[0:00] Welcome back to EtherLink AI Insights. I'm Alex, and today we're diving into something that's genuinely reshaping how enterprises think about AI. We're talking about AI agents and agentech AI systems. And spoiler alert, this isn't just the next evolution of chatbots. It's a fundamentally different approach to enterprise automation. Sam, thanks for joining me. When you first heard the term agentech AI systems, what was your initial take? [0:31] Great to be here, Alex. Honestly, I think there's a lot of confusion in the market right now about what agentech actually means. Everyone's throwing the term around, but most organizations are still deploying what are essentially fancy chatbots and calling them agentech systems. The real distinction is autonomy, genuine bounded autonomy. An agentech system doesn't just respond to a user prompt. It reasons through multi-step workflows, makes decisions, [1:01] and adapts based on real-time feedback. That's a completely different beast. That's a crucial distinction. So when you say bounded autonomy, you mean these systems aren't just doing whatever they want. There are guardrails in place, right? Exactly. An agentech system operates within defined parameters and compliance frameworks. It can autonomously decide to escalate a patient case to a human clinician or re-root a workflow based on business logic, but it's always operating within the rules the organization has set. [1:33] That autonomy within guardrails is what makes these systems genuinely valuable at enterprise scale. The numbers here are staggering. Gartner found that 73% of enterprises are actively piloting agentech AI systems as of 2026, compared to just 31% in 2024. That's more than a doubling in two years. Why the acceleration? Because the pain point is real. Traditional single-purpose chatbots hit a wall. You've got one agent handling customer service, [2:04] another managing scheduling, another validating insurance, but they're all silos. There's no coordination, no context sharing. A healthcare provider dealing with patient intake, appointment scheduling, and clinical documentation realized they needed something that could orchestrate across all those domains simultaneously. That's where agentech orchestration comes in. Orchestration is the key word here. Let's talk about that. What does orchestration actually look like in practice? [2:34] I'm imagining multiple specialized AI agents working together, but how does that actually function? Think of it like a jazz ensemble rather than a solo performance. You have specialists, a patient intake agent, an insurance verification agent, a scheduling agent, a documentation agent. They're each excellent at their specific domain, but they need a conductor, right? That's the control plane. The control plane manages which agents are deployed when ensures their sharing context appropriately, [3:06] handles handoffs between agents, and critically enforces compliance policies across the entire ecosystem. So the control plane is like the orchestration engine that keeps everything coordinated. What does that actually involve from a technical and operational perspective? It's handling several things simultaneously. You need agent lifecycle management, spinning up agents based on demand, scaling them, deprovisioning them when they're no longer needed. You need compliance enforcement, which is non-negotiable, [3:36] especially with the EU AI Act now in effect. You need cross agent communication. So when an intake agent collects patient information, that data flows seamlessly to the scheduling agent without manual intervention. And you need real-time performance monitoring, latency, error rates, cost per transaction, success metrics. Speaking of compliance, that's a huge topic right now. The EU AI Act is in effect, HIPAA requirements, GDPR. How does an orchestration system actually enforce compliance [4:09] across multiple agents? This is where control planes prove their value. Compliance isn't something you bolt on afterward. It's embedded in the control plane logic. Every action an agent takes gets evaluated against compliance policies in real time. For health care, that might mean a clinical documentation agent can't share patient notes with a billing agent without proper authorization checks. The control plane enforces those boundaries automatically. Organizations doing this properly report a 42% reduction in compliance violations. [4:41] That's significant given the regulatory penalties we're seeing. That's a concrete number. Are there other measurable outcomes we should be thinking about when evaluating whether to invest in a gentick orchestration? Absolutely. The 58% improvement in multi-agent workflow completion rates is telling. That's about end-to-end process efficiency. But you also need to look at ROI metrics that are specific to agentic systems. What's the cost per transaction? What percentage of workflows can be fully automated [5:13] versus requiring human intervention? What's the latency improvement compared to sequential manual processes? And critically, what's the impact on employee productivity and customer satisfaction? Let's talk about what we're seeing in different verticals. You mentioned health care, which seems like an obvious use case. Where else are we seeing strong adoption? Financial services is huge. Loan processing, compliance verification, fraud detection. These are complex, multi-step workflows [5:43] that are perfect for agentic orchestration. Insurance companies are deploying agents to handle claims processing with different agents validating claims, assessing coverage, calculating payouts, and flagging anomalies. Manufacturing is using agent orchestration for supply chain management and quality control. The common thread is that these are all processes with multiple interdependent steps that benefit from coordinated autonomous decision-making. So it's not just about replacing humans. [6:14] It's about making complex processes work more efficiently. When someone is evaluating whether to invest in this, what should they be looking for in a platform or implementation partner? First, demand visibility into their control plane architecture. How do they handle compliance? How do they implement agent communication? Are they actually orchestrating multiple specialized agents? Or are they just marketing a chatbot as agentic? Second, ask about their monitoring and observability capabilities. [6:46] You need to see what agents are doing, why they're making decisions, and how they're impacting your business metrics. Third, insist on clear escalation paths. When an agent encounters something outside its expertise, how does it hand off to a human? That's crucial for safety and compliance. Those are really practical criteria. One thing I'm curious about, as these systems become more sophisticated and more autonomous, what's the risk profile we should be thinking about? [7:17] The key risk isn't the autonomy itself, it's opacity. If you deploy an agentic system without proper monitoring and you don't understand why agents are making decisions, you can end up with compounding errors or compliance violations that you don't catch until it's too late. That's why control planes need to be transparent. You need to be able to audit agent decisions, understand the reasoning chain, and intervene when necessary. The safest agentic systems are the ones where humans [7:48] maintain meaningful oversight without being bottlenecks. That's a really important balance, so you're not replacing human judgment. You're augmenting it with autonomous systems that can handle complexity at scale. If someone's just starting their agentic AI journey, what should be their first priority? Pick a process that's clearly multi-step involves coordination between departments and has strong ROI potential if you can automate it. Insurance verification and healthcare is a great example. [8:18] It requires pulling data from multiple systems, making decisions based on policies and escalating exceptions. Start with a pilot, measure the metrics carefully, and ensure you've got proper compliance safeguards from day one. Don't try to build a fully autonomous system immediately, prove the value with semi-autonomous workflows where humans remain in key decision loops. That's excellent advice. Start focused, measure rigorously, iterate based on results. [8:49] Sam, thanks for breaking down something that's genuinely transforming enterprise operations. For our listeners who want to go deeper into control plane architecture, multimodal AI agents, compliance frameworks, and real world implementation strategies, the full guide is available on etherlink.ai. Check it out for more detailed insights on building your enterprise-agentic AI strategy for 2026. Thanks for listening to etherlink AI Insights. [9:19] Thanks for having me, Alex. It's an exciting moment for enterprise AI and getting these implementation details right is going to be the difference between organizations that successfully scale AI and those that hit compliance or operational walls.