Conviva CEO on Turning Experimental AI Agents Into Reliable Systems

In this episode of Tech Talks Daily, I sat down with Keith Zubchevich, CEO of Conviva, to unpack one of the most honest analogies I have heard about today’s AI rollout.

Keith compares modern AI agents to toddlers being sent out to get a job, full of promise, curious, and energetic, yet still lacking the judgment and context required to operate safely in the real world. It is a simple metaphor, but it captures a tension many leaders are feeling as generative AI matures in theory while so many deployments stumble in practice.

As ChatGPT approaches its third birthday, the narrative suggests that GenAI has grown up. Yet Keith argues that this sense of maturity is misleading, especially inside enterprises chasing measurable returns. He explains why so many pilots stall or quietly disappoint, not because the models lack intelligence, but because organizations often release agents without clear outcomes, real-time oversight, or an understanding of how customers actually experience those interactions.

The result is AI that appears to function well internally while quietly frustrating users or failing to complete the job it was meant to do.

We also dig into the now-infamous Chevrolet chatbot incident in which a $76,000 truck was sold for one dollar, using it as a lens to examine what happens when agents are left without boundaries or supervision.

Keith makes a strong case that the next chapter of enterprise AI will not be defined by ever-larger models, but by visibility. He shares why observing behavior, patterns, sentiment, and efficiency in real time matters more than chasing raw accuracy, especially once AI moves from internal workflows into customer-facing roles.

This conversation will resonate with anyone under pressure to scale AI quickly while worrying about brand risk, accountability, and trust. Keith offers a grounded view of what effective AI “parenting” looks like inside modern organizations, and why measuring the customer experience remains the most reliable signal of whether an AI system is actually growing up or simply creating new problems at speed.

As leaders rush to put agents into production, are we truly ready to guide them, or are we sending toddlers into the workforce and hoping for the best?

Useful Links