Latent Space team member Swyx
Swyx is recruiting writers, reviewers and reporters for Latent Space’s new editorial team, directing interested people to the ricmac and adlinzainal accounts he tags. Part-time contributors and one-off guest posts are welcome. He plans to put a version of the linked material into the welcome package for new hires, without describing that material in the post. The two updates explain how to contribute but do not specify compensation or job locations.
Practical AI tutorial and interview creator Peter Yang
Peter Yang shares his experience of using the Once reading app with his four-year-old for several months, calling it the only learning app they have managed to stick with. He credits the daily habit of a parent learning alongside the child. Lightweight gamification also helps, such as taking a silly photo together after a lesson. After 30 lessons, Once sends books the child can read independently, creating a strong sense of accomplishment. He says his child now sometimes chooses the Once books over YouTube or a phone, and explicitly states that the post is not sponsored. This is one parent’s experience, without comparative outcome data from other children or courses.
Vercel CEO Guillermo Rauch
Guillermo Rauch enjoys the engineering vocabulary models use, including gate, evidence, culprit, durable, provenance and contract. He connects it to the fact that much engineering consists of debugging and understanding existing systems, where people act as detectives or auditors. He asks how much English is actually necessary to work productively, noting that ASD-STE100 Simplified Technical English contains roughly 900 approved words. He suspects fewer might suffice but offers no validation. In another post, he describes the balance of guiding AI: people can over-guide a model that knows better, or under-guide one that goes off track, carries the wrong biases or lacks context. Managing that capability is the top skill he looks for. The posts concern the cost of expression and the degree of guidance in human–AI collaboration, without proposing a universal procedure.
Box CEO Aaron Levie
Aaron Levie predicts that cyber agents may soon process 1,000 times as many tokens as they do today. He argues that the expanding attack surface already strains large companies, while most businesses have even fewer resources to keep up with new risks. With so many libraries, tools, infrastructure components and codebases to secure, agent swarms could run security tools in parallel, test attack permutations, read logs and assess enterprise data for risk. He links the opportunity to recent progress in long-running agents, computer use and cyber-focused model training. He sees openings across identity, networking, firewalls, code scanning and data protection. This also gives him some optimism about the balance between offense and defense; the 1,000-fold figure is his forecast, rather than measured industry growth.
Ryo Lu, a designer who has worked at Cursor, Notion and Stripe
Ryo Lu reflects on learning programming and design through repeated attempts, failures, taking things apart and asking people for help. Those difficulties gradually became judgment: he could sometimes sense that something was wrong before he could explain why. AI has helped him make things he might never have made otherwise and removed tedious work, but he worries that eliminating difficulty before understanding what it teaches can reduce learning to obtaining a result. A finished-looking result makes it easier to miss the questions one never learned to ask. Human collaboration has a similar difficulty, as different histories, priorities and rhythms need time to align. Shared struggle gradually teaches a team who notices its blind spots and who can challenge an idea without killing it. He worries that an endlessly patient, always-available agent makes it easier to avoid the beginning of human relationships, leaving each person in a small world at a screen. Those worlds also run on company-owned servers and are shaped by interests people may not see. He does not consider pain inherently valuable; he wants tools that help people enter the world, share unfinished work and solve problems together. The difficulty he hopes to retain is that of learning and creating with other people, rather than obtaining a good-enough result alone.
Builder Zara Zhang
Zara Zhang observes that giving advice to other people is much easier than applying the same advice to yourself. The post supplies no example or method, so it remains a brief reflection on the gap between advice and practice.
FPV Ventures partner Nikunj Kothari
Nikunj Kothari wants to meet helpful angel investors interested in an infrastructure company’s Series A. In a separate post, he describes his focus as evolving from context to harnesses to judgment and invites people building in that space to discuss product ideas. He links to an explanation but does not detail a product mechanism or the company’s financing terms in the post.
SPC general partner and Bevel Health cofounder Aditya Agarwal
Aditya Agarwal says that becoming deeply immersed in AI has made him treasure in-person time with family and friends more. He encourages dancing with friends and taking children to the park, whether agents keep working or not. The point is to celebrate human relationships; the post does not turn this personal reflection into a productivity measurement.