Synopsis: In this Techstrong.ai Leadership Insights interview, Tray.ai CEO Rich Waldron examines why deploying AI agents alone is not enough to deliver meaningful business outcomes.
Waldron separates “agent success” into two distinct questions that too often get blended together. First is technical execution: Is the agent doing what it’s supposed to do accurately and verifiably? Second is business impact: Even if the agent works, is it producing measurable value? Without that distinction, organizations can end up optimizing prompts and workflows while never answering the real question—what improved and by how much.
A major theme is cost control. Vizard notes that many early adopters run agents multiple times, throw away most of the output, and keep the “best” response—an approach that quickly becomes expensive at scale. Waldron argues the fix isn’t just better prompts; it’s engineering discipline. He points to two levers: tight feedback loops that teach agents what “good” looks like, and narrower scopes that reduce variability and error rates. In Waldron’s framing, agents behave less like traditional software and more like high-powered interns—capable, but needing ongoing guidance to become reliable.
On the value side, Waldron advocates measuring ROI the same way teams should measure any serious automation effort: pick a business problem, define success metrics, and compare results before and after deployment. He cites practical examples like support-ticket deflection and time saved for account teams preparing for customer calls—metrics that can translate directly into labor efficiency and, ultimately, revenue impact.
Waldron also warns of an “agent divide.” Organizations that simply “open the budget and let everyone experiment” may see inconsistent outcomes. The winners will be those that treat agent deployment like a real software initiative—balancing deterministic workflows with agentic components, investing in context engineering, and building governance that keeps agents aligned with business objectives.
His bottom line: there’s no magic wand for AI agents. The outcomes depend on execution—training, monitoring, integration, and measurement—so automation efforts stay connected to real business value.
