Swyx, affiliated with SMOL AI, Cognition and Latent Space
Swyx distinguishes experience with data-intensive applications from a role focused only on RAG, arguing that the gap in demand is widening. He summarizes the hiring emphasis as “Data engineers > AI engineers.” He invites developers whose blogs or GitHub profiles demonstrate that work to contact the Latent Space team. San Francisco is preferred, with relocation and visa support possible. The contact address is jobs@latent.space, and the team will assess fit.
OpenAI Codex and ChatGPT team member Thibault Sottiaux
Thibault Sottiaux says his team worked with OpenCode to get users who are also OpenAI API customers back in business. He says this included refunding any credits that might have been drawn during the attack. He also recommends requesting Daybreak Blue access in advance. He describes it as a cyber-permissive setup for defense. His message is not to wait until an incident before seeking help.
AI tutorial and interview creator Peter Yang
Peter Yang argues for making work trials the norm in hiring rather than relying only on interviews. He quotes Anthony Pompliano’s example: a team chose a feature it genuinely wanted and gave a candidate two weeks to build it. The candidate asked questions, learned the context and delivered the feature. Pompliano sees that as more direct evidence of ability than spending four weeks trying to assess fit. The team also shipped a real feature while evaluating the candidate. Yang uses the case to support assessing people through actual work.
Anthropic Claude Code developer Thariq
Thariq finds that he now often prefers vibe-coding small parts of games to playing them, especially games that require thought. He keeps returning to a possible hook: making the game and its characters could itself be the main attraction. He imagines distributing the game as open source, letting players vibe-code their own characters and having the game balance them. He is also exploring whether characters might fight autonomously, as in Salty Bet, or whether combat could offer cinematic choices like Persona or Heavy Rain. He presents these as open questions rather than settled mechanics. He plans to share more as he works them out.
Vercel CEO Guillermo Rauch
Guillermo Rauch tried the new deepseek-v4.1-flash-1017-fast through AI Gateway to improve a game project. He says it added sound and better collision detection. It also improved the ramp and made the camera follow the car. He estimates that the changes cost about 30 cents. This is his account of one build session, not a cost or performance guarantee for other tasks.
Box CEO Aaron Levie
Aaron Levie argues that agent swarms matter because they can make compute a critical variable in many more economic tasks. Research, data processing, discovery and testing have been constrained by people’s capacity to plan, execute, review and coordinate, leaving much less work done than he believes its economic value warrants. Once models are sufficiently capable, swarms could parallelize that work. His examples include cybersecurity, life sciences and materials research, trading, QA testing and data processing. He imagines agents that use tools, write code, process data, have domain depth and coordinate with one another tackling more complex problems through clear goals and more tokens. His claim that compute could become the only parameter “in theory” depends on models first being good enough at the tasks. He expects this capability to change the competitive dynamics of those fields as well.