Week of March 1, 2026

This Week's Top Videos

Every AI Model Explained

By Tina Huang

Grok breaks the AI cost-performance rule: users vote it most capable yet it's faster and cheaper than GPT-4o and Claude, with 2M token context window that can analyze entire books. The airplane analogy reveals why 80% of your AI queries should use mid-tier workhorses, not flagship models that cost 10x more.

• Use the airplane framework to select AI models: flagship models (GPT-4o, Claude) are like commercial jets—powerful but expensive and slow, while light models are like private jets—fast and cheap but limited capability.
• Grok is an anomaly that breaks normal AI trade-offs—currently rated most capable by users but also faster and cheaper than competitors, with highest emotional intelligence for user interactions.
• Mid-tier AI models handle 80% of queries efficiently, just like Boeing 737s handle most flight routes—focus on these workhorses instead of always using expensive flagship models.
• Use Perplexity AI as a model aggregator to access multiple premium AI models without paying for separate subscriptions to OpenAI, Anthropic, and others.
• Claude Opus excels at code generation and writing but lacks multimodality—can't generate images directly, making it specialized rather than well-rounded like GPT-4o.
• Context window size determines how much data you can input—Grok's 2 million token window lets you analyze entire books in a single prompt.

The SaaS Apocalypse: Who Lives & Who Dies | Insight Partners Co-Founder, Jerry Murdock

By 20VC

Insight Partners co-founder Jerry Murdock claims autonomous agents have made Cursor obsolete in just 6 weeks—his $90B portfolio companies are now using OpenClaw agents to write code natively. The "SaaS apocalypse" isn't about AI bolt-ons but autonomous agents creating an orchestration layer that will shift compute from Nvidia to cheaper ASICs.

• Portfolio companies using autonomous agents for 2-6 weeks report that Cursor ($27-30B valuation) is already obsolete as a product category
• AI-native startups like E2, Eventual, and Lotus AI are using OpenClaw and Nano Claw autonomous agents to write code, not just assist with it
• The coming orchestration layer will let agents triage workflows between expensive Claude tokens and cheaper open-source models like Deep Seek
• Just like the LAMP stack (Linux, Apache, MySQL, PHP) enabled the 2004-2005 website explosion, we need a 'Claw stack' for autonomous agents
• ASIC chips will rise as autonomous agents put models directly on chips, making them cheaper and more tunable than Nvidia's expensive GPUs
• The AI tsunami analogy: it's harmless at sea but devastating when it hits the beach—companies need to move to higher ground now, not bolt on AI later

How I Use Obsidian + Claude Code to Run My Life

By Greg Isenberg

Forget basic AI prompting—combining Obsidian's interconnected markdown files with Claude Code creates a 'thinking partner' that remembers your context across sessions. Instead of re-explaining projects repeatedly, you pass specific files to Claude Code for instant context, enabling complex automation and idea generation on tap.

• Claude Code can control your computer through natural language commands, creating files and automating tasks you'd normally do manually through GUI interfaces.
• The key to unlocking Claude Code's power is feeding it rich context files instead of starting fresh each session—this prevents having to re-explain complex projects repeatedly.
• Obsidian creates interconnected markdown files that show relationships between ideas, unlike regular folders—each note can link to others forming a knowledge graph.
• By combining Obsidian's context-rich files with Claude Code's automation, you can pass specific project descriptions and contexts directly to the AI for immediate understanding.
• The system helps you notice patterns in your thinking that you wouldn't catch on your own by creating persistent, searchable records of your thoughts and connections.
• This workflow transforms AI from a basic chat tool into a persistent thinking partner that remembers your projects, preferences, and thought processes across sessions.

The New Design Portfolio (No Case Studies)

By Designer Tom

Hiring managers decide on design portfolios in just 55 seconds—and they're not reading your case studies. With 73% now requiring AI tool proficiency and 58% prioritizing visual craft over process, designers are ditching traditional portfolios for interactive "playgrounds" that demonstrate building ability in public.

• Design hiring decisions happen in 55 seconds, not enough time to read process case studies—focus on immediate visual impact and interactivity instead.
• 73% of hiring managers now consider AI tool proficiency a requirement, not a bonus, as single designers can create full experiences that previously required teams.
• Interactive portfolio "playgrounds" that let visitors click and explore for 10+ minutes outperform static case studies for getting callbacks.
• Only 25% of design job openings are junior level while 56% are senior, indicating entry-level roles may not return as AI handles foundational work.
• Building and sharing experiments publicly on Twitter is becoming a more effective portfolio strategy than traditional websites for landing design engineering roles.
• Visual polish and craft (58% priority) now matters more than process documentation as hiring criteria shifts toward demonstrated building ability.

How I use Figma MCP + Claude to Vibe Code at Intercom

By Sneak Peek Design

Intercom built a revolutionary workflow where designers can copy any Figma component link and instantly generate production code in their IDE using Claude + Figma MCP. Every single component is code-connected, eliminating the traditional design-to-dev handoff and letting designers prototype directly in the codebase. This is the future of design-engineering collaboration.

• Every single Figma component at Intercom is connected to code via Code Connect - buttons, icons, dialogs, inputs, text areas, and switches all have direct code mappings accessible in developer mode
• Designers can copy a Figma link and tell Claude 'build the dialogue from this link' - AI automatically imports the correct components from the codebase instead of creating new ones
• The design playground gives every designer access to the codebase with a dedicated branch and single script setup to prototype directly in production environment
• Figma MCP integration provides Claude with component names, tokens, and full context from Figma files, solving the problem of AI creating duplicate components
• Engineers also benefit from this designer tooling, using the dedicated prototype pages without worrying about integration groundwork and connections
• The workflow eliminates design-dev friction by letting designers start in Figma OR code, with seamless transitions between both environments using the same components

How to Combine Cells in Excel (Without Losing Data)

By Kevin Stratvert

Excel's Ctrl+E flash fill can automatically complete cell combinations after typing just one example, eliminating manual data entry. For dynamic updates, TEXTJOIN function with TRUE parameter ignores blanks and refreshes when source data changes. This transforms tedious spreadsheet work into one-keystroke automation that saves hours weekly.

• Excel's flash fill feature activates with Ctrl+E after typing one example, automatically completing the rest of your data combination pattern without formulas.
• TEXTJOIN function with TRUE parameter automatically ignores blank cells when combining data, preventing empty spaces in your results.
• Unlike flash fill, TEXTJOIN creates dynamic formulas that automatically update whenever your source data changes, maintaining accuracy over time.
• Adding delimiters like spaces between combined data requires specifying the separator as the first parameter in TEXTJOIN function.
• Manual cell combination is inefficient and error-prone compared to Excel's built-in automation features that handle patterns instantly.