Tech Videos — 2026-06-02#

Watch First#

How Lovable self-improves every hour — Benjamin Verbeek, Lovable: A highly pragmatic look at continuous agentic learning in production, showing how Lovable gives their AI a “vent tool” to directly report API friction, bad schemas, and platform incident alerts into a developer Slack channel.

Highlights by Theme#

Developer Tools & Platforms#

Google Cloud Tech released two solid, no-nonsense architecture demos showing the Model Context Protocol (MCP) in action. Automate project intake with multi-agent AI using MCP, Google ADK, Cloud Run, and BigQuery demonstrates an orchestrator and four specialist agents dynamically writing SQL at runtime against a BigQuery MCP server. A companion video, Automate SAP order-to-cash with multi-agent AI & Google Agent Development Kit (ADK), shows a similar multi-agent setup interacting with SAP’s OData endpoints, highlighting the power of parallel tool calling in ADK v1.1. On the NVIDIA side, How We Built Self-Evolving Hermes Agents With NVIDIA NemoClaw demos deploying an internal Noose Research Hermes agent securely via the OpenShell sandbox to establish trust boundaries that prevent internal data leakage.

AI & Machine Learning#

The AI Engineer Melbourne 2026 Keynote Livestream | Day 1 offers a goldmine of signal, particularly Notion’s engineering lead Sarah Saxs detailing their multi-model routing strategy to avoid vendor lock-in and optimize token costs across different cognitive tasks. In Task Fidelity Scaling Laws — Kobie Crawdord, Snorkel, Snorkel shares empirical RL training data showing a massive 5x performance uplift (a 6% vs 1% improvement) when models hill-climb on high-quality, explicitly verifiable tasks rather than lower-quality data. Finally, the OpenAI CFO Sarah Friar on IPO, AI Rivalries, New Device, and Spending $100B+ on Compute episode of the All-In Podcast is worth the time for its sheer macro-scale business realities, revealing OpenAI’s multi-CSP strategy to shift CapEx to OpEx and a multi-chip pipeline (Nvidia, AMD, Cerebras, Broadcom) to overcome the massive compute deficits expected through 2026.

Hardware & Infrastructure#

If you’re curious about local inference capabilities, Architectural Design With Agents on NVIDIA RTX Spark features an impressive demo of an agent autonomously manipulating local instances of Rhino and Blender via the Hermes harness to perform CAD tasks. Elsewhere, the Conversation with Qualcomm President and CEO Cristiano Amon: Satya Nadella at Microsoft Build 2026 touches on the architectural shift toward power-efficient CPUs and dense sensor arrays required to run continuous, real-time agentic context across edge devices.

Everything Else#

Much of the remaining upload volume from the day was pure marketing fluff, corporate vlog content, or placeholder music (e.g., Apple Design Awards, Google Dublin office tours, Microsoft’s “Project Solara”). However, we have to respect the absolute absurdity of the Official Keynote Closing Video | GTC Taipei 2026, where NVIDIA summarized their new Verub chip architecture and NVLink capabilities via a fully produced rap video.