How Agentic AI Is Reshaping the Enterprise in 2026

In this episode, Mike and Alex take a deep dive into agentic AI — the technology that's transforming how enterprises operate. They explore what agentic AI actually means, how it differs from traditional AI, and why 2026 is a turning point for adoption.

Topics covered include:

• What agentic AI is and how it works — autonomous AI agents that plan, decide, and act
• Real-world implementations at JPMorgan Chase, Wells Fargo, MNT-Halan, Siemens, and PepsiCo
• NVIDIA's NemoClaw announcement at GTC 2026 and the OpenClaw phenomenon
• Critical security risks: chained vulnerabilities, prompt injection, privilege escalation
• The gap between hype and reality — why 40% of agentic AI projects may fail
• Practical steps for organizations: data readiness, workforce investment, and governance

Transcript

SPEAKER_01 0:00

Welcome to the DX Today podcast, where you get facts and no hype. I'm Mike.

SPEAKER_00 0:05

And I'm Alex. Today we're doing a deep dive into one of the biggest shifts in enterprise technology right now: Agentic AI. If you've been hearing this term everywhere and wondering what it actually means for businesses and frankly for your career, this is the episode for you.

SPEAKER_01 0:21

That's exactly right. So let's start with the basics. Agentic AI refers to artificial intelligence systems that don't just generate text or answer questions. They can actually plan, make decisions, and take actions autonomously. Think of it this way: traditional AI is like having a really smart assistant who waits for you to tell them what to do. Agentic AI is more like having a colleague who understands your goals and goes out and accomplishes tasks on their own.

SPEAKER_00 0:46

And that's a huge distinction. MIT Sloan's professor Sinan Aral puts it well. He defines agentic AI as systems that incorporate multiple different agents orchestrating a task together. So it's not just one bot doing one thing, it's a network of AI agents that can coordinate across tools, databases, and workflows to complete complex multi-step processes. And importantly, these agents don't just operate in the digital world. They can monitor real-time video, interact with physical systems, and take actions that change things happening in the real world.

SPEAKER_01 1:21

Right, and the scale of adoption we're seeing is remarkable. Gartner predicts that by the end of this year, 2026, 40% of enterprise applications will embed task-specific AI agents. That's up from less than 5% just last year. And the market itself is projected to grow from about$7.8 billion to over$52 billion by 2030. So this is not experimental anymore. This is mainstream enterprise technology.

SPEAKER_00 1:46

But here's where it gets really interesting, Mike. While the numbers sound impressive, the reality on the ground is more nuanced. Deloitte's 2025 emerging technology trend study found that while 30% of organizations are exploring agentic options and 38% are piloting solutions, only 14% have systems ready for deployment, and just 11% are actually using them in production. So there's a massive gap between interest and execution.

SPEAKER_01 2:14

That's such an important point, and Gartner actually predicts that over 40% of agentic AI projects will fail by 2027 because legacy systems simply can't support modern AI execution demands. Accenture and WIPRO have found similar patterns, showing that 70 to 80% of agentic initiatives haven't made it to enterprise scale. So the challenge isn't the technology itself, it's the infrastructure, the data architecture, and frankly, the organizational readiness to support autonomous AI systems.

SPEAKER_00 2:44

Exactly. And Deloitte's research highlights something really crucial here. They found that many enterprises are making a fundamental mistake. They're trying to automate existing processes, processes that were designed by and for human workers, without actually reimagining how the work should be done. The organizations finding real success are the ones that redesign their operations to be agent compatible from the ground up.

SPEAKER_01 3:09

Let's talk about some real-world examples because I think that helps make this concrete. JP Morgan Chase is using agentic AI to automate legal and compliance processes. Their agents can plan workflows, detect issues, replan when something goes wrong, and deliver final outputs. They're reporting up to 20% efficiency gains in their compliance cycles. That's significant when you're talking about a bank of that size. And it goes beyond financial services. Siemens and PepsiCo unveiled their digital twin composer at CES 2026, where AI agents simulate and test supply chain changes with physics level accuracy before any physical modification is made. That's a game changer for manufacturing and logistics.

SPEAKER_00 3:52

Wells Fargo is another great example. Their virtual assistant called Fargo has completed over 242 million fully autonomous customer interactions. These are complex requests that previously required human agents. And Gardner projects that by 2029, AI agents will resolve 80% of common customer service issues without human intervention, leading to a 30% reduction in operational costs. So the trajectory here is clear.

SPEAKER_01 4:20

Now let's talk about one of the biggest developments this month because it ties directly into where AgenTic AI is headed. At NVIDIA's GTC conference on March 15th, Jensen Huan unveiled NemoClaw, a new enterprise agent platform built on top of the open source system OpenClaw.

SPEAKER_00 4:37

This is a really important announcement. OpenClaw has exploded in popularity since its launch about four months ago. It powers autonomous AI agents and has gained massive traction, including over 318,000 stars on GitHub in about 60 days. That's faster than Linux or React ever achieved. Now, Nvidia is essentially saying, okay, OpenClaw is powerful, but enterprises need security and governments around it. That's what Nemo Claw provides.

SPEAKER_01 5:04

Nemo Claw adds sandboxing, privacy controls, and policy-based guardrails so that AI agents that need full access to files and data can run more safely in corporate environments. Jensen Huang said, and I quote, every company in the world today needs to have an open claw strategy and an agentic system strategy. And he's reportedly been pitching this to companies like Salesforce, Cisco, Google, Adobe, and CrowdStrike.

SPEAKER_00 5:30

Now I want to flag something for our listeners here. There is some healthy skepticism around this. Some observers have pointed out that NVIDIA strategy follows a classic playbook. They highlight a genuine open source movement, underscore its security vulnerabilities, and then offer their proprietary solution, one that conveniently runs best on NVIDIA hardware. The privacy routing in NemoClaw directs sensitive tasks to local NemoTron models running on NVIDIA's DGX Spark or RTX systems. So while the security benefits are real, there is an element of hardware lock-in. Enterprises should evaluate this with clear eyes.

SPEAKER_01 6:12

Now let's talk about something that I think is critically important and often underappreciated: the security implications of agentic AI. McKinsey published a detailed playbook on this, and the risks they outline are genuinely concerning.

SPEAKER_00 6:26

This is where listeners really need to pay attention. McKinse identifies several new risk categories that traditional security controls weren't designed for. First, there's chained vulnerabilities. A flaw in one agent can cascade across tasks to other agents, amplifying the damage. Imagine a credit data processing agent that misclassifies debt as income. That error flows downstream to scoring and approval agents, leading to risky loan approvals.

SPEAKER_01 6:54

Then there's cross-agent task escalation. Malicious agents can exploit trust mechanisms to gain unauthorized privileges. McKinsey gives the example of a compromised scheduling agent in a healthcare system that requests patient records by falsely escalating the task as coming from a licensed physician. The clinical data agent releases sensitive health data without triggering any security alerts.

SPEAKER_00 7:17

And Obsidian Security adds more to this picture. They highlight prompt injection attacks, where attackers manipulate agent inputs to trigger unauthorized actions. There's identity spoofing and token compromise. There's model poisoning where training data is corrupted to subvert the agent's behavior. And privilege escalation through chaining, where agents integrate with multiple systems, each granting incremental permissions, and attackers chain these together to achieve access levels no single human would have.

SPEAKER_01 7:49

The fundamental issue is this. We're moving from systems that enable interactions to systems that drive transactions. These agents directly affect business processes and outcomes. The OASP Gen AI Security Project has been documenting these threats, and the security community is actively working on mitigation frameworks. But honestly, the tooling and best practices are still catching up to the pace of deployment. That makes total sense. You need clean, accessible, well-governed data as the foundation. And CloudKeeper's analysis adds that enterprises need mature orchestration frameworks, governance models, and observability platforms before deploying agents at scale. The infrastructure has to be ready before the agents can deliver value.

SPEAKER_00 8:42

The economic promise of agentic AI, as MIT Sloan describes it, is that agents can dramatically reduce transaction costs. The time and effort involved in searching, communicating, contracting, and monitoring. Tasks that humans have traditionally done can now be executed at machine speed and scale.

SPEAKER_01 9:04

But that doesn't mean people become irrelevant. Far from it. What Nuvento's research emphasizes is that agentic AI replaces manual execution, not accountability. Humans define the intent, the policies, the risk thresholds, and the escalation paths. AI handles the execution. Humans remain responsible for the decisions. The roles change, but human judgment becomes more important, not less.

SPEAKER_00 9:31

Exactly. The organizations seeing the best results are the ones that invest in upskilling their workforce to manage and oversee AI agents. Think of it like the transition from manual manufacturing to automated production lines. The factory workers didn't disappear. Their roles evolved into overseeing, maintaining, and optimizing the automated systems.

SPEAKER_01 9:52

So let's bring this together with some clear takeaways for our listeners. First, agentic AI is real and it's moving fast. With 40% of enterprise apps expected to embed AI agents this year, this isn't a future trend. It's happening now. Second, don't just automate existing processes. Redesign your operations for agent compatibility. That's where the real value comes from. Third, take security seriously from day one. The risks of prompt injection, privilege escalation, and chain vulnerabilities are real and growing. Build governance and security frameworks before you deploy, not after. Fourth, data readiness is your foundation. If your data isn't accessible, well organized, and properly governed, fix that first. And fifth, invest in your people. The technology is powerful, but it needs human oversight, human judgment, and human accountability. The most successful agentic AI deployments are the ones where humans and AI agents work as a team, each doing what they do best. Well said, Mike, this is going to be one of the defining technology shifts of the next several years, and understanding both the opportunities and the challenges is essential for anyone working in or around technology today. Thank you for joining us today for the DX Today podcast. Stay curious.