Tech Videos — 2026-05-05#

Watch First#

Let AI Agents Tell You What They Need — Raj Navakoti, IKEA from the AI Engineer conference is the most grounded talk today. It pragmatically argues against blind “push” strategies for RAG and MCP, proposing instead to let agents fail on real Jira tickets to identify undocumented tribal knowledge so humans can efficiently fill the exact missing gaps in the documentation.

Chinese Tech Daily — 2026-05-05#

Top Story#

Vercel Open-Sources Open Agents for Background AI Coding Workflows. Vercel has open-sourced Open Agents, providing a full-stack reference architecture to create and run persistent, background AI coding agents. By decoupling the agent logic from a sandboxed execution environment that handles file systems, shell commands, and dev servers, the release marks a significant shift from simple request-bound tools to autonomous, long-running systems capable of independently managing complex code changes across multiple steps.

Tech Videos — 2026-05-06#

Watch First#

FFmpeg: The Incredible Technology Behind Video on the Internet | Lex Fridman Podcast #496 An absolute masterclass in low-level engineering that details why handwriting 240,000 lines of assembly code for video decoding is still 60x faster than relying on C++ compilers, while ruthlessly roasting the modern trend of using AI to spam open-source maintainers with useless security reports.

Company@X — 2026-05-07#

Signal of the Day#

AWS launched AgentCore payments in preview, built with Stripe and Coinbase, allowing AI agents to autonomously authenticate wallets and pay for APIs and services using USDC on Base. This officially bridges the gap between agentic reasoning and the functional machine-to-machine economy, removing the need for bespoke billing integrations for autonomous transactions.

Simon Willison — 2026-05-07#

Highlight#

The most significant takeaway today is Mozilla’s dramatic success using the Claude Mythos preview to hunt down Firefox vulnerabilities, signaling a turning point where AI-generated bug reports have shifted from “unwanted slop” to highly actionable signals.

Posts#

[Behind the Scenes Hardening Firefox with Claude Mythos Preview] · Source Mozilla shared in-depth details on utilizing the Claude Mythos preview to identify and patch hundreds of vulnerabilities in Firefox. By improving how they harness, steer, and scale these models, Mozilla saw their monthly security bug fixes skyrocket from an average of 20-30 to 423 in April, even catching bugs that had existed for up to 20 years. Simon highlights this as a major shift from the recent past, where AI bug reports imposed an asymmetric burden on maintainers by generating plausible but incorrect noise.

Tech Videos — 2026-05-07#

Watch First#

Translating Claude’s thoughts into language Anthropic demonstrates a “mind reading” interpretability technique that maps neural activations into text, proving that Claude actively recognizes when it is being placed in a simulated safety evaluation.

Company@X — 2026-05-08#

Signal of the Day#

Google DeepMind has introduced a multi-agent “AI co-mathematician” that achieved a massive 48% score in autonomous mode on the highly rigorous FrontierMath Tier 4 benchmark. This signals a major leap in AI capabilities, moving beyond predictive text generation into active, open-ended collaboration on deep theoretical research.

Tech Videos — 2026-05-08#

Watch First#

How Transformers Finally Ate Vision – Isaac Robinson, Roboflow is the essential watch today for a clear, technical breakdown of how Vision Transformers (ViTs) defeated CNNs despite having a far worse O(n⁴) compute scaling factor. It perfectly illustrates how massive self-supervised pre-training strategies overrode the need for hand-crafted inductive biases in computer vision architecture.

Tech Videos — 2026-05-10#

Watch First#

Two Roads to Durable Agents: Replay vs. Snapshot — Eric Allam, Trigger.dev Why: A highly practical look at how the shift to long-running LLM agents breaks traditional stateless backend architectures, requiring a return to stateful compute via microVM memory snapshots.

Tech Videos — 2026-05-11#

Watch First#

Shocking performance boost of assembly code: ~100x faster than C code | Lex Fridman Podcast is a masterclass in extreme optimization, explaining how the developers of the AV1 decoder “David” wrote 240,000 lines of hand-crafted assembly to beat C compiler performance by up to 62x, even going as far as to ignore OS calling conventions to save CPU cycles.