TL;DR
Claude taught agents to grade themselves and remember, while Peter Yang said AI products have to ride the next model wave. Zara Zhang and Training Data both argued coding turned into literacy, with agents doing the overnight grind.
Claude says its new Outcomes feature lets you define a rubric, have a separate grader score the result, and keep the agent iterating until it hits the bar. It’s also teasing Dreaming, which mines past sessions for patterns and turns them into memory so agents improve over time.
Dario and Daniela’s take is blunt: build for the exponential, because some products only become viable as models improve. The real action is shifting from chatbots to agentic and Claude Code-style workflows, and even software engineering itself is starting to look like a stepping stone to growing a business.
She highlights Boris’s take that coding is becoming the new literacy, with thousands of agents running overnight and Claude Code now used mostly on his phone. The bigger point: software work is shifting from hands-on typing to steering fleets of agents, which is a pretty wild change for builders.
Coding is becoming a general skill, not a specialist one
The Takeaway: Boris Cherny thinks the real shift isn’t “AI writes code” — it’s that coding is turning into a universal, domain-first skill.
Cherny, the creator of Claude Code at Anthropic, comes off less like a futurist and more like someone already living in the next workflow. He says the tool started “accidentally” inside Anthropic Labs, and for months it barely worked. The bet was simple: build for the next model, not the current one. That patience paid off once Opus 4 hit and usage inflected hard.
His bigger claim is sharper: coding is basically solved for him, and the remaining bottleneck is no longer syntax but context. “The best person to write accounting software,” he argues, “is not an engineer, it’s a really good accountant.” In other words, domain expertise is becoming more valuable than raw programming skill.
That logic extends to teams and business models. He expects more generalists, but not just engineers who can do web plus mobile. He means people who can span product, design, data, and code. On moats, he thinks AI will make switching costs and process power weaker, while network effects and scale still matter. And he’s blunt about the pace: software is headed toward the same kind of democratization reading and writing got after the printing press.
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.