Tech Videos — 2026-07-30#

Watch First#

Jeff Dean: The 1% Rule for Building in AI is the clear standout, featuring a deeply pragmatic discussion on the physics of data movement, hardware efficiency, and why context engineering often beats model scaling for long-running agentic systems.

Highlights by Theme#

Developer Tools & Platforms#

GitHub launched native support for stacked pull requests, featuring atomic merges and a terminal UI via the gh stack CLI, as discussed on the GitHub channel’s Stacked PRs are LIVE on GitHub! | Rubber Duck Thursdays. This offers a critical structural fix for reviewing large, agent-generated code diffs sequentially. Google Cloud Tech explored context management in Skills vs MCP: How AI tools have evolved, explaining how “Skills” use progressive disclosure to load prompts into an LLM’s context only when needed, avoiding the token bloat often seen with the Model Context Protocol. Additionally, Google for Developers detailed the stable release of Android Studio Quail 2, the Build with Gemini XPRIZE, and more! - Google Developer News July 2026, which notably shifts Leak Canary heap analysis off constrained test devices and natively onto the host machine.

AI & Machine Learning#

At the AI Engineer summit, Shawn Chan delivered a highly necessary reality check in Build for the Memo, Not the Demo — Shawn Chan, China Resources Holdings, arguing that enterprise AI fails when it optimizes for fluent text rather than the audited, receipt-backed trust required by actual decision-makers. On the architecture side, Divakar Kumar’s talk Let’s integrate AI Agents in Event-Sourced Systems — Divakar Kumar, FlyersSoft broke down how to build materialized views across bounded contexts so a saga orchestrator can route complex cases to a multi-agent tier for real-time fraud detection. In physical AI, Google DeepMind unveiled new generalist models in Gemini Robotics 2 brings whole body intelligence to robots, showcasing impressive fine-motor dexterity like tying knots and dynamic multi-robot collaboration run by independent, localized neural networks.

Hardware & Infrastructure#

Bloomberg Tech featured K2 Space Lands $500 Million to Scale Satellite Production, detailing the startup’s plan to mass-produce massive 100-kilowatt class satellites for just $15 million each, specifically aiming to enable high-compute orbital data centers. Qualcomm’s CEO outlined a distinct shift away from mobile handsets in Qualcomm Sees AI Driving Its Next Growth Wave, highlighting a $3.5 billion edge/industrial AI pipeline and the active development of custom data center ASICs for two major hyperscalers.

Everything Else#

For a historical palate cleanser, Computerphile explored the 1960s symbolic AI era in ITS, Eliza & MiT’s AI Lab - Computerphile, showcasing a restored recording of the SHRDLU 3D block-world parser—a stark reminder of early AI breakthroughs that eventually preceded the first AI winter.