← Collected sources
BUILDERS · EDITED DIGEST

Builders’ Picks | 2026-03-26

2026-03-26 · Historical edition

X / Twitter

Anthropic Claude Code team member Thariq

Thariq announced that Claude Code now supports iMessage as a new channel, allowing interaction through iMessage, and linked to documentation. He also issued an important announcement about Claude usage limits: to address growing demand, Anthropic will adjust the 5-hour session limits for free, Pro, and Max plans during weekday peak hours of 5–11 a.m. Pacific Time, or 1–7 p.m. Greenwich Mean Time, causing allowances to be consumed faster than before. Weekly limits remain unchanged. Around 7% of users will be affected, especially Pro users running token-intensive background tasks, who are advised to move such tasks to off-peak hours.

Vercel CEO Guillermo Rauch

Guillermo argued that every company will become an “AI factory” whose production unit is the token, but token metering and billing differ fundamentally from traditional SaaS. He announced that Vercel AI Gateway now handles cross-model and cross-provider metering through a single `/v1/report` API call. He also shared a new Vercel Sandbox feature: agents need stateful compute environments, and Vercel Sandbox supports persistence simply by giving a sandbox a name, with Vercel handling the rest. Deployment through Vercel CLI is also supported, letting agents freely call tools, install dependencies, write code, and run and debug software.

Deep learning researcher Andrej Karpathy

Using his experience building menugen about a year ago, Karpathy argued that the hardest part of AI development today is not code itself but DevOps: assembling services such as payments, authentication, databases, security, and domains like IKEA furniture. He envisions a future where telling an agent “build menugen” is enough for it to browse service documentation, obtain API keys, debug, and deploy autonomously, turning the entire DevOps lifecycle into code. He acknowledged that this vision is “only just technically feasible now, but needs to be designed from scratch,” and finds the direction exciting.

Box CEO Aaron Levie

Aaron believes we seriously underestimate the complexity of change management required to automate knowledge work. Data in legacy systems or environments without good APIs, missing task context, and uneven technical backgrounds across teams all make enterprise AI transformation harder than expected. He sees good news for founders: opportunities lie in software bridges that reduce this transition friction, or new service firms offering change-management consulting.

Y Combinator CEO Garry Tan

Garry shared results showing GStack winning a direct comparison with Superpowers. He also emphasized a central idea for the new era: creating outstanding things requires aggressive token use and a willingness to “go for it.” Access to token credits is crucial for breaking down geographic and wealth barriers to entrepreneurship, giving people anywhere a chance to participate.

Google Labs VP Josh Woodward

Josh announced Gemini Live’s biggest upgrade yet: faster responses, smarter content, stronger emotional understanding (EQ), more languages, and a context window 2 times larger, powered by Gemini 3.1 Flash on both Android and iOS.

OpenClaw founder Peter Steinberger

Peter announced a partnership between OpenClaw and Telegram, with direct support from Telegram founder Durov, and izhukov joining as a new OpenClaw maintainer. The immediate priority is investigating intermittent duplicate messages when the bot streaming API is enabled. Resolving this should greatly improve Telegram support. He also said Microsoft Teams had completed its OpenClaw integration and delivered excellent improvements, with Slack collaboration expected to materialize within days.

FPV Ventures Partner Nikunj Kothari

Nikunj demonstrated an AI workflow that excited him: open Claude Code, connect Google Calendar MCP, install Ramp CLI, and complete 3 months of expense categorization and notes in around 12 minutes, with personally identifiable information removed. He also reposted a joint Stripe and Railway demo, calling it well worth watching.

Builder Zara Zhang

Zara quoted Brandon Chen, founder of “one of the most AI-native startups”: “We would rather waste tokens than waste time.” She also reflected on where product ideas originate: not from “thinking,” but from “playing,” continuously experimenting with models, and “talking,” broadly engaging with different people. Ideas arise from collisions and exploration without an immediate practical goal.

Anthropic philosopher Amanda Askell

Amanda offered a sharp workplace observation: technology companies spend millions hiring talented people, then place them in open-plan offices where they cannot concentrate. She thinks the most effective recruiting strategy may simply be providing a private office with a door. She added that widespread remote work may make things harder for those who dislike working from home, because “you can work at home” has become the default assumption.

South Park Commons Partner Aditya Agarwal

Aditya shared highlights from investor Elad Gil’s interview at South Park Commons. Elad offered several views counter to Silicon Valley convention: co-founders are unnecessary, winning is the only startup culture that matters, more market opportunities are open now than ever, product breadth builds a stronger moat than data, and micromanagement is severely underrated.

Replit CEO Amjad Masad

Amjad shared figures showing high App Store approval rates for Replit-built apps, calling this a strong signal of Apple’s recognition of their quality. He also described a conversation with US Senator Josh Hawley about how vibe coding could unlock a new wave of American entrepreneurship, emphasizing that platforms should not erect barriers to the emerging category’s development.

Linear Head of Product Nan Yu

Nan argued that sometimes there is no “third option balancing both sides,” citing Instagram’s copying of Snapchat Stories and its well-known outcome. He also reflected on design’s essence, wishing the word would return to its original meaning: “a plan for arranging elements for a particular purpose.” The industry obsesses over arranging elements while forgetting the purpose. He repeatedly mentioned Heroku’s missed opportunity, lamenting that “such a great opportunity was wasted.”