Swyx, affiliated with SMOL AI, Cognition and Latent Space
Swyx uses a parable-like exchange to examine repeated calls for the end of chat-based AI interfaces. He distinguishes what people publicly applaud from what they keep returning to. He compares that gap with people announcing that they will leave X while continuing to use it. The exchange suggests that users of the platform often seek novelty rather than truth. His focus is the distance between stated preferences and repeated behavior, rather than a claim that a new interface will inevitably replace chat.
Anthropic Claude Code developer Thariq
Thariq thinks many people multitasking with agents would benefit from slowing down. He asks them to identify their highest-priority P0 task and consider what giving it their full attention would look like, since much of the other work may not matter.
Vercel CEO Guillermo Rauch
Guillermo Rauch announces an open-source builder for producing HTML artifacts quickly. He says it is powered by libfx, which makes the agent durable. Building the underlying infrastructure for v0 to survive server crashes, unreliable models and function limits had required substantial effort. Those capabilities are now available as free open-source software. He also shows a floating container he built in a few minutes, emphasizing that the result is simply HTML.
Box CEO Aaron Levie
Aaron Levie argues that stronger models may increase, rather than eliminate, opportunities for domain-specific differentiation in applications. When models were limited, applications often did little more than return output through a custom prompt or limited RAG, making them simple wrappers. Agents that use tools and work for hours or days require workflow triggers, coordination, domain-specific data and model routing. Enterprise implementation also involves changing how work gets done, rather than merely asking employees to type into a chat box. These requirements create opportunities for specialized systems to address jobs that were previously impractical. He points to security, life sciences, healthcare, finance and manufacturing. He also draws a boundary: the closer a use case is to a model lab’s core focus, the harder this strategy becomes.
Y Combinator CEO Garry Tan
Garry Tan sees multiplayer agents delivered through group chat as a likely development. He points to Wabi as an implementation of that interaction model, focusing on how people use agents together.
FPV Ventures partner Nikunj Kothari
Nikunj Kothari shares an early experience with Wabi 2.0: the first version was beautifully designed but did not retain his attention, while the new version quickly engaged him. He distinguishes a single personal-agent conversation from multiple user-built persistent agent chats, finding the former difficult to compartmentalize as tasks accumulate and the latter too permanent. Wabi combines a main orchestrator with episodic chats for specific purposes. Those chats can be shared, such as a school-related conversation with his wife, and can be generated proactively. His examples include a New York trip chat created from a flight the system noticed and archived when the trip ended, and a similar chat for his 2025 taxes. He still finds it too proactive and sees room to improve the apps. He explicitly describes this as an unpaid user recommendation rather than an investor endorsement.
SPC general partner and Bevel Health cofounder Aditya Agarwal
Aditya Agarwal announces a Wednesday SPC event with Trae Stephens. He identifies Stephens as an early Palantir employee, an Anduril cofounder and a Founders Fund general partner. Stephens last visited SPC in 2022. Agarwal notes how much the world has changed since that visit. The post includes a registration link.