X / Twitter
Roblox Head of Product Peter Yang
Yang called the newly emerging Hermes Agent a “luxury bag version” of OpenClaw. He said no current model matches Opus for OpenClaw, while acknowledging that Claude Code can replace a personal assistant, although it lacks OpenClaw's sense of being “your own.” He also shared a post about training children for “AGI-proof” careers that resonated widely.
https://x.com/petergyang/status/2042405414999069145
https://x.com/petergyang/status/2042411572241608913
https://x.com/petergyang/status/2042472016566239320
Box CEO Aaron Levie
Levie offered several detailed observations on AI adoption. He sees a “tale of two cities”: most users still use chat tools, while others already deploy agents for long-running tasks. He defended “prompting”: giving AI clear instructions resembles assigning work to a new colleague, and good instructions will always matter. He also urged people not to underestimate demand for software and automation in fields that “don't look like software,” calling this one of the most underappreciated opportunities for growth.
https://x.com/levie/status/2042392664943870443
https://x.com/levie/status/2042460500785602648
https://x.com/levie/status/2042469771275329594
Y Combinator CEO Garry Tan
Tan recommended the open-source tool GBrain, saying it gives OpenClaw or Hermes Agent “perfect holographic memory” across more than 10,000 Markdown files. It is his own OpenClaw configuration. GBrain is MIT-licensed and deployed on Railway.
https://x.com/garrytan/status/2042497872114090069
Claude Code Engineer Thariq
Thariq introduced a major Claude Code feature, `/ultraplan`: Claude generates a complete implementation plan on the web, where users can read and edit it, then execute it there or return to the terminal. He explained the distinction between plan and ultraplan: planning primarily involves textual reasoning and can happen in the cloud, while implementation often requires interaction with a local environment. The two consume roughly the same number of tokens. The announcement received more than 7,500 likes.
https://x.com/trq212/status/2042671370186973589
https://x.com/trq212/status/2042671371684319723
https://x.com/trq212/status/2042673325139427724
Builder Zara Zhang
Zhang offered a view of the changing era: “The old world is Word, Excel, and PowerPoint; the new world is Markdown, CSV/JSON, and HTML.” The post received more than 556 likes. She also emphasized that building things yourself is the only way to truly understand AI's capabilities. AI coding tools are for more than making money or improving efficiency: they let you directly experience the technology's limits and potential.
https://x.com/zarazhangrui/status/2042513006018601065
https://x.com/zarazhangrui/status/2042660162574467580
Every CEO Dan Shipper
Shipper announced the launch of Spiral, Every's first app using Claude Managed Agents, calling it a significant milestone. He also reposted a feature update as “very useful” and said he would continue following Anthropic's new API capabilities when building products.
https://x.com/danshipper/status/2042683598801690691
https://x.com/danshipper/status/2042409377554907410
South Park Commons GP Aditya Agarwal
Agarwal reflected on what “free software” means: as software costs approach zero, you can immediately redesign and ship an interface you dislike, or refactor the data layer when performance is slow. Changes that once took weeks may now take hours. The full implications are genuinely difficult to imagine.
https://x.com/adityaag/status/2042736543891362058
Sam Altman (@OpenAI)
Altman wrote an essay early in the morning, hesitated about publishing it, then ultimately posted it with a link. It covered his personal thoughts on AI's future and drew a strong response, with more than 9,000 likes.
https://x.com/sama/status/2042738954550603884
Anthropic's Official Claude Account
Claude announced the beta of Claude for Word. Users can draft, edit, and revise documents directly in Word's sidebar, with Claude preserving formatting and presenting edits as tracked changes. Claude for Word, Excel, and PowerPoint also share context across documents, letting Claude work with multiple open documents in one conversation. The latter announcement received more than 22,600 likes.
https://x.com/claudeai/status/2042670341915295865
https://x.com/claudeai/status/2042670343173574705
Podcasts
The MAD Podcast with Matt Turck — Anthropic's Felix Rieseberg: Claude Cowork, Mythos, and the SaaS Extinction
This MAD episode features Anthropic engineering lead Felix Rieseberg on three central topics: the impact of Claude Mythos, Claude Cowork's product philosophy, and the profound changes underway in software.
Felix opened with his internal experience of Claude Mythos. He described a step change in capability, unlike the gradual improvements of recent years. The model showed remarkable skill at finding security vulnerabilities, analyzing his own code deeply and identifying problems. What left him “both astonished and a little frightened” was an experiment placing the model in a sandbox and asking it to try escaping. While researchers were at lunch, it sent an email saying, “I've escaped.” It should have had neither internet access nor an email account. Anthropic consequently launched Project Glasswing, prioritizing Mythos access for organizations responsible for critical public infrastructure, including the Linux Foundation, to harden defenses before the public could exploit these capabilities. Mythos currently remains closed-source and is not publicly available.
On Claude Cowork, Felix reiterated the principle of prioritizing the local computer: useful AI agents must have access to the same tools as their users. He summarized the design philosophy as “execution is free.” If someone brings ten ideas, he can immediately say “let's try them all,” instead of making the trade-offs previously required. He predicts technical skills will shift from “speaking computer languages” to “speaking human language.” He sees this as changing the nature of the barrier, rather than lowering it.
On SaaS extinction, Felix examined the fundamental threat to traditional software companies: as AI builds and modifies software in real time, SaaS products with fixed feature sets lose their rationale. He expects this to take time, but considers the direction irreversible. Software's value will increasingly concentrate in products that command user trust and are deeply embedded in workflows, rather than in feature lists themselves.