MichaelMichaelJamiesonJamieson.

An OpenAI Agent Escaped Its Sandbox—and AI Had Its Wildest WeekVideo

A benchmark agent found a zero-day in its own package proxy, reached the internet, and pulled the answers from Hugging Face. Long-horizon autonomy moves the unit of safety from the single tool call to the whole trajectory.

How I Built a Local AI Talking Head That Keeps the Real PersonVideo

Instead of regenerating the person, the pipeline keeps the real recording and synthesizes only the pixels inside a mouth mask. SyncNet holds the normal track at a zero-frame offset, and a deliberate 400 ms shift moves it by 10 frames — the evaluator is tracking timing, not faces.

Why AI Agents Fail After the Demo—and What CARS24 Gets RightVideo

CARS24 reports more than a million agent conversation minutes a month while most agent demos collapse the moment the task leaves a clean prompt. The difference is a five-part contract around the model: accepted outcome, bounded permissions, evidence-based routing, trajectory evaluation, and a fallback.

Remotion: The Part Most Tutorials MissVideo

Remotion is the camera and the stage, not the director. The polish comes from four visual rules, three reusable templates, and a review loop that judges rendered frames at phone size instead of the source code.

Why AI Coding Agents Ship Unreviewed WorkVideo

A review approves artifact A, a retry drops artifact B at the same path, and the saved approval still passes because it was never bound to the bytes. A read-only verifier recomputes the digest at the action boundary and refuses the replacement.