X

Original signal: Linear’s head of product Nan Yu captured a new default in software development: everyone pair programs now, with a robot. The point is less about any single IDE plugin and more about a shift in collaboration norms. AI coding assistants are moving from optional helpers to default teammates, which will reshape workflow design, review expectations, and how teams decompose work.

Original signal: Replit CEO Amjad Masad endorsed what he called an inspiring positive-sum vision for enterprise AI. In the context of this week’s broader discussion about learning loops, the signal is clear: enterprise AI becomes strategically valuable when it compounds people, data, processes, and agents together, rather than being framed only as task or job replacement.

Original signal: Vercel CEO Guillermo Rauch surfaced two infrastructure signals. Starlink-enabled flights are making connectivity feel like a core travel experience, while the OpenAI ecosystem’s skills directory has passed 700,000 skills through organic community growth. Together they point to the same pattern: infrastructure is increasingly experienced directly by users, and the supply side of AI/agent tooling is expanding quickly.

Original signal: Box CEO Aaron Levie framed enterprise AI around IP, institutional knowledge, and architecture. The companies best positioned for the future will be those that can put unique data, business knowledge, and processes into systems that learn over time, rather than merely calling a model API. He also argued that model-layer availability risk will push more countries and businesses toward open weights and sovereign AI, because dependence on a single country or vendor becomes a real operational risk if access can be with…

Original signal: YC CEO Garry Tan highlighted both organizational sovereignty and the next generation of AI-native skill. Open source, in his view, is the escape hatch that lets businesses control their own destiny over the long term. He also argued that the next generation of world-changing young people will be those who can run long-running, multi-stage, multi-team agent tasks extremely well and at high volume. The skill frontier is moving from prompting to orchestrating durable work.

Original signal: Zara Zhang offered a practical counterintuitive rule for building skills: you do not start by writing a skill; you do the work, fix it twenty times, and then bottle up what you learned. A good skill is less a design artifact than a distilled record of real execution, repeated fixes, and edge cases. That explains why polished skills written upfront often perform worse than ones extracted from lived workflows.

Original signal: Peter Steinberger shared a concrete but useful remote-work tip: on unreliable in-flight internet, Mosh plus tmux or zellij can be a lifesaver. For builders increasingly dependent on remote agents, long-running tasks, and cloud development environments, resilient sessions matter because productivity loss often comes from connection drops and broken context, not from the editor itself.