Tech Videos — 2026-05-01#

Watch First#

Agents on the Canvas in tldraw — Steve Ruiz, tldraw is the must-watch of the day. Ruiz ditches the tired chat UI by wrapping the canvas in an Electron app that gives Claude direct script-injection access to the DOM, effectively demonstrating what it looks like when AI models break out of the sidebar to collaborate natively on a visual canvas.

Highlights by Theme#

Developer Tools & Platforms#

The Download: Linux 486 retirement, DeepSeek v4, TanStack AI & more by GitHub highlights the alpha release of TanStack AI, an open-source, framework-agnostic toolkit with isomorphic typing that ensures type safety across server and client boundaries. Meanwhile, Strands Agents just shipped TypeScript SDK 1.0 from AWS Developers brings Zod-validated tools and built-in Model Context Protocol (MCP) to the browser, unlocking truly client-side agent execution. On the noisier side, OpenAI’s Bring your work into Codex in a few clicks is a polished 60-second marketing reel showing seamless Slack and Google Drive integrations, but offers zero technical depth on how the underlying context routing is managed.

AI & Machine Learning#

In a high-signal interview on Baseten CEO Tuhin Srivastava on Custom Models, and Building the Inference Cloud, the CEO notes that over 95% of their served tokens now come from custom, post-trained models rather than vanilla open-source weights. To help operationalize these complex multi-stage flows, Shipping complex AI applications — Braintrust & Trainline provides a gritty, hands-on workshop about replacing vibe-based prompts with nested tracing and deterministic scoring functions to prevent autonomous agents from breaking production systems. Additionally, GitHub’s open-source news notes the arrival of DeepSeek v4, a model heavily optimized for agentic tool use that pushes open-weights inference costs down to an order of magnitude cheaper than closed-source frontier models.

Hardware & Infrastructure#

The All-In Podcast’s OpenAI Misses Targets, Codex vs Claude, Elon vs Sam Trial, Big Hyperscaler Beats, Peptide Craze notes that hyperscalers like Amazon, Microsoft, Google, and Meta are projecting over $700 billion in capital expenditures to build out AI data centers, sacrificing their once-legendary free cash flow. However, the actual bottleneck is shifting from GPU supply chains to the raw power grid capacity required to run these massive clusters. For everyday cloud practitioners, You Deleted Everything and AWS Is Still Charging You offers a highly practical reminder to clean up unattached EBS volumes, elastic IPs, and NAT gateways that silently accumulate monthly charges long after instances are terminated.

Everything Else#

Mastering AI Pricing: Flexible & Agile Monetization — Mayank Pant, Stripe points out that pure subscription SaaS pricing is dangerous for AI products due to power users destroying hardware margins, leading 56% of top AI companies to adopt hybrid billing models. On the startup strategy front, The One Thing Nobody Can Copy About Corgi explains how enduring massive regulatory friction to become a licensed insurance carrier—rather than just another API broker—created a nearly unassailable two-year infrastructure moat.