Week of February 15, 2026

This Week's Top Videos

Claude Opus 4.6 vs GPT-5.3 Codex: How I shipped 93,000 lines of code in 5 days

By How I AI Podcast

Developer claims to have shipped 93,000 lines of code in 5 days using OpenAI's new GPT-5.3 Codex and Anthropic's Claude Opus 4.6 models. OpenAI's new Codex desktop app introduces Git-native workflows with work trees for parallel agent development, while positioning skills as first-class visual components. This represents a massive leap in AI-assisted development velocity that could redefine how fast teams can iterate.

• OpenAI's Codex desktop app is built around Git primitives like work trees, enabling multiple AI agents to work on separate full copies of your codebase simultaneously without conflicts
• The new models enable shipping more code in 5 days than previously accomplished in a month, suggesting 6x+ productivity gains for complex codebases
• Codex transforms 'skills' from zip file uploads into visual, first-class UI components with icons and buttons, making AI workflows more accessible to non-technical users
• Testing methodology focuses on complex, established codebases rather than simple one-shot landing pages to properly evaluate model capabilities for real-world development
• Both OpenAI and Anthropic released competing coding models within the same week, indicating intense competition in the AI development tooling space

The New Way To Build A Startup

By Y Combinator

The best startups are now "20x companies"—tiny teams beating 100x larger competitors through total internal automation. Anthropic engineers manage 3-8 Claude instances each, while 5-person GigaML closed DoorDash against 100+ engineer competitors. Smart founders are building AI teammates, unified dashboards, and custom agents for every employee before hiring humans.

• Anthropic's own engineers manage 3-8 Claude AI instances each to build their product, with humans only meeting for foundational decisions while AI handles implementation, bug fixes, and research
• GigaML's 5-person team beat 100+ engineer competitors to win DoorDash by building Atlas, an internal AI agent that can use browsers, edit policies, write code, and handle all boilerplate work
• Legion Health grew 4x in revenue while keeping ops headcount completely flat by building a unified AI interface that gives instant access to patient history, scheduling, and insurance across their entire system
• Phase Shift's 12-person team competes against companies with hundreds of employees by documenting every manual task employees do, then building custom AI agents to automate those specific workflows
• 20x companies automate across all functions—code, support, marketing, sales, hiring, QA—allowing them to postpone hiring additional staff much longer and keep culture from drifting
• Companies are avoiding entire hiring categories by using AI tools—Phase Shift delayed hiring designers by using Magic Patterns for all front-end designs

Waymo Co-CEO on the Road to 1 Million Robotaxi Rides a Week

By Bloomberg Technology

Waymo just raised $16B at $126B valuation after hitting 400k paid rides weekly across 6 cities—quadrupling 2025 trips to 15M total. With 90% fewer serious injury crashes than human drivers across 127M miles, they're expanding to 20 cities this year including NYC and international launches in London and Tokyo.

• Waymo quadrupled trips in 2025, reaching 15 million rides with over 20 million lifetime rides, demonstrating rapid scaling after proving product-market fit
• The company achieved 90% fewer serious injury-causing crashes compared to human drivers across 127 million autonomous miles driven
• New investors Sequoia, DST, and Dragoneer joined at this funding round, signaling an inflection point as the company moves from R&D to commercial scale
• In welcoming regulatory environments like Miami, Waymo can launch from mapping to full commercial service in just a couple of months
• The company is diversifying its fleet beyond the i-Pace to include Ohi vehicles and Ioniq 5s to improve unit economics while scaling operations
• Waymo is planning first international expansions to London and Tokyo while simultaneously scaling across 20 US cities in 2025

Once You Learn This, Clients Stop Choosing Your Competitors

By The Futur

Nike's underdog pitch to Michael Jordan reveals the ultimate competitive strategy: teach prospects exactly what questions to ask your competitors about their biggest weaknesses. By prepping Jordan's mom to ask Converse "how will Michael stand out?" and Adidas "who's really in charge?", Nike positioned their rivals' strengths as fatal flaws. This psychological warfare tactic is crucial for startups competing against established players.

• When you're the underdog, don't highlight your strengths—expose your competitors' weaknesses by teaching prospects what specific questions to ask them
• Turn your competitor's greatest strength into their weakness: Nike reframed Converse's star roster as making Jordan just "number four"
• Prepare clients to expose organizational dysfunction by asking "who makes the final decision?" to multi-stakeholder competitors
• Use cultural and operational differences as wedges—Nike highlighted Adidas being German-run versus American understanding
• Always prepare your own slam-dunk answer when teaching prospects to question competitors—anticipate they'll ask what to ask you
• Identify the real decision maker early and craft your entire pitch around what matters most to them, not just the obvious contact

Every programming language teaches us something

By GitHub

A language designer reveals why every successful programming language—from Python's AI dominance to Rust's borrow checker innovation—teaches us something valuable. Languages only survive if they solve real problems, making cross-language learning essential for any serious developer. This matters now because the fastest way to level up your technical skills is studying what made each language successful.

• Any programming language that gains traction has inherent value because getting adoption is extremely difficult—respect the journey even if you don't like the syntax.
• Rust's borrow checker represents a breakthrough approach to memory management that's less costly than automatic garbage collection while maintaining safety.
• Go succeeds as a 'memory safe and type safe C'—proving that simple languages with modern safety features can win over complex alternatives.
• Python's dominance in AI and machine learning demonstrates how language design choices can capture entire industry segments.
• Language designers must borrow concepts from other languages—creating something entirely new from scratch is a fool's errand.