Tech Videos — 2026-06-27#

Watch First#

Build a multi-agent system: A2A & Agent Registry from Google Cloud Tech is the most practical watch today, demonstrating how to standardize multi-agent systems without hardcoding custom glue. It shows how treating agents similarly to HTTP servers and using a central Agent Registry can successfully manage the fragmentation of sprawling, real-world AI deployments.

Highlights by Theme#

Developer Tools & Platforms#

GitHub’s pitch in Why you need the new GitHub Copilot desktop app addresses the headache of agent context fragmentation by isolating parallel Copilot sessions into separate git work trees to prevent them from stepping on each other. It features an “agent merge” capability that automatically guides pull requests through your team’s existing CI checks and required reviews. On the web framework side, Google Cloud Tech’s Building a dynamic personal site with Antigravity and Gemma 4 shows how to generate vector embeddings at compile time to serve dynamic offline recommendations without needing a live backend server. Additionally, ThePrimeagenHighlights covers practical developer integrations in My New AI Workflow.

AI & Machine Learning#

The All-In Podcast’s Socialists Sweep NYC, China Catches Up in Coding, AI Memory Crunch, Micron’s Blowout Quarter highlights the release of GLM 5.2 by China’s Z.AI, a massive open-weights model featuring 744B parameters and a 1M context window that currently beats GPT 5.5 on the SWE coding benchmark. The hosts pragmatically note this rapid catch-up is largely achieved via distillation and harvesting reasoning traces from Frontier APIs, proving that composable models mixing frontier and open-weights runtimes are the likely enterprise future. Google Cloud Tech’s A2A video is technically notable for establishing “Agent Cards” (metadata identities) to dynamically route tasks between disparate tools and agents via an MCP server. Also in the AI space, NVIDIA Developer dropped What 5,000 Kagglers Taught Us About Improving AI Reasoning | Nemotron Labs, while Y Combinator explored emerging tech markets in India Can Create The Largest AI Companies.

Hardware & Infrastructure#

The hardware segment of the All-In Podcast episode digs into Micron’s massive earnings growth, asserting that HBM (High Bandwidth Memory) DRAM remains the definitive physical bottleneck for scaling AI data centers. It also unpacks the highly speculative but theoretically viable economics of “orbital compute,” where SpaceX’s reusable Starship could hypothetically make space-based data centers (with an estimated $5B launch cost) economically competitive against terrestrial $60B gigawatt deployments that are currently struggling with extreme power and cooling inflation.

Everything Else#

For those interested in historical philosophy, Dwarkesh Patel interviews Ada Palmer in What Machiavelli thought freedom actually was - Ada Palmer. If you tolerate political segments in your tech pods, the All-In Podcast includes a lengthy debate on the Democratic Socialists of America sweeping New York City primaries, framing the movement as a localized reaction to downward mobility and economic stagnation.