Agentic AI: Cutting Through the Hype to Find the Reality

Here's what I keep hearing: "2025 is the year of the agent!" "Agentic AI will replace knowledge workers!" "It's fully autonomous and delivers immediate ROI!"

And here's what I'm actually seeing: A lot of hype. Some reality. And a massive gap between the two.

I spent my latest livestream diving into what's really happening with agentic AI—not the vendor promises, but the actual state of the technology and what business leaders need to know right now. Because if your IT department is coming to you saying "let's build an agent," you need to understand what that really means.

The Hype vs. Reality Check

The market numbers look impressive. We're talking about growth from $6 billion to $107 billion. And the promises are even bigger: Fully autonomous AI. Seamless development. Universal applicability. Instant ROI.

Sound familiar? It's just another regular day in the tech world.

Here's the reality: This has been a year of tremendous development with agents, but we're looking at the next decade of agents, not the next quarter. Some organizations are achieving ROI within the first year, but they're doing it with very specific, structured workflows and clear boundaries. Think claims processing, not strategic planning.

The agents that work today excel at high-volume, fairly routine tasks that require a bit of reasoning. They shine when there's clear success criteria and well-defined rules. What they can't do yet is open-ended strategic thinking or handle level four and five autonomy.

And that's the disconnect. People expect self-improving, fully autonomous agents when the technology is mostly at level two and three autonomy. 

We want the bright shiny object, but it doesn't exist yet.

What's Actually Working Right Now

Let me give you some real examples of where agents are delivering value:

📄 Insurance 
Agents are processing straightforward claims, recognizing when something's too complex, and handing it to a human. They're tackling a major administrative burden that's been bogging down claims adjusters.

🏥 Healthcare
Cleveland Clinic and Mayo Clinic are using AI tools to analyze entire medical histories. The AI categorizes symptoms, identifies patterns, and gives doctors a preliminary analysis so they can focus on patient care.

🏦 Banking
For loan approvals and fraud rejection, agents can handle the clear-cut cases and escalate the murky ones. They're not replacing loan officers, they're making them more effective.

➡️ The pattern is that agents work when you have volume, clear rules-based processes, and the ability to escalate exceptions to humans. The key difference from old-school robotic process automation is that these agents can reason through gray areas using logic and criteria, not just follow strict "if this, then that" rules.

Why Most Agents Fail

I talk to a lot of companies who say agents aren't working for them. Here's what I'm seeing:

▶️ Data quality issues. Your agents are only as good as your data. But you don’t need to wait for your data to be perfect. My “Winning with AI” book co-author, Katia Walsh, is adamant about this, given her experience as a Chief Data and AI Officer. Start with what you have. If your data isn't ready for your big strategic initiative, find a smaller project where it is ready and start there. 

▶️ Governance gaps. Do you have rules of conduct for your agents? Who has permission to access what data? What decisions can agents help with and what requires human oversight? You need governance from day one.

▶️ Legacy infrastructure. Agents don't understand silos. When they need data from three different systems to complete a task, they expect to access it. If your infrastructure isn't API-driven and integrated, you've got problems.

▶️ Pilot purgatory. Too many organizations are running pilots that never scale because no one's prioritizing which use cases actually matter. This isn't an IT decision, it’s a CEO mandate. AI isn't a strategy; it's a technology that enables your business strategy. 

▶️ Poor use cases. If you're trying to use agents for open-ended strategic thinking, you're setting yourself up for failure. Go for proven use cases with high volume, routine work that requires some reasoning.

What You Should Do Now

In our book, “Winning with AI”, Katia Walsh and I go through how to get started creating value with AI. Those principles apply even more so when it comes to deploying agentic AI. 

First, start with strategy. Your CEO, not your CIO, should be driving agentic AI strategy. Connect it to one of your top three strategic objectives. Set growth and innovation goals, not just efficiency targets. If you're only chasing productivity, you're missing the point.

Second, don't just automate … transform. Pick one process to completely redesign, not just automate. How would you do this differently if agentic AI was handling it instead of humans?

Third, build the foundation. You need good data, strong APIs that connect your platforms, clear governance, and a plan for managing agents at scale. 

And here's the most important piece of advice: Use commercial platforms first. Don't start by building agents from scratch. There are enterprise-grade platforms from Salesforce, IBM, Microsoft, Google, SAP, and more, all with libraries of ready-to-deploy agents. There are also builder platforms that sit on top of your existing systems. Tools like n8n, Zapier, Crew AI, and LangChain make building agents dramatically easier.

Deploy an agent. See if it works. If it doesn't, you'll learn something. Once you understand what you need and how to deploy agents effectively, then you can consider custom development.

The Timeline Everyone Gets Wrong

Here's what people need to understand: We're early. 

The technology is developing fast, but your organization also needs to get ready. That readiness requires strategic planning, integration, and fundamentally reimagining how your organization works. These two things—technology advancement and organizational readiness—need to happen in parallel.

Think of this as building organizational muscle. Each agent you deploy teaches you something about how to deploy the next one in a better way. You learn about your data, your processes, your governance, and your infrastructure. That learning compounds.

🗣️ Your Turn

What has your experience been with agents? Are you building AI agents in your organization? What's worked? What hasn't? 

I'm heading off the grid to Antarctica for a few weeks (yes, really!), but I want to hear from you when I'm back. Your experiences, the good, the bad, and the lessons learned. That's how we all get better at this.

I'll be back December 2nd with a deep dive into something fascinating that came up in our research for the book: The rise of superhumans who integrate the best of AI and the best of human intelligence. It's not about efficiency. It's about creating value with a capital V—the kind that helps us do what we uniquely can do as humans.

What I Can’t Stop Talking About

• We might be thinking about time all wrong. Timelines spanning decades don’t cut it anymore. Here’s what happens when you scale back and look ahead 18 months at a time. 

• It’s time to debunk myths about AI, and in our forthcoming book, “Winning With AI,” my co-author Katia Walsh and I are breaking down some of the biggest mysteries surrounding AI. 

• Modular building is the future, and the long-term “cathedral” building approach is out. It’s not just a trend; it’s the only way to keep up with AI’s constant evolution.

 My Upcoming Appearances/Travel

• Dec 3: AWAAI webinar (Asian Women Advancing AI), virtual

• Jan 26: Private executive leadership retreat, Houston, TX

• Feb 3: Ragan All-Horizons Conference, Keynote, Ft. Lauderdale, FL

• Feb 27-28: OrthoForum 2026, Keynote. Tampa, FL

If you found this note helpful, please forward and share it with someone who needs the inspiration today. If you were forwarded this, please consider subscribing.

Join me on Facebook, Instagram, LinkedIn, and Twitter.