Agents require narrow, specific, structured tasks with strong guardrails—broad autonomy with loose constraints immediately causes unpredictable failure even in virtual environments, so real enterprise use of autonomous agents is years away from the promises being made.