Imagine your customer opens a support chat to resolve a payment glitch—and instead of a quick fix, the bot loops generic responses, escalating frustration. Now imagine the opposite: an intelligent digital agent that not only solves the issue in minutes but also anticipates the next step, offering proactive guidance and empathy that feels human. It’s just a difference having our not having Agentic AI in CX.
In reality, Agentic AI in CX is not a distant vision. In fact, it’s the reality whose architect is Richard Smullen, Founder and CEO of Pypestream. Under his leadership, Pypestream has evolved into a global leader in conversational AI, powering autonomous, secure, and measurable customer experiences for enterprise giants like Google, AT&T, and HBO. With over 50 million monthly interactions processed across industries, Pypestream’s Agent Platform brings a balance of intelligence, compliance, and creativity to modern CX design.
A member of YPO and the founder of SouthWinston, Smullen’s vision extends beyond technology. He’s a builder of future-ready enterprises—championing AI that enhances both customer and employee experiences. In this exclusive CXQuest.com interview, he shares how agentic AI redefines automation while preserving the essential emotional thread of human connection.
Agentic AI in CX: Most Meaningful Recent CX Win
Q1. What recent CX win surprised you most at Pypestream?
RS: Having our conversational AI agent get the following unsolicited comment was not the most surprising, but it was the most meaningful recent CX win for our team recently. “I’m [elderly] and not exactly text savvy. The agent stayed by my side and directed me… So we’re all set. I don’t know his name or her name but whoever it was did a wonderful job.”
This valued customer of our client had a long exchange with the AI Agent that responded empathetically and affirmatively, offering patience and repetition as needed -likely exceeding what a multi-tasking human would have provided. This is a perfect example of the types of interactions our industry should strive to achieve and are the wins we should be celebrating most. Not because “we fooled ‘em,” that isn’t the point, but because we helped in a way that made the customer feel validated, supported and ultimately delighted.
Q2. What core difference do you see between agentic AI and traditional chatbot systems?
RS: Chatbots hold a conversation. AI Agents take interactions to completion. One answers a pre-defined question with pre-defined answers. The other reasons, decides and executes within defined boundaries. It is our position that clients should be demanding solutions that encompass both: solutions that can distinguish between when to follow a strict workflow, never deviating from a defined process, and when to allow well-trained AI models to operate more flexibly and offer the best reasoned response based on interpretation of inputs.
Agentic AI in CX Platform Maintaining Enterprise-grade Security
Q3. How does Pypestream’s Agent Platform interpret “autonomy” while maintaining enterprise-grade security?
RS: Autonomy is scoped, not unchecked. Agents are empowered to act only within clearly defined rules, permissions and auditability. AI agents should intelligently route users and answer questions based on data and context, but to maintain a strong security posture when it comes to high-value actions such as processing payments or completing refund transactions, the AI agents should handoff to a locked-down system like a secure workflow or API. This prevents the AI itself from becoming a direct access point for malicious actors.
Q4. What frameworks or ethical guidelines guide your AI orchestration across industries like healthcare and finance?
RS: Start with a simple principle: if a decision requires judgment, accountability or regulatory interpretation, the human stays in the loop. AI accelerates, while humans authorize.
Successfully rolling out secure AI capabilities requires thinking about agents as only one component of the solution, relying on other common sense strategies and locked-down systems to complete sensitive tasks. Humans do not get unfettered access to every system within a company on their first day and neither should AI agents. Organizations should adopt a security model similar to that used for human agents. They should apply minimal required access with guidelines on what information AI agents can and cannot access and give out.
Agentic AI in CX: AI Agents Preserving Brand Voice
Q5. How do you ensure AI agents preserve brand voice without sounding mechanical or generic?
RS: Brand voice is treated as a system asset, not a prompt. We design conversational intent, tone and escalation paths with the same rigor brands apply to human training.
Q6. What strategies help large enterprises integrate agentic AI into legacy ecosystems without disruption?
RS: Large enterprises integrate applied AI solutions most successfully by treating agentic AI as an orchestration layer. Instead of ripping out legacy platforms, the AI sits across them – connecting CRMs, ERPs, and on-prem systems through secure APIs to automate workflows and unify data. This modular approach avoids disruption and lets AI capabilities evolve independently of legacy release cycles. Built-in analytics, monitoring and auditability allow teams to understand how AI is interacting in real-time with legacy systems and where improvements can be made. By shifting routine, rules-based tasks into AI-enabled automation, enterprises can gradually decommission brittle legacy workflows, reducing the maintenance burden that consumes the majority of IT spend.
Q7. How do you measure ROI on large-scale AI-driven CX deployments?
RS: We encourage our clients to look beyond KPIs like containment, cost-to-serve, speed-to-resolution and employee productivity and focus instead on the most critical business outcomes: cost savings, revenue growth, operational efficiency and customer loyalty.
Reduced Escalations Paired with Higher Agent Satisfaction
Q8. What metrics best prove that automation enhances—not replaces—human service?
RS: Reduced escalations paired with higher agent satisfaction as well as customer satisfaction is the formula we look to most. When experienced humans handle fewer but more meaningful interactions, everyone wins.
Q9. How do you balance innovation velocity with enterprise risk tolerance when deploying new CX models?
RS: We recommend balancing innovation velocity with enterprise risk tolerance by piloting applied AI incrementally. Every new capability earns trust through performance before it earns scale.
A few specific strategies organizations can adopt as they roll out Applied AI capabilities include:
* Establish an AI Governance Risk Committee to oversee deployment of AI models and monitor emerging threats.
* Run scenarios that combine technical, procedural and human dependencies. Assume that breaches will happen and ensure systems are designed to recover quickly and limit damage.
* Hold ongoing employee training on phishing, malware and AI misuse.
* Adopt data governance standards. Ensure that sources of data used for training models are verified and high quality. Incorporate validation pipelines before training models to detect anomalies, and to reduce sensitivity to malicious inputs. Conduct adversarial testing so that the models are hardened against poisoning.
Winning Internal Alignment
Q10. What leadership insight helped you win internal alignment around automation-led transformation?
RS: We recommend that our clients position AI-enabled automation as a competitive differentiator and workforce multiplier, not a cost-cutting mandate. While cost savings, revenue growth, operational efficiency and customer loyalty should be expected from applied AI solutions, stakeholders should see this as the opportunity to replicate the outcomes of your high-performers while freeing them to perform more high-value, rewarding tasks.
Q11. How do EX (employee experience) and CX reinforce each other in AI-enabled customer journeys?
RS: When applied AI removes repetitive tasks and low-reward work, employees show up more engaged and customers feel that difference immediately.
Q12. How do you envision agentic AI evolving customer trust over the next three years?
RS: Consumer behavior is already evolving to prefer interacting with intelligent agents over the traditional user experience of navigating website pages. Every day we are seeing how AI is becoming the new UI, increasingly chosen over apps, websites and call centers. In three years, expectations around trust will have shifted from healthy suspicion to reactionary. Customers won’t care that AI is involved, they’ll care that things get done correctly, consistently and proactively. Only when something goes drastically wrong will situational distrust be activated.
Reconciling CX-Cost Conflicts
Q13. How should enterprises reconcile CX-cost conflicts in AI-orchestrated service models?
RS: Enterprises need to look beyond a single metric and recognize that cost efficiency and experience quality need not have an inverse relationship. The best customer experiences eliminate wasteful repetition, not care.
Q14. Finally, what excites you most about the next frontier of conversational technology and cognitive design?
RS: What excites me most about what’s coming with next-wave adoption of applied AI in customer experience is the reliability of Voice AI. Conversational AI agents have always been designed to help enterprises contain interactions within digital-first chat channels. But for more traditional organizations, voice remains a necessary touchpoint. Voice AI capabilities meet people where they are today while giving organizations a seamless path toward a digital engagement model over time.
Reactive Automation to Agentic Orchestration
Finally, as the conversation with Richard Smullen reveals, the evolution of customer experience is shifting from reactive automation to agentic orchestration—where AI acts with intent, empathy, and measurable accountability. In fact, Pypestream’s story isn’t just about speed or cost. It’s about building trust at scale, uniting data, dialogue, and design into one secure framework that empowers both consumers and businesses.
Actually, for CX leaders seeking to explore next-gen AI frameworks, ethical deployment models, and success stories at scale, discover more at our dedicated CX Trends and CX Benchmarks.
