Week of January 25, 2026

This Week's Top Videos

Claude Code Is Taking Over (And We Don't Hate It)

By AI For Humans

Anthropic's Dario predicts AI models will handle 'most maybe all of what SWEs do end to end' within 6-12 months, while Claude Code users are already spinning up entire fleets of programmers. The middle tier of software engineers faces elimination as entry-level coding gets automated and only top-tier architects remain valuable.

• Engineers at Anthropic report they 'don't write any code anymore' - they just let models write code and edit it, signaling the immediate shift in how senior developers work
• The middle lane of software engineering jobs will evaporate, leaving only entry-level individual makers and high-tier genius architects that companies pay premium salaries to retain
• Junior engineers must adopt latest AI tools as their competitive edge since traditional entry-level coding tasks are now what machines output automatically
• Claude Code is maintaining discussion dominance over OpenAI, Google, and major IDEs like Cursor, creating potential moat and escape velocity in the AI coding space
• Users are 'blowing through' $200 Claude Code Max subscriptions by building fleets of automated programmers and complex video editing tools with Remotion
• At least 3 major frontier model improvements expected this year, with Claude 5 likely arriving by end of year, accelerating the automation timeline

Harvey CEO Winston Weinberg: How to Make Mega Deals | Lessons from Rabois, Halligan & Grady

By 20VC

Harvey CEO Winston Weinberg reveals B2B SaaS value is about to become 'astronomical' as AI plateaus for consumers but explodes for enterprises. His framework: run 1 mile daily to build stress tolerance, zero out Slack every 15 minutes for decision-making speed, and transition from founder heroics to building a 'well-run machine.' Critical insight for builders scaling AI companies right now.

• B2B SaaS value will become astronomical as AI performance plateaus for consumer use cases but explodes for enterprise applications, creating massive opportunities for vertical-specific solutions.
• Zero out Slack every 15 minutes in early stages - catch up to yesterday daily for faster decision-making, but transition away from this as you scale to focus on P0 priorities.
• Run one mile daily at maximum effort first thing in the morning to build stress tolerance and reduce decision-making pressure throughout the day - physical stress absorption improves company building.
• Know when to not negotiate in deals - if you want one specific thing (like hiring top talent), give them whatever they want instead of going back and forth on terms.
• Successful founders transition from heroics to building machines by jumping into random meetings across departments to understand how each function operates, then optimizing the entire system.
• Three-stage scaling framework: Stage 1 is product-market fit, Stage 2 is company-market fit (building structures different from traditional SaaS), Stage 3 is reinventing product-market fit again.

Vinod Khosla and Keith Rabois on Building and Investing in Enduring Companies | Ep. 40

By Uncapped

Keith Rabois made 70% of his investments in AI after rejoining Khosla Ventures just two years ago, having invested zero AI companies before. He credits working with Vinod Khosla for avoiding either complete irrelevance or reckless investing in the AI wave. Their partnership works because both prioritize first-principles thinking over political correctness, spending less than 5% of time on firm operations.

• In just two years at Khosla Ventures, Keith Rabois shifted from zero AI investments to 70% of his portfolio being AI companies, crediting institutional knowledge for avoiding either missing the wave entirely or investing recklessly.
• Top VCs spend less than 5% of their time on firm operations, focusing almost entirely on investing and portfolio companies rather than management or LP relations.
• Every Monday partner meeting starts with existing portfolio review before discussing new opportunities, emphasizing the 10-20 year partnership mindset over deal sourcing.
• First-principles thinking eliminates political debates between partners—isolating key variables (A, B, C factors) makes investment decisions clearer and faster.
• The most important hiring principle is 'the team you build is the company you build'—a core lesson that shaped how Keith approached both Square and venture investing.
• Successful VC partnerships thrive on 'brutal honesty over hypocritical politeness'—direct communication style that ambitious founders actually prefer and process better.

Build Hour: Apps in ChatGPT

By OpenAI

OpenAI just launched a ChatGPT apps platform with SDK, UI components, and MCP server integration that lets developers build custom apps directly inside ChatGPT. Apps like AllTrails and Adobe Express show how to pull external data and create dynamic UIs within chat conversations. This opens a massive new distribution channel for builders to reach ChatGPT's millions of users without building standalone apps.

• ChatGPT apps use MCP (Model Context Protocol) servers to pull contextual data that doesn't exist publicly on the web, giving access to private databases and APIs directly within chat conversations.
• OpenAI provides a complete UI component library and app SDK with pre-styled assets that automatically match ChatGPT's design system, eliminating frontend design work for developers.
• Apps can dynamically render content and switch to full-screen mode within ChatGPT, creating immersive experiences that go beyond simple text responses.
• The platform includes a public app submission flow and marketplace, allowing developers to distribute their apps to ChatGPT users globally without separate app stores.
• OpenAI launched a new Docs MCP server specifically designed to help developers build apps faster by providing documentation and examples directly in the development process.
• Apps can take actions on behalf of users and refine requests contextually, like filtering dog-friendly trails or filling event details into templates automatically.

AI coding’s impact on open source software is not great

By Designer Tom

Free AI Video Generator on Your PC (No Subscriptions, No Limits)

By Kevin Stratvert

You can now run Sora-quality AI video generation locally with just 6GB VRAM using open-source models like LTX-2, generating unlimited videos with audio narration for zero ongoing costs. Tools like Pinokio eliminate complex Python/CUDA setup with one-click installs. This breaks the subscription stranglehold of commercial AI video tools.

• LTX-2 model produces video quality comparable to Google's Veo 3 and OpenAI's Sora 2 while running locally on consumer hardware with 6-8GB VRAM minimum
• Pinokio acts as a Steam-like one-click installer for AI tools, eliminating the complexity of managing Python versions, CUDA dependencies, and model installations
• The distilled version of LTX-2 is only 20GB vs 40GB for full model, offers nearly identical quality but significantly better performance on consumer GPUs
• LTX-2 uniquely generates both video and synchronized audio narration from text prompts, including character dialogue specified in prompts with 'She says' commands
• The 19 billion parameter LTX-2 model requires 35-40GB disk space for model weights, making local storage planning crucial for deployment
• Multiple generation modes available: text-to-video, image+text prompts, video continuation, and end-frame specification for precise control over outputs