Week of March 22, 2026

This Week's Top Videos

Building AI for better healthcare — the OpenAI Podcast Ep. 14

By OpenAI

OpenAI worked with 250 physicians to train healthcare models while 40 million people already use ChatGPT daily for health queries—one in four weekly users. Their new ChatGPT Health creates secure, contextual healthcare conversations with guaranteed data protection and no training on user health data. This matters now because consumer healthcare AI demand is exploding while most builders are missing the clinical validation piece.

• 40 million people per day use ChatGPT for health-related queries, representing one in four of ChatGPT's 900 million weekly users—revealing massive untapped healthcare AI demand
• OpenAI collaborated with a cohort of 250 physicians across every stage of healthcare AI model training, showing the critical importance of clinical validation in AI development
• AI is discovering new applications for existing medications that have been 'sitting on a shelf,' creating direct patient value from previously unused pharmaceutical assets
• ChatGPT Health implements a 'one-way valve' security system with guaranteed no training on user healthcare data, setting new standards for healthcare AI privacy
• Context matters in healthcare AI—traditional search engines have 'amnesia' and are one-size-fits-all, while personalized context dramatically improves healthcare AI outcomes
• Healthcare systems are reactive rather than proactive, with patients left 364 days per year without engagement from organizations holding their centralized health information

Gokul Rajaram on the 8 Moats Companies Need & Why Dropouts are "AI Maxing" the World

By 20VC

Gokul Rajaram's "8 moats" framework reveals that companies need 4+ moats to be secure, ranging from data and workflow to network effects and scale. His key insight: remarkable products need multiplayer elements—Figma succeeded because sharing designs was effortless, creating natural viral distribution. Critical for today's AI-first world where single-player tools are easily commoditized.

• Companies need at least 4 of 8 moats to be secure: data, workflow, regulatory, distribution, ecosystem, network, physical infrastructure, and scale moats
• Gmail's 1GB free storage vs Yahoo's 10MB (100x improvement) exemplifies the 'remarkability' threshold—products must be 10-100x better than alternatives
• Multi-product companies win: Square grew from 1 payment product to 11 products each doing $50M+ revenue, with retention tied to median products used per merchant
• Not every product needs profit—some are retention tools. Square Capital barely made money but dramatically improved merchant stickiness through natural product adjacency
• The best PLG software is multiplayer by design. Facebook taught that distribution genius means making products that require multiple users to function properly
• Vertical products must own the full stack to reach $10B+ valuations—single-layer horizontal solutions face commoditization pressure in AI-dominated markets

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit

By Pragmatic Engineer

Simon Willison codes more on his phone than laptop using AI agents that now write most of his code—he shipped a feature mid-interview and got 49% performance improvements in 30 minutes. The game-changer: red-green TDD with agents plus manual verification makes it safe to trust AI without reading every line. This matters NOW because Claude Opus 4.5 crossed the reliability threshold where agents consistently produce production-quality code.

• AI agents crossed a reliability threshold with Claude Opus 4.5 and GPT-4o in November—they now consistently write good solutions instead of janky code that needs fixing
• Start every coding session with 'use red green TDD' prompt—agents will write tests first, preventing over-engineering and dramatically improving code quality
• Tests are now effectively free with AI agents, making them non-optional for any serious development work—the old excuse of 'extra work' no longer applies
• The progression: first agents help occasionally, then write more code than you, then you stop writing code entirely, finally you stop reading code—but require proof it works
• Manual verification is critical even with passing tests—agents must actually run and demonstrate the working application, not just pass unit tests
• Trust AI agents like you trust other professional teams' services—focus on outcomes and documentation rather than inspecting every line of implementation

Every company ‘needs an OpenClaw strategy’: Jensen Huang claims at Nvidia GTC 2026

By TechCrunch

Jensen Huang declares 'OpenClaw' as the operating system for AI agents, claiming it's as foundational as Windows was for PCs. Every company now needs an 'OpenClaw strategy' just like they needed Linux, HTML, and Kubernetes strategies in previous tech waves. This positions agentic systems as the next mandatory infrastructure layer for all software companies.

• OpenClaw has open-sourced the operating system for agentic computers, making it as foundational as Windows was for personal computers
• Every company needs an 'OpenClaw strategy' - positioned as the next mandatory tech infrastructure after Linux, HTML, and Kubernetes
• OpenClaw enables the creation of 'personal agents' similar to how Windows enabled personal computers
• The strategic imperative applies to all software and technology companies, not just AI-first startups
• Agentic systems are being positioned as equivalent to previous foundational shifts like mobile cloud adoption

A few of our favorites plugins

By Figma

Figma designers reveal their productivity-boosting plugins: pixel art generators, dynamic shadow tools, and noise gradients for consistent thumbnails. These micro-tools show how specialized plugins can solve specific creative bottlenecks that generic design software misses. For product builders, this highlights the massive opportunity in creating niche workflow tools that solve narrow but painful problems.

• Pixel art creation tools tap into nostalgia-driven design trends, especially for gaming and retro aesthetics
• Dynamic lighting plugins with moveable light sources enable precise shadow control without manual adjustments
• Noisy gradient generators solve file organization by creating consistent thumbnail aesthetics across projects
• Gamified design tools like FigWordle show opportunity for productivity apps that include entertainment elements
• Specialized micro-plugins outperform generic tools by solving specific creative workflow bottlenecks

Master 80% of Claude Code in 26 Minutes

By Futurepedia

Claude Code lets non-developers build websites, apps, and automation tools by writing prompts in plain English—no terminal or coding skills required. Available as a $17/month desktop app, it creates and edits files directly on your computer, generating production-quality landing pages in minutes. This democratizes app development for founders who can't code but need custom tools.

• Claude Code works as a desktop app for $17/month, eliminating the need for technical terminal installations that intimidate non-developers
• Upload screenshots of websites you like as references—Claude will match colors, fonts, and layouts just like working with human designers
• Set up separate folders for each project so Claude only accesses files related to that specific build, maintaining organization and security
• Use Opus 4.6 model for initial builds and complex planning, then switch to Sonnet for revisions to save credits and costs
• Claude Code creates files locally on your desktop, unlike web-based tools that host online—giving you full control but requiring local setup
• The tool connects to existing data sources like Google Drive and Slack, providing context for building custom dashboards and automation