AI agents cannot be trusted to run autonomous back-and-forth coordination — they require constant hand-holding

Summary

Users trying to offload multi-step, multi-party coordination (scheduling, negotiating terms, chasing confirmations) to AI agents report that anything beyond a simple task 'goes sideways.' Instead of waking up to a settled decision, they end up holding the agent's hand step by step, checking every piece, and doing the back-and-forth themselves — the exact manual work they wanted to eliminate. This is the core blocker to the promise of 'AI handles the back-and-forth and only brings me the final decision to approve.' Existing agents have no reliable orchestration for long-horizon, multi-party tasks with a clear 'done' state, so autonomy remains a demo, not a dependable workflow.