Three enterprises—Citi, Home Depot, and Capcom—have moved AI agents from proof-of-concept into production environments handling real financial transactions, customer purchases, and creative workflows. What distinguishes their deployments is not the technology itself but the governance infrastructure layered beneath it. Citi treats its Citi Sky agent as an employee subject to securities regulations, embedding compliance controls and audit trails into every customer interaction across voice and video channels. Home Depot has standardised its Magic Apron agent across web, in-store, and phone systems using a unified framework, achieving measurable improvements in conversion rates and resolution speed. Capcom has redirected 30,000 hours monthly per project away from repetitive testing tasks, allowing developers to concentrate on creative work rather than regression passes. Each company recognised early that customer-facing agents demand more than model selection—they require architectural decisions about data access, consistency, and accountability.
The implications for CX teams are substantial. These deployments reveal that agent reliability at scale depends on treating AI as infrastructure rather than experimentation, which raises a critical question: how should teams already managing omnichannel support platforms like Zendesk or Freshdesk integrate agentic workflows without fragmenting their existing audit and compliance frameworks? Citi's approach—layering custom safeguards over a base model rather than adopting off-the-shelf solutions—suggests that organisations managing significant customer risk cannot outsource governance to vendors. Home Depot's cross-channel consistency model demonstrates that the real competitive advantage lies not in individual agent performance but in standardising customer experience logic across every touchpoint, a capability that demands deeper integration with backend systems than traditional chatbot deployments. For support leaders, this signals a shift from managing agent responses to managing agent architecture, where data governance, model transparency, and audit trails become as critical as first-contact resolution rates.
The Capcom case introduces a secondary implication often overlooked in CX discussions: agents unlock value not by replacing human judgment but by eliminating the friction that prevents humans from exercising it. When testing automation frees developers to focus on design, the organisation gains creative capacity rather than cost reduction. This reframes how CX teams should measure agent success—not solely through deflection metrics or cost per interaction, but through whether agents are liberating skilled staff to handle higher-value customer problems. The question becomes whether your current support platform architecture can surface the data and decision logic needed for agents to operate autonomously whilst maintaining the human oversight that complex or high-stakes interactions require.
Executives from Citi, Home Depot, and Capcom describe early work with AI agents While AI agents have moved from experimental tools to customer-facing workers in a matter of months, the next challenge is governance and reliability once those agents touch real money, real shoppers, and real creative o