AI Agents Shift Labor Into Liability, Not Eliminate It, Analysis Argues

An opinion piece on DEV Community argues that AI agents relocate work rather than remove it, with humans still needed to audit and handle failures. It cites a German security firm's review of Lovable projects, which found tens of thousands of Supabase databases readable without authentication, exposing chat messages, profiles and passport records. It also reports that a Cursor marketplace compromise let trojanized copies of an extension stay live for about 11 hours, potentially reaching 5,000 developers. The author proposes treating agent autonomy as a dial calibrated by reversibility, blast radius and named human accountability. The piece concludes that companies succeeding with agents will be those able to demonstrate control, not those deploying the most of them.