AI news ·
Cyara CEO says AI agents should be managed like employees, not software
Enterprises must manage AI agents like employees, not software, because they fail silently in customer-facing roles-Cyara validates over 350 million customer journeys annually across 140 countries.
Sushil Kumar, CEO of Cyara, argues enterprises should manage AI agents like employees rather than traditional software. His company, which validates over 350 million customer journeys annually across 140 countries, sees a fundamental mismatch between conventional testing and how AI agents actually fail in production.
"Unlike deterministic systems, AI agents make unpredictable decisions, commitments, and may fail silently," Kumar said. He brings over 25 years in enterprise software, including leadership roles at Oracle, Broadcom, and as CEO of RelicX.ai before its acquisition by Harness.
Managing AI agents as a workforce
Kumar's core argument centers on treating agent autonomy as a progression. Enterprises must define clear roles, permissions, and escalation paths for AI agents - the same structural thinking applied to human employees. Granting more responsibility requires evidence of competence, not just successful lab tests.
This shift matters because AI agents operate in customer-facing roles where failures don't produce system errors. They produce dissatisfied customers. A chatbot that confidently gives wrong information hasn't technically crashed, but the business outcome is worse than a server outage.
Production failures as mandatory release gates
Traditional software testing metrics miss these failures entirely. Kumar said enterprises must evaluate outcomes, not just responses. When an AI agent makes a mistake in production, that failure should act as a release gate - blocking further deployment until the underlying issue is fixed. This creates an iterative improvement loop rather than a one-time pre-launch checkpoint.
Cyara's platform focuses on testing, monitoring, and validating AI-driven customer interactions at scale. The company's position in the market gives Kumar direct visibility into how enterprises currently stumble when moving AI agents from test environments to live customer interactions. For professionals working in AI quality assurance, the shift from response validation to outcome evaluation changes what testing workflows look like.
QA teams accustomed to pass/fail metrics now need frameworks that capture customer satisfaction signals, commitment accuracy, and silent failures. This requires different instrumentation and different escalation logic than traditional software pipelines. Teams focused on AI software testing face the same challenge: the test surface area expands beyond code correctness into behavioral reliability.
Why this matters for customer support and operations leaders
If your organization deploys AI agents in customer-facing roles, the employee management model changes how you staff oversight teams. You need people who can define role boundaries, audit decisions, and build escalation paths - not just engineers monitoring uptime dashboards. Kumar's framework means hiring and training for agent management skills that didn't exist in contact centers three years ago. The platform that validates 350 million journeys yearly suggests the scale of failure risk is already material.