TL;DR
Karpathy said to ask for HTML, not plain text. Cat Wu showed Claude Code with a terminal control plane, Swyx said realtime got a brutal upgrade, and No Priors argued AI rollups win by making service firms scale like software.
He says the easiest win with LLMs is to ask for HTML, then open it in a browser — suddenly the model can give you layout, interactivity, even slideshow-style output. Bigger picture: he thinks AI interfaces are moving from text to markdown to HTML, and eventually to interactive visual simulations, because vision is the natural output channel for humans.
They’re pushing `claude agents` as a control plane you can run from your terminal, with `<-` letting any CLI session register back into it. The pitch is simple: manage all your Claude Code agents from one root directory, instead of juggling them across scattered shells and sessions.
He says @thinkymachines just "framemogged" Google DeepMind and OpenAI, and that everyone’s definition of realtime got a massive upgrade. The subtext: whatever they shipped is pushing the bar hard enough that even the big labs look slow by comparison.
AI rollups win by making service firms grow like software
The Takeaway: Long Lake’s bet is that AI isn’t mainly a cost-cutting tool—it’s a growth engine that turns sleepy service businesses into compounding machines.
Key Insights
The Story
Alex Taubman, cofounder and CEO of Long Lake Management, is building an AI-native acquisition platform aimed at services businesses that have been under-optimized for decades. Long Lake has already completed around 30 acquisitions and recently announced a $6.3 billion take-private of American Express Global Business Travel, a 111-year-old franchise Taubman sees as a perfect fit for long-term AI transformation.
His philosophy is simple but sharp: buy businesses where AI can change the operating model, not just automate a few tasks. Nexus sits between the models and the business workflows, and Long Lake’s engineers work directly in the field with employees to find pain points and build tools around them. That tight loop matters because, as Taubman puts it, “AI is very, very underpenetrated”—and adoption only works when change management is built in.
The payoff is not just efficiency. In Long Lake’s HOA business, growth has reportedly jumped from 0–5% to 20%+ annually. The logic is that if AI makes teams 30–40% more productive, service firms can scale like software companies: higher margins, faster growth, better retention, and more attractive economics. Taubman’s long game is to be the best owner in each industry, then compound that advantage for decades.
Peter Steinberger called AI code review the killer dev workflow, while Thariq said Claude Code turned files into a workbench. Cursor bet on specialized models plus brutal RL infra, and Anthropic detailed how it contained Claude across products.
Steinberger said AI agents need lean skills, not bloated prose; Levie argued AI won’t kill jobs, it’ll raise the bar. Yang said Codex tested itself while Claude still owned frontend, and Garry Tan pushed hard evals as the way agents got better fast.
Levie warned AI demos fool CEOs; the hard work starts after. Tan said high agency plus taste beat everything, while Yang urged builders to build the system that builds the MVP first. LeCun kept the pressure on: LLMs help, but they’re not the road to real intelligence.
Levie said AI kills tasks, not jobs, so roles just expand. Masad said Replit’s AI stack can ship MVPs in a weekend, while Tan showed Qwen fine-tunes in hours and personal AI got real. No Priors said Cerebras won by betting on speed before the market cared.
Levie said AI will surface more bugs, making humans the bottleneck. Steinberger argued GitHub’s native PR limits kill bot policing, while Tan pushed the “ship the 60%” rule and Swyx said AI coding should harden codebases, not just move faster.