Tech Videos

Watch First

The single video most worth your time is When Will The Benchmaxxing Plague End? because it offers a highly pragmatic, technical teardown of the gamification of AI benchmarks and explains why current leaderboard scores rarely translate to real-world utility.

Highlights by Theme

Developer Tools & Platforms

In MCP Tasks (async): Why Aren’t Any Agents Supporting Them? from AI Engineer, Cornelia Davis explains why agent clients have avoided the Model Context Protocol (MCP) async tasks specification, noting that the v1 stateful architecture breaks under distributed workloads. Backed by a live Temporal demo of a long-running purchase order workflow, she details the shift to a stateless v2 protocol for better reliability at scale. Also from AI Engineer, MCP Apps: Extending the Frontier pitches a new spec that allows servers to return interactive UI components (like a PostHog widget or Shopify view) rather than just textual data over MCP, enabling dynamic widgets inside chat clients.

AI & Machine Learning

In a grounded takedown of current AI metrics, When Will The Benchmaxxing Plague End? from AI Engineer exposes how labs are gaming leaderboards like the Chatbot Arena. Nick Heiner highlights specific technical failures in popular tests—such as data contamination in SWE-bench and unsolvable, contradictory prompts in IFEval—arguing that genuine evaluation requires expensive, high-fidelity human expert analysis. Meanwhile, a brief clip from GitHub, Why OpenClaw feels like the Linux of AI, notes that OpenClaw is being treated as a foundational open-source layer that engineers can deeply modify and build upon, contrasting it with standard, thin API shims.

Hardware & Infrastructure

You’re Underestimating America from the No Priors podcast highlights the overlooked physical bottlenecks in the AI hardware supply chain. The discussion emphasizes massive concentration risks in components beyond GPUs—such as precision reducers, server moss, rare earth magnets, and actuators—particularly in the robotics sector, which is currently dominated by China.

Everything Else

On Lenny’s Podcast, This CPO regrets that product management exists features Whatnot’s Tom Verrilli arguing that the default PM-to-engineer ratio infantilizes technical teams, advising engineers to instead use AI tools to parse data and bypass PM bottlenecks. For career perspectives, Are software engineers really engineers? validates software development as a rigorous discipline, while Grant Sanderson’s Advice for Students highlights teaching as a highly resilient, post-AGI job. Finally, history buffs can catch Lex Fridman’s 5-part Civil War series with historian Gary Gallagher, covering The most important battle of the Civil War, The shocking brutality of the Civil War, Ulysses S. Grant vs Robert E. Lee, how Historian debunks The Lost Cause mythology, and modern political parallels in Extreme political division in US - Are we on the verge of a Civil War?.