AI digest: Agents optimising themselves while humans lose trust
AutoAgent lets AI systems engineer themselves overnight, Netflix releases video object removal tech, and Microsoft admits Copilot is just for entertainment.
This week’s stories highlight the growing gap between what AI can do and what humans are comfortable with it doing.
AutoAgent builds better agents than humans do
AutoAgent is an open-source library that lets AI systems engineer and optimise their own agent configurations automatically. Instead of the usual prompt-tuning grind, it runs benchmarks, analyses failures, and iterates on system prompts without human intervention. This feels like a proper step towards recursive self-improvement, even if it’s just optimising prompts for now.
Netflix open-sources video object removal
Netflix released VOID, an AI framework that removes objects from videos and automatically adjusts the physics in the remaining scene. The clever bit isn’t just erasing things, it’s rewriting how light, shadows, and movement would look without that object. Expect this to show up in every video editing tool within six months.
Microsoft quietly admits Copilot is entertainment
Buried in Microsoft’s terms of service, Copilot is now labelled “for entertainment purposes only”. This legal hedge comes as Americans are using AI more whilst trusting it less, according to recent polling. The disconnect between marketing AI as a productivity tool and legally classifying it as entertainment tells you everything about where we actually are with reliability.
Japan deploys physical AI for real work
Japan is moving physical AI from pilot projects into actual deployment, driven by labour shortages rather than efficiency gains. Robots are filling roles nobody wants rather than replacing existing workers. This pragmatic approach to automation feels more sustainable than the usual “AI will replace everyone” narrative.