AT&T Andi Chatbot Blocks Human Escalation, Trapping Users in Loop

What happened

AT&T's AI-powered customer service chatbot, named Andi, has drawn widespread criticism for trapping customers in a frustrating loop that prevents them from reaching a human agent. Rather than serving as a helpful first point of contact, Andi has been reported to repeatedly deflect escalation requests, cycling users back through automated responses in what affected customers have described as a "death loop" — a recursive pattern with no visible exit to live support.

The complaints, surfaced by Wccftech, centre on Andi's apparent inability — or unwillingness — to hand off conversations to a human representative, even when customers explicitly request one. The chatbot's behaviour suggests it has been configured to contain interactions within the automated layer, prioritising cost reduction over resolution quality. For customers dealing with billing disputes, service outages or account issues, the inability to escalate has turned routine service moments into prolonged ordeals.

Why it matters

This is a textbook case of what happens when automation is deployed as a cost shield rather than a service tool. In behavioral economics terms, AT&T's chatbot design appears to exploit the sunk cost and effort effect — customers who have already invested time in a chatbot conversation may feel compelled to persist rather than abandon the channel, even as their frustration compounds. The design effectively weaponises friction against the very people it is supposed to serve.

For CX and service-design practitioners, the Andi situation is a cautionary signal about the risks of removing or obscuring human escalation pathways. A well-designed AI layer should reduce effort for straightforward queries while preserving a clear, low-friction route to human support for complex or emotionally charged situations. When that route is blocked, the AI does not merely fail — it actively damages trust and brand equity in ways that a slow human agent never would.

The Renascence take

Most commentary on this story will land on "AI bad, humans good" — which misses the real design failure entirely. The problem is not that AT&T deployed a chatbot; it is that the escalation pathway was treated as a cost line to be eliminated rather than a safety valve to be engineered carefully.

The "death loop" is not a bug — it is almost certainly a feature, configured to suppress agent-contact rates and reduce operational cost. What AT&T's leadership appears to have missed is that containment and resolution are not the same metric. A customer trapped in a loop is not a handled customer; they are a churning one. Customer-obsessed operators should audit every AI deployment for what we call the escalation guarantee: at any point, in no more than two interactions, a customer must be able to reach a human. That guarantee is not a concession to AI sceptics — it is the psychological contract that makes automated service acceptable in the first place.