In a compelling session at the 2026 ICMI Digital Event, Brent Barbara and Amanda Hughes from Clearest Blue LLC demystified Agentic AI, providing contact center leaders with a practical framework for understanding and implementing this transformative technology.
The Fundamental Shift
The session opened with a critical distinction: Agentic AI is about action, not just answers. While traditional conversational AI focuses on responding to customer questions, Agentic AI is designed to complete tasks end-to-end. As Hughes explained, the shift is from asking "What should I say back?" to "What needs to happen to resolve this?"
This represents a fundamental evolution in how automation works within customer experience environments. Rather than simply providing information or routing customers, Agentic AI systems can verify identity, check systems, apply business rules, and complete transactions—all within a single interaction.
A Practical Example
To illustrate the difference, Barbara and Hughes walked through a common scenario: rescheduling a package delivery.
With traditional automation, a bot might gather basic information but ultimately hand off to a human agent or redirect the customer to a web page. The bot helped, but didn't complete the task.
With Agentic AI, the system can:
- Verify customer identity
- Check the delivery system to confirm the existing order
- Apply business rules (such as no Sunday deliveries)
- Confirm the request change is eligible
- Send confirmation to the customer
- Log the interaction in the CRM
All of this happens within the same interaction, without requiring the customer to look up tracking numbers or navigate multiple channels.
Where Agentic AI Creates Value Today
The speakers emphasized that the earliest production value appears in high-volume, repeatable, bounded workflows—interactions customers want completed quickly and accurately rather than discussed at length. These include:
- Status requests and delivery updates
- Appointment scheduling and changes
- Simple exception handling
- Password resets and account inquiries
- Proactive follow-up and notifications
Hughes noted that success isn't defined by whether the AI sounds impressive, but by whether customers accept it, employees trust it, and workflows perform as designed.
The Operating Model: Human-in-the-Loop to Human-on-the-Loop
A key operational insight shared was the progression from "human-in-the-loop" to "human-on-the-loop":
Human-in-the-loop means the system does much of the work, but a person reviews or approves key decision points before certain actions are completed—particularly for identity-sensitive changes, refunds above thresholds, or policy exceptions.
Human-on-the-loop represents supervision at scale, where AI operates within defined boundaries while humans monitor outcomes, investigate exceptions, and improve workflows based on evidence.
This approach allows organizations to move faster without sacrificing control or customer experience quality.
The Technical Foundation
The session provided clarity on three critical technical concepts:
Probabilistic Layer: Helps the system understand requests flexibly (recognizing that "I need to change my delivery" and "I can't be there for that delivery window" are versions of the same request)
RAG (Retrieval Augmented Generation): The "look-up layer" that pulls in the right information at the right moment—current order details, available windows, company policies
Deterministic Layer: The rules and control layer where businesses define what AI is allowed to do, what must be verified, and when escalation is required
As Hughes emphasized, "The model's language and reasoning is probabilistic—there's variability. The controls around actions must be deterministic—clear, predictable, auditable."
Channel-Agnostic and Always-On
Barbara highlighted that well-designed Agentic systems operate consistently across all channels—voice, web chat, text, email, social messaging, and dozens of global messaging platforms. The system maintains context across channels, allowing customers to start in chat, switch to voice, and follow up via email without losing the thread of the conversation.
Additionally, these systems work 24/7/365, handling interactions in multiple languages with automatic translation capabilities that go far beyond "press 1 for English."
The Economics That Matter
The financial case for Agentic AI is compelling:
- Cost per interaction: Agentic AI interactions cost approximately $0.50-$1.00, compared to $9.00 for human-assisted interactions
- Containment rates: Organizations typically target around 20% containment in the first phase, with potential to grow to 30% and beyond
- Handle time reduction: Automating caller identification and verification can save nearly a minute per call—a 10% efficiency gain on a 10-minute average call
- First Call Resolution: When systems finish tasks, FCR improves naturally
Critically, Barbara emphasized that this investment doesn't require increasing budgets or layoffs. Instead, it's funded through natural attrition (20-40% annually in most contact centers). Rather than replacing every departing agent, organizations redirect a portion of labor expense into automation that permanently reduces workload.
The Crawl-Walk-Run Approach
Hughes outlined a staged adoption path:
Crawl: Choose one narrow workflow and make it "boringly reliable." Define success clearly, keep guardrails tight, and make escalation clean and predictable.
Walk: Expand that workflow to handle real-world variation, simple exceptions, and continuity across channels.
Run: Connect adjacent workflows, introduce proactive operational messaging, and use oversight data to identify what to automate next.
The pattern that works best starts where risk is low and outcomes are clear, proves reliability, measures impact, and then expands deliberately.
Critical Success Factors
The speakers emphasized three foundational requirements:
- Data Quality: Poor data leads to poor decisions
- Integration: Systems must connect cleanly or work will stall
- Governance: Weak governance makes automation risky quickly
As Hughes noted, "Agentic AI doesn't remove foundational problems—in most cases, it exposes them."
What Success Actually Looks Like
In closing, Barbara shared that successful Agentic AI implementations often feel invisible to customers—the experience just works. The practical test is simple: when automation isn't working, customers tell you quickly by asking for a human, abandoning the interaction, or switching channels.
Success is defined not by whether AI sounds impressive, but by whether:
- Customers accept it
- Employees trust it
- Workflows perform as designed
- Operational signals improve (leaner resolution paths, lower avoidable effort, more predictable outcomes)
Three Key Takeaways
The session concluded with three essential points for contact center leaders:
Agentic AI is about action, not answers—the shift is from conversational automation that responds to outcome-driven automation that completes work
Value comes from applying automation to the right workflows—where speed, consistency, and completion improve both customer experience and the operating model
Readiness and governance matter as much as technology—data quality, integrations, guardrails, and oversight separate interesting pilots from production capabilities
Session Title: Agentic AI Explained: What It Really Means for the Contact Center
Speakers: Brent Barbara, Managing Director & Amanda Hughes, Founder and Managing Director, Clearest Blue LLC
Event: ICMI Digital Event 2026
Want to dive deeper? Watch the full session recording here to hear all the insights from Brent Barbara and Amanda Hughes.
Explore more from the 2026 ICMI Digital Event: Check out other recorded sessions here to continue your learning journey.
