AI Builders Digest — March 26, 2026
Claude Code auto mode: a safer way to skip permissions
Claude Code now offers a new “auto mode” that uses model-based classifiers to automatically approve safe commands while blocking dangerous ones. This addresses a key UX challenge: users typically accept 93% of permission prompts anyway, leading to “approval fatigue” where they stop paying attention. Auto mode targets high autonomy at low maintenance cost—a middle ground between manual review and the unsafe –dangerously-skip-permissions flag. The classifiers catch actions not aligned with user intent (like deleting git branches or uploading auth tokens) while letting routine operations through without prompts.
https://www.anthropic.com/engineering/claude-code-auto-mode
Andrej Karpathy, AI researcher (formerly Director of AI @ Tesla, founding team @ OpenAI, PhD @ Stanford)
Observes that LLMs tend to overfit to whatever context is in the window—a bias likely developed during training where relevant information is present. Also notes that personalization memory in LLMs becomes distracting; a single question from months ago can resurface as a perceived “deep interest” inappropriately.
https://x.com/karpathy/status/2036841069636370467
https://x.com/karpathy/status/2036836816654147718
Swyx, builder at dxtipshq / cognition / temporalio / aiDotEngineer
Shares that the upcoming aiDotEngineer London event in 2 weeks is extremely challenging to run despite sold-out booths and tickets. Thanks sponsors GDM, OAI, Braintrust, and WorkOS for keeping them afloat. Still has Afterparty, Leadership Luncheon, and Expo Cafe available for last-minute sponsors.
https://x.com/swyx/status/2036925527441965164
https://x.com/swyx/status/2036919167639839193
https://x.com/swyx/status/2036918746854678830
Kevin Weil, VP Science @OpenAI
Shared an interesting link (no further context in the tweet itself).
https://x.com/kevinweil/status/2036809393405317545
Peter Yang, Product at Roblox
Excited about something new that he describes as “like crack”—likely referencing a new AI tool or feature. Also built his own fitness app while setting up a squat rack. Shares a practical post about how VP at Meta solved the “predicting what your leader wants” problem by building an AI skill that reviews docs in their voice.
https://x.com/petergyang/status/2037042313361588701
https://x.com/petergyang/status/2037029503177023705
https://x.com/petergyang/status/2036812870365917284
Nan Yu, Head of Product @Linear
Reflects on design terminology: wishes “design” could reclaim its meaning of “a plan for arranging elements to accomplish a particular purpose” rather than just being about arranging elements. Notes that sometimes there are only two options with no third balance—you have to commit fully (e.g., should Instagram fully copy Snapchat Stories or not?).
https://x.com/thenanyu/status/2037042617213481410
https://x.com/thenanyu/status/2037035153566953870
https://x.com/thenanyu/status/2037008518515953803
Cat Wu, Claude Code team @Anthropic
Announces that Auto mode is now available for Claude for Team users. Calls it a “step change improvement” balancing autonomy and safety, with almost everyone on the team using it as a daily driver.
https://x.com/_catwu/status/2036852880624541938
Thariq, Claude Code @Anthropic
Announces iMessage is now available as a channel in Claude Code.
https://x.com/trq212/status/2036959640022638922
https://x.com/trq212/status/2036959638646866021
https://x.com/trq212/status/2036954677666021582
Amjad Masad, CEO @Replit
Notes Apple seems happy with apps made with Replit, as evidenced by high acceptance rates.
https://x.com/amasad/status/2037004600893472936
Guillermo Rauch, CEO @Vercel
Argues every company will become an “AI factory” where the unit of production is the token. Tokens create usage tracking and billing challenges unlike traditional SaaS. Announces AI Gateway now solves metering across models and providers with a single /v1/report API call.
https://x.com/rauchg/status/2036963706576527623
Aaron Levie, CEO @Box
Articulates Jevons paradox in action: companies now realize they can afford software projects they couldn’t before because AI makes production cheaper. Predicts marketing teams will have engineers, life sciences will automate research, small businesses will hire engineers for the first time. Argues that human oversight of agents (prompting, reviewing, guiding back, maintaining, fixing bugs) means engineering jobs aren’t going away—the world will be increasingly made of software, and those who understand it best will be in a strong position.
https://x.com/levie/status/2036832183131033977
Ryo Lu, Design @Cursor_ai
Makes a key observation: as agents make it easy to add features, design matters more, not less. The role is no longer just pushing pixels—it’s deciding what should exist, how it fits together, how humans stay in control, and how intelligence feels clear, trustworthy, and useful. Taste, craft, and judgment have always been the bottleneck; the game is who makes the right thing for humans, not who ships fastest.
https://x.com/ryolu_/status/2036886854805709097
Garry Tan, President & CEO @YCombinator
Emphasizes that in this new age, you have to use tokens aggressively to create something remarkable. If you do, and have agency and taste, the result will be remarkable. Token credits for AI are key to making startups accessible regardless of where you grew up or family wealth. Also shared that GStack beat Superpowers in a head-to-head comparison.
https://x.com/garrytan/status/2037055498974093629
https://x.com/garrytan/status/2037054022994321725
https://x.com/garrytan/status/2037051410857283967
Matt Turck, VC at FirstMarkCap
Observes the evolution of what VCs do on company boards: 2016 was governance/guidance/support, 2021 was cheerleading, 2026 is being Anthropic and OpenAI sales representatives.
https://x.com/mattturck/status/2036747586468499889
Nikunj Kothari, Partner @FPV Ventures
Co-hosting a founder dinner on April 7th in SF with Andru Yeung.
https://x.com/nikunj/status/2036936135541481923
https://x.com/nikunj/status/2036935343199707640
Peter Steinberger, ClawFather @OpenClaw
Announces new OpenClaw beta with better MS Teams integration, OpenWebUI support, and more.
https://x.com/steipete/status/2036824286988816737
Dan Shipper, CEO @Every
Teases a big announcement for the next day at Every. Shares an episode of AI & I podcast featuring Mike Krieger (cofounder of Instagram, now at Anthropic Labs) discussing how to build agent-native products, why teams need to throw out their product every 3-6 months as AI progresses, and how Anthropic Labs structures experiments with just 2-person teams.
https://x.com/danshipper/status/2036923805273129160
https://x.com/danshipper/status/2036831785154183190
https://x.com/danshipper/status/2036827118915485942
Aditya Agarwal, GP @SouthPkCommons, Co-Founder @Bevel_Health
Shares observations from Hill Valley Forum: it takes three to tango (Builders + Investors + Policy Makers) for hardtech/deeptech. Notes insane energy around drones, energy, robotics, atoms+bits—not just Silicon Valley but dispersed across the US. Positive about Washington support. Optimistic about US going from bits to atoms+bits diversifying capital sources.
https://x.com/adityaag/status/2036824253970923925
Claude, AI assistant @Anthropic
Announces that work tools in Claude (Figma, Canva, Amplitude) are now available on mobile.
https://x.com/claudeai/status/2036850783526719610
Training Data — “Biology’s Waymo Moment: Ginkgo Bioworks’ Jason Kelly”
Jason Kelly, founder and CEO of Ginkgo Bioworks, believes AI in biology is fundamentally different from previous tech revolutions. Unlike internet and social media which were “meaningless to biotechnology,” AI is changing the fundamentals of how science is done and will disrupt biopharma. He bootstrapped Ginkgo for 5-6 years (2008-2014) without raising capital—unusual in biotech VC which typically avoids young founders. In 2014, Sam Altman persuaded him to raiseVC funding after his blog post arguing the Silicon Valley model could work for deep tech and biotech. Kelly sees biology as becoming programmable, with AI enabling a new approach to engineering living systems.