X / Twitter
Sam Altman, OpenAI CEO
Altman had two noteworthy posts this week. He announced Codex's arrival in the ChatGPT mobile app, receiving more than 7,700 likes and bringing OpenAI's core agent tool to hundreds of millions of mobile users. He also linked to an announcement of further new features. Separately, he reflected that fatherhood has exceeded his expectations more than anything else in life, surpassing even his already high hopes; the post received more than 2,600 likes.
https://x.com/sama/status/2055034461591588916
Aaron Levie, Box CEO
Levie commented on the growing Forward Deployed Engineer role, urging university career centers to help students understand these positions and how to obtain them. He said they require deep technical ability, typically a CS background, alongside strong business understanding and the ability to serve enterprise customers directly, making them one of the engineering paths with the greatest growth potential in the AI era. The post received more than 760 likes.
https://x.com/levie/status/2054729966630441007
Guillermo Rauch, Vercel CEO
Rauch shared two practical updates. Notion's developer platform is built on Vercel Sandbox, letting users seamlessly extend Notion or bring it into workflows through MCP; he also highly praised its landing-page design. He demonstrated the terminal command `npx ai-cli image '...'`, which renders AI-generated images directly in the terminal and accesses all major image, video, and text models through Vercel AI Gateway. The post received more than 450 likes.
https://x.com/rauchg/status/2054734851086155778
https://x.com/rauchg/status/2054989456189648918
Dan Shipper, Every CEO
Shipper shared a vivid street-level observation from 2026: while discussing Codex with his barber, the barber replied that he used Claude for crypto-trading strategies and Gemini plus nano banana to learn jiu-jitsu moves. A customer in the next chair interjected that their company had just bought 1,300 Claude seats. The post received more than 250 likes. He also shared his central criterion for whether an organization has embraced agents: the most important leading indicator is whether leadership personally uses Codex, Claude Code, or Cowork in daily work.
https://x.com/danshipper/status/2055030691180503152
https://x.com/danshipper/status/2055015466054410400
Aditya Agarwal, South Park Commons Partner
Agarwal reposted an article that resonated widely, receiving more than 950 likes without adding specific commentary. The high engagement suggests the content closely relates to his longstanding interest in AI entrepreneurship.
https://x.com/adityaag/status/2054976557702066520
Nikunj Kothari, FPV Ventures Partner
Nikunj identified a major ongoing shift: large companies with data advantages are moving to headless architectures and opening access because agent usage will be 100 times human usage. Companies continuing to design closed systems for humans will shrink in value by a factor of 100. He also commented on an unwritten rule of VC: large funds cannot afford to miss any important category, so they inevitably invest in competitors simultaneously.
https://x.com/nikunj/status/2054924517135540320
Peter Yang, Roblox Product
Yang shared an understated, practical use for Claude Code and Codex: merging and editing PDFs. Working with PDFs in Preview or Adobe Acrobat is painful, while asking AI is highly efficient, including detailed operations such as cropping blank margins from scanned PDFs. He also commented on a product announcement discussed by developers, emphasizing that communication should state constraints candidly and directly rather than vaguely.
https://x.com/petergyang/status/2054736772312293466
https://x.com/petergyang/status/2054767559933927737
Builder Zara Zhang
Zhang announced that she will co-host an event in San Francisco on May 23 about "managing and curating context for agents." She invited experienced practitioners to give 5-minute lightning demos of approaches that actually work in real workflows.
https://x.com/zarazhangrui/status/2054981832408760782
Podcasts
The MAD Podcast with Matt Turck — Why Every AI Agent Needs Its Own Computer | Ivan Burazin (Daytona)
First Mark partner Matt Turck speaks with Daytona CEO Ivan Burazin about why AI agents need their own computers, or sandboxes, and Daytona's product thinking in this infrastructure category.
Ivan's central argument is that agents are "digital knowledge workers," and knowledge workers need computers to work effectively. Simple questions may not require tools, but accessing real data, executing code, reaching the network, or operating a file system requires a dedicated machine. He defines a sandbox as "a composable computer designed for AI agents," combining an isolated secure environment with full computing capabilities.
Ivan illustrated the need through personal experience: when he asked a Claude agent to retrieve banking data, it replied, "Sure, please give me your login credentials." This immediately highlighted the risk of putting agents on his own computer. His solution was to give the agent its own Daytona account, separate phone number for 2FA, and credit card with a spending limit—like equipping a digital employee with a complete but restricted set of work accounts. In his framing, even if compromised, the worst outcome would be data exposure rather than financial loss, and he could always "turn off the machine."
On stateful versus stateless systems, Ivan noted that traditional clouds such as AWS, GCP, and Azure were built to deploy stateless applications, separating code from application state. Agent work is inherently stateful: agents need to install tools, retain work context, and persist memory across tasks. This is why existing cloud infrastructure cannot simply be used directly for agents, and why Daytona rewrote its scheduler rather than reuse Kubernetes.
Ivan also credited the OpenClaw Mac Mini example with educating the market: "People saw it and said, oh, I get it—an agent needs a computer, like a Mac Mini." He believes this concrete analogy substantially lowers the barrier to understanding sandboxes. He expects every agent to need at least one sandbox, sometimes several, and warned that global CPU supply could tighten sooner than expected.
https://www.youtube.com/watch?v=kMXJrzAa5fM
No Priors — Pax Silica: Inside the Trump Administration's Tech Strategy with US Under Secretary of State for Economic Affairs Jacob Helberg
No Priors speaks with US Under Secretary of State for Economic Affairs Jacob Helberg about Pax Silica's core framework and latest developments.
Helberg explained that Pax Silica is an economic-security alliance currently involving 14 countries, primarily aiming to secure AI supply chains through an ecosystem-based approach. Unlike the United States' traditional emphasis on isolated policy interventions, it seeks distributed supply-chain resilience through allied cooperation.
The latest development is an agreement with the Philippines, which is providing 4,000 acres—approximately one-third of Manhattan's area—for the US State Department to take over as diplomatic property, providing embassy-level legal protection. Over the next two years, the parties will negotiate investor protections, taxation mechanisms, and a long-term legal framework, aiming for arrangements lasting decades. The goal is to bring production of key AI supply-chain components—including precision reducers, server motors, rare-earth magnets, and actuators—into the allied ecosystem.
Helberg stressed that while many think AI supply chains mean chips alone, they actually involve thousands of component inputs, where US concentration risk is extremely high. The Philippines became the first implemented pilot because of its deep alliance with the United States, domestic manufacturing base, and potential robotics-supply-chain contribution.
Asked how it differs from China's Belt and Road Initiative, Helberg identified the model as the central distinction: the United States does not pursue government-led supply chains—"that's not our strength"—but relies on private companies and commercial logic. Government sets frameworks and builds platforms while markets lead actual construction. He compared it to Apple's product thinking under Jobs: American companies' products win by "captivating and delighting," not through government instructions.