AI VIDEOS
One-sentence takeaways and concise summaries of important AI videos.
Just-in-Time Ontologies: What Cheap Intelligence Means for Data - Hrishi Olickel, Southbridge AI
Hrishi Olickel argues that reliable AI data pipelines should build decision-specific structure, separate model roles and test their outputs against explicit failure criteria.
Please... Enough with the Sloppy, One-Shot SaaS Replacements - Mahmoud Abdelwahab, Railway
Mahmoud Abdelwahab explains how reusable architecture, deliberate interaction design and hands-on verification turn AI-generated internal tools into software people actually want to use.
Unboxing the Surface Laptop Ultra With RTX Spark. Will I Switch?
Ray Fernando examines the hardware of a pre-release Surface Laptop Ultra with RTX Spark, but cannot yet test whether it delivers the local AI performance that would make him switch from a Mac.
Your Agent Is a Lethal Trifecta (And What to Do About It) - Kim Maida, Keycard
Kim Maida demonstrates how delegated identity and short-lived, task-scoped credentials can constrain an agent's access to private data and external tools while preserving useful autonomy.
How Financial Planners Build and Review AI Agent Skills
Matthew Ragone explains how Range’s financial planners build reusable AI skills from their own workflows while keeping every client recommendation under human review.
Ben Affleck’s AI Explanation: Tensors, Vision and Transformers
The Pretrained Pod uses Ben Affleck’s explanation of visual-effects machine learning to clarify tensors, convolutional networks and vision transformers.
AI Adoption and Small-Business Hiring - Nich Tremper
Nich Tremper discusses Gusto research linking AI adoption with faster hiring at small businesses, especially firms with fewer than ten employees.
Building a Managed Research Agent with Claude
This workshop builds a Claude-managed research application using streamed events, specialized subagents, reusable skills, shared memory and rubric-based outcome checks.
Claude Managed Agents: Evals, Isolation and CI/CD
Anthropic's workshop Q&A explains how managed-agent applications can version resources, evaluate behavior, isolate customer data and coordinate long-running tasks.
Scheduling Agents and Improving Memory with Dreaming
Anthropic demonstrates recurring managed-agent deployments and experimental memory improvement, then discusses observability, evaluation and access-control limits.
Agent Identity on Existing Communication Networks
Adi Singh argues that persistent contact points and existing communication networks can help agents gain useful identities, while introducing new security risks.
Separating Agent State, Harnesses and Sandboxes
Anthropic's engineers explain why Claude Managed Agents separates durable session history from replaceable execution components and introduce its core APIs.
Agent-Native Registration with auth.md
Michael Grinich explains an open proposal for discovering signup requirements and registering agents as legitimate service users.
Inference Tradeoffs from Batching to Multi-Silicon
Daniel Kim and Michelle Nguyen explain how runtime optimizations and heterogeneous hardware can trade off token latency, throughput and memory use.
Finding Agent Failures at the Sandbox Boundary
Kevin Orellana describes using coding agents to exercise cloud sandboxes and distinguish interface bugs from resource and permission limits.
How AI Businesses Build, Price and Sell Globally
Nilofer Rajpurkar uses Stripe's observations and an agent-assisted payments demo to discuss faster launches, global sales and evolving AI pricing.
Four Ways to Isolate Coding Agents
Robert Brennan compares four ways to isolate coding agents, showing how environment lifetime, access scope and triggers change the risks of autonomous work.
Why Production AI Needs More Than Stronger Language Models
Mario Claudio Martone explains how mixed modeling, internal evaluations and full-stack experiments support consequential housing and healthcare workflows.
Why AI Bills Rise While Equivalent Capability Gets Cheaper
Theo Browne argues that higher AI bills can coexist with falling costs for equivalent capability because users run more demanding tasks and consume more tokens.
Building Reinforcement Learning Tasks for AI Research
Jennifer Zhou explains how Preference Model designs verifiable machine-learning tasks, hardens reward functions and evolves environments as agents improve.