Building agentic contact centers that survive contact with a real support org

Situation

Every contact center vendor now ships an agentic product, and every one of them demos beautifully. A prospect asks a clean question, the agent resolves it, the room nods. Then it goes into production against a real support org, with a knowledge base nobody has pruned in years, several systems of record that disagree about who the customer is, and a queue where a good share of the volume is people who already tried the bot and want a human.

That gap is where I work. The engagements that succeed are the ones where somebody treats the deployment as an integration and change-management problem rather than a configuration exercise.

What I did

Build the whole environment, not just the agent. In Zoom I build contact center environments from the ground up: agentic voice and chat agents, quality management, workforce management, and AI expert assist. The agentic layer is one component in a system that also has to handle staffing, scoring, and supervisor tooling. Teams that buy only the agent discover the rest later, at a worse moment.

Route on intent, then hand off to real systems. The builds use intent-based skill paths that pass the conversation to connected tools and APIs. That’s the difference between an agent that answers and an agent that resolves: without a path to the order, the ticket, or the entitlement, it’s a search box with better manners.

Design the guardrails alongside the workflow. In Dialpad I build the agents themselves: the workflows, the skills, and the guardrails behind a natural, reliable experience. Human-in-the-loop escalation is specified as part of the design rather than bolted on when something goes wrong, because the moment a customer moves from the agent to a person is where the whole deployment gets judged.

Certified to deliver on both platforms. Zoom ZCX/ZVA AI Deployment and Dialpad Contact Center & Agentic Delivery Partner, one for each platform I build on. Real credentials with real scope: each vendor’s own delivery standards are the baseline I work to, rather than a bar I set for myself.

Advise the partner side too. I work with reseller and partner teams on aligning CCaaS and UCaaS solutions to CX outcomes across RingCentral, Twilio, Zoom, and Dialpad, with enough landscape fluency across NICE, Five9, Genesys, and Cisco to say plainly when something else is the better fit. I also build the enablement content that gets partner teams certified and sales-ready, the same work I did at Cloud Elements and Twilio, pointed at a new problem.

Outcome

Currently delivering agentic contact center environments on both Zoom and Dialpad, with each vendor’s delivery certification behind the work.

The through-line from more than a decade in this industry is that the technology was rarely the constraint. Adoption was. The same is true of agentic AI, just faster and more expensively. A badly deployed agent damages the customer relationship at machine speed. Knowing where the routing breaks and where a human has to catch it is the part that takes the background.

If you’re evaluating an agentic build and want a second opinion on the deployment plan before you commit to it, that’s a conversation worth having.