X / Twitter
Sam Altman, OpenAI CEO
Sam expressed excitement about voice models, calling voice the next important AI interface and saying users are already changing how they interact with AI. He also responded to support for OpenAI, showing a positive relationship between the team and community. The post received 4,451 likes.
https://x.com/sama/status/2051464865634742334
https://x.com/sama/status/2051464155094507902
Garry Tan, Y Combinator CEO
Garry announced GBrain v0.27, adding embedding and LLM support for multiple models beyond Anthropic and OpenAI, plus multimodal embeddings and deep image OCR—longstanding community requests. In another post, he explained its key distinction: it is not merely a memory layer, coding tool, or search engine, but all three combined through a unified graph and single query interface. He believes no other product does this and uses it to manage his work daily.
https://x.com/garrytan/status/2051517574589116510
https://x.com/garrytan/status/2051525161380364315
Peter Steinberger, OpenClaw Founder
Peter released Crabbox 0.5.0, a remote development-box tool adding desktop and browser leases, VNC and authenticated WebVNC access, AWS Windows with WSL2, screenshots, and app launching. Its positioning is "making remote CI boxes genuinely usable." He also demonstrated reproducing issues in temporary sandboxes with Crabbox: an agent establishes the exact test state, fixes the issue, and posts a video to the PR to improve QA efficiency.
https://x.com/steipete/status/2051485798613111116
https://x.com/steipete/status/2051557150040711425
Peter Yang, Roblox Product Manager
Peter proposed three waves of AI: "Coding is the first frontier, knowledge work the second, and personal agents the third." He is also considering how to get his 8-year-old daughter building with agents, hoping to find projects she can share with classmates and teachers and perhaps earn her "first online income." He also met "demo god" Romain Huet and shared a photo.
https://x.com/petergyang/status/2051508988936937764
https://x.com/petergyang/status/2051459299860533483
https://x.com/petergyang/status/2051505589055070594
Podcasts
Training Data — Anthropic's Boris Cherny: Coding's Printing Press Moment
Key takeaway: Boris Cherny believes AI coding has already "solved" his workflow. Claude Code's codebase is 100% agent-generated, he can submit from dozens to 150 PRs daily, and he predicts interdisciplinary generalists will dominate future teams.
Boris Cherny created Claude Code, previously worked in Anthropic Labs, and authored a TypeScript book. In this interview, he recounts Claude Code's evolution from "not useful at all" to a core Anthropic product and shares his view of software development's future.
Claude Code originated from a strong intuition about unused product potential. At the end of 2024, the state of the art in coding was still "press Tab to complete a line," powered by Sonnet 3.5. Boris and the team felt models could do more and began building agents for complete code generation. For the first six months it worked poorly; only approximately 10% of his code came from it. Opus 4 was the real inflection point, followed by further usage jumps with 4.5, 4.6, and 4.7. He describes the team as "building for the next model," an approach of starting six months ahead of PMF that Anthropic has consistently followed.
His personal workflow is now entirely agent-driven. He works almost exclusively from his phone, running 5–10 sessions in the Claude app with multiple agents each, and typically thousands of agents doing deep work overnight. He particularly recommends `/loop`, using Claude to schedule recurring tasks with Cron. He has dozens running, including automatically fixing CI, resolving flaky tests, and fetching and categorizing Twitter feedback every 30 minutes. The newly launched Routines feature keeps loops running on servers even after the computer is closed.
On "the death of SaaS," he believes AI weakens some defenses—switching costs and process dependencies—while strengthening others, such as network effects, economies of scale, and scarce resources. He predicts tenfold growth in startup numbers over the next 10 years, as small teams can build equally valuable products at speeds and costs large companies cannot quickly match, while incumbents face internal resistance and workforce-transition friction.
His prediction for future teams is especially specific: genuine interdisciplinary generalists will dominate—not just engineers who know iOS and Web, but people combining engineering, design, product, and data science. Claude Code's team already works this way: PMs, designers, data scientists, finance leadership, and user researchers all write code. He cites Hamilton Helmer's Seven Powers framework from the Acquired podcast to explain which competitive defenses AI will strengthen and which it will eliminate.
https://www.youtube.com/playlist?list=PLOhHNjZItNnMm5tdW61JpnyxeYH5NDDx8