Pendo Launches Tool to Monitor AI 'Zombie Agents' in CX

What happened

Digital adoption platform Pendo has launched a new offering designed to help enterprises monitor the AI agents embedded in their products, targeting what the company calls "zombie agents" — AI agents deployed inside digital experiences that operate without real-time visibility into how users are actually behaving. The tool is aimed at giving product and CX teams the ability to track agent performance and ensure agents have live context about user behaviour when resolving issues, rather than acting on stale or incomplete data.

According to reporting from Diginomica, the offering responds to a growing problem as enterprises rush to embed AI agents into customer-facing and internal workflows: many of these agents are shipped and then left largely unmonitored, meaning organisations often have limited insight into whether the agents are helping or quietly degrading the experience they were built to improve.

Why it matters

The launch reflects a wider shift in enterprise AI adoption: the initial wave of excitement around deploying agents is giving way to a more operational concern about governance, observability and accountability. As agents take on more customer- and employee-facing tasks, organisations need the same kind of monitoring discipline they apply to any other production system — visibility into what the agent is doing, why, and whether it is actually helping the person on the other end.

For leaders running digital transformation and CX programmes, this points to agent monitoring becoming a standard requirement alongside agent deployment itself. An agent without real-time context on user behaviour risks making decisions that feel disconnected from what the customer is actually experiencing — undermining trust in exactly the kind of automation that was meant to build it.

The Renascence take

The idea of a "zombie agent" is a useful shorthand for a problem behavioural science has long understood: automation that looks functional on paper can still feel broken to the person using it, because functionality and experience are not the same thing.

Most organisations measure whether an AI agent completed its task, not whether the customer felt understood while it did so — and that gap is where trust quietly erodes. An agent that resolves a ticket without registering the customer's frustration, or repeats a question the person already answered elsewhere in the journey, is technically working and experientially failing. The operators who get ahead here won't just monitor agent uptime; they'll instrument agents for behavioural signals — hesitation, repetition, sentiment shifts — the same cues a good human agent picks up on instinctively. Treat agent observability as a customer experience discipline, not just an engineering one.