From Better Models to Better Knowledge Infrastructure
Knowledge infrastructure in AI customer service is the combination of structured content, connected data, and governed workflows that AI customer service agents use to understand context, perform multistep tasks, and deliver consistent resolutions at scale. For years, customer service AI was framed as a race for smarter models and more natural language. That era is fading. Most enterprises now have access to capable models, but they struggle to connect those models to reliable knowledge, clean data, and repeatable agentic workflows. This is why agentic customer service success increasingly depends on how knowledge is captured, tagged, and maintained, and how workflows are orchestrated across tools. The differentiator is not whether an agent can answer a single question, but whether it can act as a dependable part of a broader service operation that keeps learning and improving.
Zendesk Relate: Agentic Service Starts with Operational Readiness
Zendesk Relate put a spotlight on how enterprise AI adoption in service now hinges on operational readiness and knowledge structure. The company’s Resolution Platform links AI agents, copilots, knowledge, workflows, and governance into a single system that aims to resolve interactions and feed a learning loop. According to Forrester’s coverage of Relate, Zendesk has assembled this platform through more than 15 acquisitions, spanning QA, workforce management, CCaaS, AI agents, analytics, conversational AI, and knowledge retrieval. That breadth has forced customers to think beyond isolated AI features. New roles, such as knowledge owners and automation designers, are emerging to oversee agent behavior, content quality, and outcome-based pricing models that charge for verified resolutions. The message to customer service leaders is clear: without organized knowledge and aligned operations, even sophisticated AI agents cannot deliver reliable resolutions or measurable business value.
Optimizely Opal Shows Why Agentic Workflows Beat Raw AI Power
Optimizely’s Opal platform offers a practical view of what happens when enterprises move from AI pilots to production agentic workflows. Optimizely reports nearly 1,700 customers using Opal have built more than 4,000 custom AI agents and run more than 172,000 executions across marketing workflows. One notable pattern is that more than 97% of Opal activity is driven by customer-built agents, a sign that teams want automations tuned to their specific stacks and processes rather than generic assistants. Around 32% of executions involve multistep tasks, indicating that agents are now handling end-to-end workflows instead of single prompts. Combined with higher experiment throughput, campaign production, and digital asset reuse, these metrics show that the advantage lies in how well a company structures its data and workflows so AI agents can execute reliably, not in marginal gains from another model upgrade.
Why AI Customer Service Agents Stall Without Strong Knowledge Foundations
As companies push beyond proofs of concept, many discover that AI customer service agents hit limits not because the models fail, but because knowledge infrastructure is weak. Fragmented FAQs, siloed CRM data, and inconsistent process definitions make it hard for agents to maintain context or complete multistep resolutions. Zendesk’s context graph and connectors are a response to this gap, aiming to tie together front-, middle-, and back-office data so agents can work across channels and departments. In parallel, marketing platforms such as Opal show that agents thrive when they have standardized workflow patterns and access to shared assets. These examples underline a common lesson for enterprise AI adoption: the bottleneck is often knowledge management, governance, and integration. Without those in place, agentic workflows remain fragile experiments instead of dependable, measurable parts of the service organization.
Early Infrastructure Choices Will Decide Long-Term Advantage
The shift in AI customer service is now less about who deploys AI first and more about who builds the right customer infrastructure early. Decisions about how to structure knowledge bases, connect data sources, and govern agent behavior will shape long-term competitive positioning. Platforms like Zendesk’s Resolution Platform and Optimizely’s Opal indicate that buyers are rewarding vendors who help design repeatable agentic workflows rather than selling one-off copilots. As John Kelleher of Zendesk notes, customers are tired of generic AI promises and want AI tied to genuine business transformation programs. For enterprises, this means treating knowledge infrastructure as a strategic asset: investing in taxonomies, context graphs, and cross-functional roles that keep agents aligned with service goals. Those that do will turn AI customer service agents into compounding advantages, while laggards remain trapped in pilot mode.