TL;DR
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.
He says the big mistake is treating task automation like job elimination. In coding, legal, sales, and marketing, AI will mostly let people do more, do it better, and serve customers who couldn’t afford that level of work before — which means the job grows, not disappears.
He built the GitHub dashboard he always wanted: one view for repos, open issues/PRs, last release, and commits since release — then shipped it for everyone. He also suggests a practical Codex workflow: keep a scratch log of decisions, tradeoffs, and fixes so you can audit what the agent actually did later.
A user says Replit let them build Dial’s MVP in a single weekend and get Apple approval on the first try — a pretty brutal benchmark for “fast.” It’s the kind of proof point that makes the AI app-builder pitch feel less like hype and more like a real workflow shift.
He says the shift from chatting with models to having them produce more code than humans have written in total happened in just six months. The point is blunt: AI moved from novelty to actual software output fast, and founders should treat that as the new baseline.
He says he fine-tuned a Qwen3.5-397B model in a couple hours and thinks fast, usable multimodal models are about to unlock genuinely mind-blowing personal AI. He also called Thinking Machines impressive, which is a pretty strong signal from the YC president who lives and breathes this stuff.
He says more B2B companies are finally waking up to the power of narrative, and that waiting another year is a mistake. The point is blunt: in a crowded market, vibes and positioning matter a lot if you want to stand out.
Cerebras won by betting on speed before the market cared
The Takeaway: Cerebras bet that AI speed would matter before anyone else did—and that patience, not hype, would decide the winner.
Key Insights
The Story
Andrew Feldman, cofounder and CEO of Cerebras, has spent more than a decade building what he calls AI computers—machines optimized for inference and training, not general-purpose computing. The company’s wager was contrarian from day one: wafer-scale chips, a 46,000 square millimeter design the size of a dinner plate, when everyone else was building postage-stamp silicon. “They told us we were out of our mind,” he says. “It would never work.”
It almost didn’t. Between 2017 and 2019, the team burned roughly $8 million a month trying to make the thing work, with board meetings every six weeks asking the same painful question: why isn’t it built yet? Then the chip finally yielded, and the company spent the next few years in an awkward place—technically ahead, commercially ignored. Feldman’s read is simple: “When it’s a novelty, nobody cares that you’re fast.”
That changed in 2025, when AI became embedded in real workflows and speed turned into a hard requirement. Cerebras was suddenly in demand across models and customers, from supercomputing labs to OpenAI and AWS. Feldman’s philosophy is blunt: love the hard road, hire fearlessly, and don’t let a company drift into safe mediocrity. His favorite line captures it: “I’m a professional David.”
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 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.
Altman shipped Codex while asking what AI should fix. Levie saw AI pricing split into tiers, Masad pushed buy-first rewards, and Anthropic added Auto Mode to Claude Code as the reliability bar jumped.