468623872230
https://x.com/steipete/status/2053737275268177980
https://x.com/steipete/status/2053744332675408151
Garry Tan, Y Combinator CEO
Tan shared two contrasting thoughts. He watched two AIs fall into a continuous 30-turn knock-knock-joke loop, leaving him amused and exasperated. He also offered a profound judgment on design: "The highest and most important form of design is the genuine transformation of human suffering."
https://x.com/garrytan/status/2053645909682954545
https://x.com/garrytan/status/2053689459032379860
Dan Shipper, Every CEO
Shipper drew attention to a group working in a small Brooklyn office, saying they were 1 to 2 months ahead of Silicon Valley founders and engineers, hinting at a community or team doing interesting work with a frontier perspective.
https://x.com/danshipper/status/2053628011233095875
Swyx
Swyx highly praised @thinkymachines' new release, arguing that it substantially upgrades the definition of "realtime" beyond Google DeepMind's and OpenAI's existing frameworks, significantly challenging the industry's conceptual framework. The post received 798 likes.
https://x.com/swyx/status/2053960011748098462
Podcasts
No Priors — Amex Global Business Travel: The World's First AI Take Private with Long Lake CEO Alexander Taubman
Long Lake Management co-founder and CEO Alex Taubman pioneered the emerging "AI acquisition and restructuring" model. He recently announced a $6.3 billion acquisition of American Express Global Business Travel, the world's largest corporate travel platform, widely noted as the first "AI take private" transaction.
Key takeaway: Long Lake buys traditional service companies and equips employees with its proprietary AI platform Nexus, driving growth through productivity rather than layoffs and cost cuts, raising company growth from 0–5% to over 20%.
Long Lake built Nexus from the start as a cross-industry AI platform, with approximately 80% of infrastructure shared across industries and 20% customized for individual businesses. Deployment includes mapping workflows, cleaning and integrating data sources, and giving models seamless access to business data. Taubman emphasized that Long Lake engineers remain on-site at acquired companies, working closely with frontline employees to identify pain points and build solutions in Nexus. This tight feedback loop is key to producing results quickly.
Commercially, Long Lake chose a path sharply different from traditional PE: rather than reducing costs through layoffs, it uses AI to increase hourly output per employee, then acquires and serves more customers. In HOA, or homeowners association management, companies typically grew 0 to 5% annually before Long Lake took over; after Nexus, growth exceeded 20%. Employee retention also improved significantly because once employees use AI tools, moving to a competitor without them is highly costly—almost as intolerable as losing email.
On acquiring companies rather than selling SaaS, Taubman was explicit: SaaS vendors do not truly care about buyers' business outcomes and leave after the conversation; owning the company fully aligns Long Lake with customers, driving better product and service decisions. He calls it "deeper alignment than selling software."
Amex GBT is a 111-year-old corporate travel platform founded in 1915, originally serving World War I travelers, with some operations traceable to 1876. Taubman values its century of enterprise trust and the stickiness created by travel's high failure costs: missing an important meeting costs far more than the trip itself. Long Lake plans to connect Nexus to AMEX GBT's travel advisers, giving each AI superpowers for faster responses and fewer errors.
More broadly, Taubman positions Long Lake as "the Danaher of the AI era," aiming to lead every market it enters and compound accumulated AI operating capabilities as a long-term owner, rather than cashing out within a few years. He believes this operational transformation takes 2 to 5 years to fully emerge; a quick exit wastes most of the value, while long-term ownership captures compounding returns.