Agentic customer service starts with knowledge, not prompts
AI customer service agents are software-driven assistants that execute multi-step customer service tasks using structured knowledge, workflow automation, and integrations, rather than relying on isolated generative responses or human handoffs for every interaction. When firms deploy AI agents without a strong knowledge management system, the result is brittle automation: agents cannot find reliable answers, workflows break, and customer trust erodes. Industry platforms are shifting away from “ask a model a question” toward “give an agent a playbook, context, and authority.” This places knowledge architecture at the center of customer service infrastructure. Agentic workflow automation only works when agents can access up-to-date policies, product information, and transaction history encoded in reusable knowledge objects. Without that base, adding more models or better prompts does not fix failure modes such as inconsistent answers, unresolved tickets, or escalations that still depend on manual effort.
What Optimizely’s agent growth reveals about knowledge and workflows
Optimizely’s Opal platform, though focused on marketing, offers an early look at how structured knowledge and workflows unlock real agent value. Optimizely reports that customers have built more than 4,000 custom AI agents and run over 172,000 executions across marketing workflows. One quotable signal is that more than 97% of activity is driven by customer-built agents, which shows teams are encoding their own playbooks and data into reusable automation. Around 32% of executions involve multi-step tasks, moving beyond single prompts to end-to-end processes such as brief-to-asset or insight-to-test. These patterns map directly to customer service infrastructure: instead of “answer this one question,” agents need orchestrated workflows across ticketing, knowledge bases, and analytics. The lesson for service leaders is clear: build knowledge and process templates that agents can execute repeatedly, and treat AI customer service agents as workflow participants, not content generators.
Zendesk’s resolution vision: outcome-driven knowledge platforms
Zendesk’s recent Relate conference highlighted how customer service platforms are centering their strategies on knowledge and outcomes rather than isolated AI features. Its Resolution Platform brings AI agents, copilots, knowledge, workflows, and governance into one system aimed at closing the loop from interaction to resolution and learning. Zendesk stresses agents that can handle multistep workflows across messaging, email, and voice, spanning front, middle, and back office tasks. According to Forrester, Zendesk’s biggest advantage is the potential to unify service data — tickets, interaction transcripts, and knowledge — into more personalized, context-aware experiences. However, the same analysis notes that the platform still needs to prove it can move from a “collection of tools” to a fully integrated suite with a reliable learning loop. For AI customer service agents, this integration is only possible when knowledge is modeled as shared, governed, and continuously updated assets.
New roles and operational readiness for agentic customer service
As AI customer service agents take on more workflow execution, organizations need new roles and operating practices to keep them reliable. Knowledge architects must design information structures that agents can query consistently, from policy documents to troubleshooting guides. Service operations teams need to curate agentic workflow automation templates that define when agents act, escalate, or learn from outcomes. Platforms like Zendesk are already building role-specific copilots and agent builder tooling, which only pay off if teams treat knowledge as a living product with owners, review cycles, and analytics. Quality assurance, workforce management, and analytics — functions Zendesk strengthened through acquisitions — become part of an integrated feedback loop that tunes both knowledge and workflows. Without this operational readiness, AI agents risk becoming shadow tools that work in pilots but fail under production load, where edge cases, policy changes, and cross-channel complexity are the norm.
Why early investment in knowledge infrastructure wins
The trajectory of platforms like Optimizely and Zendesk shows that early investment in knowledge infrastructure gives companies an edge when deploying AI customer service agents. Optimizely’s 42% quarter-over-quarter ARR growth for its Opal agent orchestration reflects expansion behavior as customers move from pilots to scaled, repeatable agents tied to measurable execution gains. Zendesk’s push toward outcome-based pricing signals a similar shift: value is measured in verified resolutions, which depend on accurate, reusable knowledge and integrated workflows. Firms that codify their processes, build composable knowledge management systems, and connect them to ticketing, analytics, and self-service channels can roll out agents faster and with lower risk. Those that delay this work will find that adding AI on top of messy, undocumented processes only magnifies inconsistencies. In the next wave of customer service infrastructure, knowledge is not an accessory to AI agents; it is the foundation they stand on.