Tech Videos

Watch First

The Workflow of the Future With Zed on Syntax is the single most critical watch for senior engineers today. It cuts through the typical agent marketing to detail Delta, a new agent-focused editor built on a custom database layer (Delta DB) and conflict-free replicated work trees that virtualize Git repositories, enabling real-time developer-agent co-authoring with zero lag.

Highlights by Theme

Developer Tools & Platforms

In The Workflow of the Future With Zed (Syntax), Nathan Sobo explains how Delta virtualizes the work tree using copy-on-write persistent B-trees and maps rendering tasks directly to the GPU via their custom GPUI Rust framework. For corporate Java environments, Getting started with your first Java app (Visual Studio Code) demonstrates how the Spring Boot extension pack integrates the Model Context Protocol (MCP) to publish backend services as AI tools. On-device developers also get minor updates: 🎙️ Dictate in Multiple Languages in #VSCode (Visual Studio Code) showcases local multilingual dictation using on-device speech-to-text modules, while Slack Developer Sandboxes | Slack (Slack) introduces free, isolated enterprise environments for safe API prototyping away from production code.

AI & Machine Learning

Physical AI and continual learning (CL) dominated the technical talks today. In Chelsea Finn: This is the State of the Art in Robotics (Y Combinator), Finn demonstrates Physical Intelligence’s PIO7 generalist model, showing how metadata and subtask prompting achieve zero-shot skill transfer across different robot joint configurations and unseen appliances. On evaluation, UC Berkeley’s Beyond Static Intelligence: Evaluating Continual Learning — Parth Asawa, UC Berkeley (AI Engineer) introduces Continual Learning Bench 1.0 and a ‘Gain’ metric designed to isolate stateful, online learning from baseline model strength, complemented by Sakana.ai’s Memory Harnesses for Long-Running Research Agents — Stefania Druga, Sakana.ai (AI Engineer) showing that ranked-recall policies on local M3 Ultra hardware dramatically reduce token context costs and drift. To scale these CL pipelines, Scaling up Continual Learning — Ronak Malde, Trajectory (AI Engineer) details On-Policy Self-Distillation (OPSD) but flags a critical ‘buttweight problem’ where long-horizon models over-diverge and collapse into filler hedging words. On inference, Meta’s new model wants “deep access” to your personal life… (Fireship) breaks down Muse Glimmer, a distilled 30B open model optimized for consumer GPUs via 4-bit quantization and speculative decoding.

Hardware & Infrastructure

AI infrastructure remains severely undersupplied, with Morgan Stanley’s Michelle Weaver warning in Why Open-Weight AI Models Won’t Kill the Closed Model (Bloomberg Tech) of a 40-gigawatt power shortfall through 2028 alongside specialized labor bottlenecks. To hedge these massive capital expenditures, Carmen Li explains in Silicon Data Raises $30.5M to Benchmark AI Compute (Bloomberg Tech) that the CME plans to list GPU futures starting October 5, cash-settled against independent H100 indices to manage price volatility. Meanwhile, hyper-scalers are printing high margins; CoreWeave, Supermicro Lead Tech Stocks Higher on Results | Bloomberg Tech 8/12/2026 (Bloomberg Tech) highlights Supermicro’s massive $15B revenue guide and CoreWeave’s lucrative strategy of extraction pricing on older A100 inventory.

Everything Else

In Stop being skeptical about AI for development with Charity Majors (The Pragmatic Engineer), Charity Majors warns of a quiet dip in code quality across the industry and argues that since code generation is cheap, engineers should treat ‘code as cash’—replacing broken functions entirely rather than editing them in place. On consumer tech, Marques Brownlee’s Google Pixel 11/Pro/Fold Impressions: It Is What It Is (Marques Brownlee) critiques Google’s incremental upgrades and ‘Ramageddon’—a cost-cutting move that dropped starting Pro models from 16GB to 12GB RAM. Finally, we get management and cultural perspectives from Why Airbnb doesn’t have a CPO | Brian Chesky (CEO of Airbnb) (Lenny’s Podcast) on corporate reorganization, alongside Khabib Nurmagomedov: Dagestan, MMA, UFC, Islam, Conor, Fedor & Football | Lex Fridman Podcast #500 (Lex Fridman), which delivers an exhaustive masterclass on spartan work ethic and remaining grounded under massive success.

🔮 If you’d like, I can do some web research to find out how other financial institutions are responding to CME’s upcoming GPU futures and what the current sentiment is around this new risk-management layer.