X / Twitter
Replit CEO Amjad Masad
Amjad shared an important development in Python’s standard library: a “controversial” view he expressed last year has become reality, with the related functionality being incorporated officially. He also reposted SaaS MRR growth figures, arguing that the growth rate matters more than absolute MRR because of its enormous compounding potential. He quoted a line about Replit’s sales culture: “Replit’s best salesperson had never sold anything before.” Together, these posts outline his current interests in platform ecosystems, startup growth, and nontraditional talent emerging in the AI era.
https://x.com/amasad/status/2034149267871567921
https://x.com/amasad/status/2034379488948269379
https://x.com/amasad/status/2034381885129073018
Vercel CEO Guillermo Rauch
Guillermo announced Vercel’s coding-agent plugin, giving Claude Code and Cursor production deployment capabilities in two steps: install the plugin, then do nothing else. He emphasized that `npx plugins add` retrieves all Skills and keeps them automatically updated, calling the experience very smooth. He also announced that Mitchell Hashimoto, HashiCorp founder and Ghostty creator, has joined Vercel’s board, calling him the ideal person to help define agent infrastructure as AI reshapes the field. He released Next.js 16.2 “Snow Leopard,” centered on performance, agentic developer experience, and robustness.
https://x.com/rauchg/status/2034095862406901772
https://x.com/rauchg/status/2034378740197912625
https://x.com/rauchg/status/2034383612414754983
Y Combinator CEO Garry Tan
Garry emphasized that agents are blind without data and introduced YC company RunCaptainRAG’s automated file search. The team applied NLP research to its startup, delivered precise retrieval in Garry’s List, and performed a Collison install in person at YC’s MV office. He also shared Claude Code lessons: `/plan-ceo-review` and `/plan-eng-review` cover 90% of scenarios. The secret to eng-review is having the model generate diagrams—user flows, data flows, branching logic, and edge cases—because doing so itself reveals hidden bugs.
https://x.com/garrytan/status/2034096942045913401
https://x.com/garrytan/status/2034117190245618106
Anthropic Research’s Alex Albert
Alex announced a webinar that day at 10 a.m. Pacific Time on applying AI at work, especially best practices for new products such as Claude for Excel and PowerPoint. Participants would also receive a free month of Claude Pro. This marks an important step in Anthropic’s efforts to make AI tools accessible to ordinary workplace users rather than developers.
https://x.com/alexalbert__/status/2034276242317566107
Box CEO Aaron Levie
Aaron analyzed the future of AI agent payment infrastructure: agents will outnumber human users by orders of magnitude, need payment capabilities for web services, and have the distinctive advantage of frictionless payments. This enables business models previously impractical for humans, such as micropayments, payment for individual API calls, and on-demand information purchases. He noted that Stripe’s MPP (Merchant Payment Protocol) and Tempo are positioning themselves here, with new revenue models opening monetization opportunities for technology and information providers.
https://x.com/levie/status/2034283304401113453
Roblox product manager Peter Yang
Peter offered his assessment of AI coding tools: Codex is OpenAI’s best product, while Anthropic has broadly penetrated knowledge-work use cases. He also posed a question: building APIs and MCP is the obvious path, but when should you build an agent yourself? This hints at his framework for thinking about AI infrastructure.
https://x.com/petergyang/status/2034106497475743819
https://x.com/petergyang/status/2034118396191904230
Google Labs VP Josh Woodward
Josh announced another Gemini API update and invited developers to share what they build with its new capabilities. He also announced a major Stitch update, highlighting Stitch Live: users can click a design and give voice instructions such as “change this screen,” or use Stitch as a live design-review partner, with the system immediately making changes.
https://x.com/joshwoodward/status/2034324178044871002
https://x.com/joshwoodward/status/2034342669808443896
Google Labs
Google officially released Stitch by Google as a vibe design platform that turns natural-language descriptions into high-fidelity designs in one flow. Its core features include an intelligent design agent that turns business concepts or app visions into designs on an AI-native canvas; rapid iteration through connected interactive prototypes and portable design-system management; and voice collaboration for hands-free, real-time layout updates and variant exploration. It currently supports English, is restricted to users aged 18 and over, and is available in countries covered by Gemini.
https://x.com/GoogleLabs/status/2034337527293944228
FPV Ventures Partner Nikunj Kothari
Nikunj shared extensive experience with several underrated models. Gemini 3 Flash offers excellent value for long contexts and structured output, with Claude directly recommending “do it in one go; no multi-agent orchestration needed.” GPT 5.4 Pro is unmatched on deep-reasoning questions, but has high latency and is available only in ChatGPT, not yet through the API. NotebookLM excels at slides, compressing many sources into polished decks, using nano banana underneath at almost no cost. OpenRouter offers many free open-source models to explore; Minimax 2.5 and GLM stand out on price, and he looks forward to DeepSeek 4 and Kimi’s next version.
https://x.com/nikunj/status/2034267921489519009
Deep learning researcher Andrej Karpathy
Andrej recalled Jensen Huang’s NVIDIA GTC 2015 talk, when Jensen excitedly told an audience mainly of gamers and scientific-computing specialists that deep learning was the next big thing. He cited Karpathy’s doctoral work, one of the earliest image-captioning systems, coupling an image-recognition ConvNet with an autoregressive RNN language model and training end to end. Most people were still waiting or skeptical, but Jensen identified the direction extremely early with remarkable foresight. Karpathy called him “one thousand percent right.”
https://x.com/karpathy/status/2034325423358955981
OpenClaw founder Peter Steinberger
Peter announced his involvement in building the Codex community page and invited people to register as Codex ambassadors. He also reposted an update saying OpenClaw now supports “almost every token provider,” adding the naming joke “insert name joke here.” He posted an image that sparked substantial discussion, with 1655 likes and 188 replies. Its content was unspecified, but it clearly struck a nerve in the community.
https://x.com/steipete/status/2034400645630058792
https://x.com/steipete/status/2034401076204679324
Every CEO Dan Shipper
Dan shared a new Codex feature: threads can now message one another, particularly useful for managing stacked production deployments. Paste a thread ID into another Codex chat and the deployment thread can take over. This is a concrete example of agent-based workflow orchestration.
https://x.com/danshipper/status/2034368436554326394
Builder Zara Zhang
Zara announced that her `frontend-slides` skill has passed 10,000 GitHub stars, with many users making it their default slide-creation method and saying goodbye to PowerPoint. She also showed a 2-minute video created from meeting notes using OpenClaw and a Remotion skill, complete with script, visuals, voiceover, and animation. She believes a video summary is far more effective than a 1000-character written summary.
https://x.com/zarazhangrui/status/2034331675363279338
https://x.com/zarazhangrui/status/2034379412997919096
Claude (@claudeai)
Anthropic announced the return of Code with Claude this spring, expanding to San Francisco, London, and Tokyo. It offers full-day workshops, demos, and one-on-one office hours with the Claude team, with online viewing and applications for in-person attendance.
https://x.com/claudeai/status/2034308517964956051
Podcasts
AI & I by Every — How Every Edits in the Age of AI
(The transcript for this episode is currently unavailable, so a summary cannot be generated.)