Tech Videos — 2026-03-29#

Watch First#

From skeptic to true believer: How OpenClaw changed my life | Claire Vo This is a rare, pragmatic deep-dive into orchestrating local open-source AI agents that cuts through the peak hype cycle to show actual utility and system limitations,. It offers credible, technical advice on sandboxing, managing context windows, and avoiding prompt injection risks when giving models access to your local machine,,,.

Highlights by Theme#

Developer Tools & Platforms#

In her Lenny’s Podcast interview, Claire Vo details how she runs nine separate OpenClaw agents on dedicated Mac Minis, utilizing basic local cron jobs, markdown-based identity files, and strict API scoping to avoid context overload,,,,. Meanwhile, Chip Huyen argues on The Pragmatic Engineer that AI is breaking traditional developer workflows, noting that standard line-by-line pull request reviews fail when junior engineers submit AI-generated code they don’t fully understand,,. Huyen also questions the artificial boundary between terminals and IDEs in an AI-driven environment, pointing out that both simply take instructions and output code,.

AI & Machine Learning#

Andrej Karpathy makes a brief but vital architectural point on No Priors: AI, Machine Learning, Tech, & Startups, stating that developers should stop building consumer web UIs and instead expose raw APIs for LLM agents to orchestrate,. Corroborating this friction, Claire Vo points out that current “browser-use” features for agents are highly unreliable because the modern web is hardened with anti-bot defenses,. She advises engineers to stick to headless web search tools like Exa or native APIs to bypass the hostile web rather than depending on raw browser automation,.

Hardware & Infrastructure#

On Lex Clips, Jensen Huang provides a systems-level breakdown of NVIDIA’s shift to “extreme co-design,” explaining that scaling distributed workloads across 10,000 GPUs creates massive bottlenecks requiring concurrent optimization of CPUs, GPUs, networking, and power cooling,,,. Huang also discusses xAI’s Colossus supercomputer deployment in a separate Lex Clips video, attributing its record four-month build to Elon Musk’s “speed of light” first-principles engineering and presence at the point of action,,,. In another Lex Clips segment, Huang contextualizes NVIDIA’s current hardware dominance by detailing the near-existential bet of putting CUDA on consumer GeForce GPUs to brute-force a massive developer install base,,,.

Everything Else#

Terence Tao warns on Dwarkesh Patel that hyper-optimizing our schedules and relying entirely on targeted AI search destroys the serendipitous discovery essential for deep research,. On the All-In Podcast, David Friedberg argues that as AI floods us with infinite digital supply, human success will depend entirely on personal agency and the ability to intentionally reject algorithmic feeds,. Finally, Jensen Huang briefly notes that Doom remains the most influential game in history for transforming the PC from an office tool into a consumer device.