AI Builders Digest — April 4, 2026
AI Builders Digest — April 4, 2026
X / TWITTER
Swyx (AI Engineer at Cognition and Temporal) Swyx demonstrated what he calls "agentic self improvement" - successfully copying and pasting blog posts and tweets directly into Devin AI, which then one-shot implemented complete features. He was surprised it worked, noting this was "very out of distribution" for the underlying Google DeepMind Gemini Flash Lite model. He also shared a video showing recorded implementations with testing annotations, and made a pointed comment about AI Twitter's credibility issues. Tweet | Comment on AI Twitter | Testing annotation
Peter Yang (Product at Roblox) Peter Yang shared a detailed prompt for creating a personalized AI assistant that embodies "warm, sharp, dry humor" and avoids corporate speak. The prompt specifically forbids phrases like "Great question!" and "I'd be happy to help!" in favor of direct, opinionated responses with "charm over cruelty." He's also promoting an upcoming episode about Codex operations, particularly relevant given recent OpenClaw developments. Prompt details | Codex episode
Nan Yu (Head of Product at Linear) Linear's Nan Yu argued against combining design and product management roles, saying it creates "poor incentives" where design reports to product, making explorations shallow and hurting aesthetics. Instead, he advocates for combining product management with product marketing as a "more natural fit," calling PMM "a product concern." He also predicted that teams will eventually want to bulk-manage project issues using AI agents. Design/PM separation | PM/PMM alignment | AI project management
Thariq (Claude Code at Anthropic) Thariq offered a month of free Claude credits following what appears to be service disruptions, directing users to claim compensation "for bearing with us" during recent issues. Free credits offer
Amjad Masad (CEO at Replit) Replit CEO Amjad Masad celebrated a user achievement, noting they accomplished significant results in "not even a month" of building with their platform. User achievement
Aaron Levie (CEO at Box) Box CEO Aaron Levie provided a thoughtful analysis of the cognitive limits humans face when managing AI agents. He explained that just like human organizations require management hierarchies due to context limitations, current AI agents are "only as effective as the context they're provided" and the human's ability to manage them. He argued this is why job displacement fears are overblown - agents still require significant mental work from humans to be effective. Human cognition limits
Ryo Lu (Design at Cursor) Cursor designer Ryo Lu announced a new interface with updated icon set "made with care and character" and promoted Composer 2 with 2x usage available for weekend testing. He also shared design philosophy around keeping interfaces simple. New interface | Composer 2 promotion | Simple design
Garry Tan (President & CEO at Y Combinator) YC's Garry Tan celebrated achieving a 95 PageSpeed score for Garry's List, taking a shot at "codegen luddites." He also released a new developer experience review tool on GStack inspired by Addy Osmani's DX framework, emphasizing his approach of building tools for himself first. PageSpeed achievement | DX review tool
Zara Zhang (Builder) Zara Zhang observed a new trend of people "distilling colleagues, influencers, and even their exes into agent skills," highlighting how AI personalization is extending into replicating specific individuals. She also agreed with Nan Yu's take on combining product management and product marketing roles. Agent personalization trend | PM/PMM agreement
Nikunj Kothari (Partner at FPV Ventures) FPV partner Nikunj Kothari coined the term "Claude Code eyes" as the next anxiety after "token anxiety," suggesting visual fatigue from heavy AI coding assistant usage. He also shared lighter content about investigating X users' bios and visiting KAWS exhibitions. Claude Code eyes | X bio investigation
Peter Steinberger (OpenClaw) OpenClaw creator Peter Steinberger criticized Anthropic's API restrictions affecting open source developers, revealing he and Dave Morin tried to "talk sense into Anthropic" and managed to delay the restrictions by only a week. He noted the timing correlation between Anthropic copying popular features into their closed system before locking out open source. He also highlighted GitHub API quota limits as poorly designed for agent usage. Anthropic criticism | API credit recognition | GitHub API limits
Dan Shipper (CEO at Every) Every CEO Dan Shipper encouraged trying new AI tools despite their early stage, noting rapid iteration cycles make experimentation worthwhile. AI experimentation
Aditya Agarwal (General Partner at South Park Commons) Aditya Agarwal responded to a lawsuit against his health company Bevel Health by WHOOP, positioning Bevel as a free alternative to WHOOP's "$300 over-priced" device and promising to "build, not resort to lawfare" while bringing "great health to everyone." WHOOP lawsuit response
Claude (Official Anthropic account) Claude announced Microsoft 365 connectors are now available across all Claude plans, enabling integration with Outlook, OneDrive, and SharePoint to bring email, documents, and files directly into conversations. Microsoft 365 integration
PODCASTS
Latent Space - Marc Andreessen introspects on The Death of the Browser, Pi + OpenClaw, and Why "This Time Is Different"
The Takeaway: Current AI breakthroughs represent an "80-year overnight success" - seemingly sudden advances that actually draw from eight decades of accumulated research finally reaching practical deployment.
Marc Andreessen, co-founder of Netscape and a16z partner, offers a uniquely historical perspective on AI's current moment. Having coded in Lisp during the 1980s AI boom and witnessed multiple boom-bust cycles, he argues this time fundamentally differs because "now it's working."
The most compelling insight comes from Andreessen's framing of timing versus correctness: "There were AI researchers who spent their entire lives. They got their PhD, they researched for 40 years, they retired. In a lot of cases, they passed away and they never actually saw it work." The neural network architecture, controversial for 70 years since the original 1943 paper, has finally proven itself as the foundation for everything from ChatGPT to modern breakthroughs.
Andreessen traces key inflection points - AlexNet in 2013, the Transformer in 2017, and the strange four-year gap when companies like Google had internal chatbots but wouldn't deploy them. Even OpenAI initially called GPT-2 "too dangerous" for public use. "There was a year where the only way for a normal person to use GPT-3 was in AI Dungeon, where you'd pretend to play Dungeons and Dragons but really just try to talk to GPT."
While acknowledging AI's historical pattern of "excessively utopian and excessively apocalyptic" thinking, he believes the current wave breaks from past cycles because the underlying research foundation has finally matured into practical applications.
Generated through the Follow Builders skill: https://github.com/zarazhangrui/follow-builders