A look at the use of conversational AI in US call centers, which some agents find helpful and others say leaves the humans with more complex, intense workloads
Lisa Bannon / Wall Street Journal :
Context & Ripple Effects
Call centers have been automating at the front door since the pandemic: as budget cuts hit in 2020, operators turned to chatbots like IBM's Watson to filter calls and cut human operator headcount pandemic-era chatbot filtering. What the WSJ documents now is the next phase — conversational AI sitting alongside the surviving agents rather than replacing them outright.
The arc is visible in the AT&T coverage from mid-2023, where a call center worker using AI that transcribes calls and suggests technical solutions openly wondered whether she was training her replacement. By 2025, agents across Australia, Canada, Greece, and the US report being repeatedly mistaken for AI themselves, showing how blurred the human-machine line has become on live calls agents mistaken for AI.
First-order effects
- Agents handling AI-assisted calls get real help on routine queries but are left holding the complex, intense conversations the tool can't resolve — raising workload intensity even where volume falls.
Second-order effects
- Every transcribed, AI-supervised call doubles as training data, so the same tool that assists an agent today sharpens the case for staffing fewer of them tomorrow; the AT&T worker's 'training my replacement' worry is the mechanism made explicit.
Third-order effects
- As AI takes the routine tier, the agent job structurally bifurcates into high-complexity human specialists plus a thinning routine workforce — while humanlike chatbots create trust problems for callers trying to tell which side of the line they're talking to.
The trend: Call centers are moving from chatbots that deflect calls to embedded AI copilots that reshape what remains of the human agent's job — assistance now, substitution later.