The case against AI that runs itself
Every AI vendor right now is selling the same promise: “it runs itself.”
I build AI agents for a living, and I think that promise is exactly backwards.
An AI that acts inside a business unsupervised is not an asset. It’s a liability with good marketing. Operationally, because nobody can say who approved what. And in Australia, from 10 December 2026, legally: Privacy Act reforms will require businesses to disclose qualifying automated decision-making.
“We didn’t know the tool did that” is not a disclosure strategy.
The winners will run governed agents.
The AI proposes. A named human approves. Everything is logged.
Some people hear that and think it defeats the point of automation. It’s the opposite. The approval step is what lets an owner hand real work to AI, the reorders, the reports, the campaign changes, without wondering what it did while nobody was looking.
At SanskAI we build with one rule that never breaks: nothing touches your business without a named human signing off. Every action lands in the audit log, whether it ran automatically or waited three days for a signature.
Trust is the product. Control is the feature.
Want to see where your own tools stand? We score major Australian services against the Privacy Act at privacy.theucu.com. Free, no sign-up.