Let’s be honest, the first time you saw a generative AI model like ChatGPT create a perfect email or flawless block of code, it felt like a giant leap forward. But after that initial “wow” moment, a bigger question probably came to mind: “What if it could actually do the things it’s writing about?”
What if it didn’t just draft the project plan but created Jira tickets, assigned them to engineers, and sent Slack notifications? What if it didn’t just analyze sales data and suggest pricing changes but updated pricing in your ERP, monitored results, and adjusted automatically?
That’s the reality we’re seeing right now, and it’s centered around agentic AI. It’s the next logical, and frankly, quantum, evolution from the AI we’ve gotten used to over the past couple of years.
This is the very topic we explored during our recent KMS panel, “The Human vs. The Robot.” We brought together two industry titans, Tolga Tarhan, a cloud and AI thought leader who has built and scaled multiple major tech companies, and Dilip Dubey, a serial AI entrepreneur and investor who’s created over half a billion dollars in value in the space. Moderated by our own Chief Delivery Officer, Jeff Scott, the discussion was geared to cut straight through the noise.
The Big Shift: When AI Gets a To-Do List
First, let’s clear the air on what we’re even talking about. For the last couple of years, the world has been captivated by Generative AI. You give it a prompt, and it generates a response—text, images, code, you name it. It’s a phenomenal tool for analysis, creation, and summarization. It’s an expert consultant at your fingertips.
Agentic AI is different. It doesn’t stop at the suggestion. An agentic AI system takes a goal, breaks it down into tasks, makes decisions, and then uses tools to execute those tasks in the real world.
Think of it this way:
- Generative AI is the analyst who researches a problem and writes you an exhaustive, insightful report.
- An agentic AI system is the COO who reads that report, formulates a multi-step plan, coordinates with different departments via their software, allocates resources, and reports back on the outcome.
It’s the transition from a passive “thinker” to an active “doer.” And as our panelists made clear, this shift is forcing a complete re-evaluation of how businesses operate. As Tolga Tarhan put it, the most effective AI agents aren’t “bolted-on” afterthoughts, like a clunky chatbot you have to go out of your way to use. The real value comes when the AI is “wired in” to the core workflows of the business, feeling like a natural and essential part of the process.
This isn’t about building a better chatbot. It’s about building a better, more autonomous organization. The challenge is, where on earth do you start?
Finding a Foothold: Start with Pain, Not Hype
The pressure on CTOs and CPOs to “do something with AI” is immense. But chasing the hype is a recipe for expensive science projects with no clear ROI. Tolga offered a refreshingly pragmatic approach for identifying the right place to begin your agentic AI journey.
“I don’t find those to be the right questions to ask,” Tolga said, referring to the open-ended ‘what can AI do?’. “I think it’s: ‘What’s slowing you down today? What today is costing too much money? What’s affecting quality?’ We should be looking inward at the business and finding a pain point, and then asking, “Can AI fix that?”
His framework is a masterclass in strategic focus:
- Look for Repetitive Decision Cycles: Don’t try to automate a once-a-year strategic plan. Look at high-volume, rules-based tasks that humans perform over and over. Think support ticket triage, customer onboarding sequences, or dynamic pricing adjustments. These are processes with clear logic, making them perfect candidates for early AI agents.
- Start Small to Build Confidence: Forget the two-year moonshot project. Find a win you can get in 30 days. This isn’t just about technical feasibility; it’s about organizational psychology. A quick, visible win builds trust, quiets the skeptics, and secures the budget and resources for the next, more ambitious project. As Tolga noted, a demonstrable success will “get you budget support, resources, and sort of… overall buying of the organization.”
- Avoid the Big Data Quagmire (at first): It’s tempting to point your shiny new AI at your biggest, messiest data problem. Resist the urge. Those projects can take years to show value while you wrestle with data cleansing and integration. Pick a use case where the data is already in decent shape.
Dilip Dubey built on this, introducing a “portfolio approach.” He advised leaders to map out all possible use cases on a two-by-two matrix: one axis for business value, the other for your confidence in actually pulling it off. This simple exercise forces a strategic conversation that moves beyond the buzz and focuses on execution. “Early wins lead to a great strategy,” Dilip emphasized. “A great strategy which does not have wins behind it creates this believability problem.”
The Trust Equation: Giving AI the Keys Without Crashing the Car
So, you’ve identified a use case. Now comes the primal, C-level fear: how do you trust an AI system to act on its own without putting at risk your reputation, your data, or your systems? This isn’t just a technical problem; it’s a governance nightmare waiting to happen if not approached correctly.
Both panelists were adamant that blind faith isn’t the answer. It’s about building a robust framework of control and oversight from day one.
Dilip was clear: success happens “when you have a very clear human governance in place for those agents.” For complex enterprise operations, the concept of an entirely autonomous, self-governing AI is not yet a reality. Initially, you need humans in the driver’s seat.
Tolga provided a practical way to structure this, calling it a “human-in-the-loop” design. In this model, the AI agent does the work but proposes the final action, and a human gives the final “go” or “no-go.” For example, an AI might analyze customer data and draft an onboarding email, but a human has to click “send.”
“When you see that the approvals have become essentially just automatic, and you know, it’s always right,” Tolga explained, “that’s maybe when you think about taking the human out of the equation.”
This approach does two brilliant things simultaneously:
- It de-risks the process entirely. The AI can’t go rogue.
- It builds trust within the team. They see the AI working, learn its quirks, and become comfortable with its decisions, making them advocates for more autonomy down the line.
The other critical piece is building guardrails into the tools the AI uses, not just the AI model itself. Tolga used a powerful medical analogy: if an AI controls a device that administers fluid to a patient, you don’t just train the AI to be safe. You build the device so it is physically incapable of administering a dangerous dose, no matter what the AI tells it to do. “That reduction of risk is not done by the AI,” he stressed, “it’s done by the tool that you gave it. That is regular code, procedural, auditable, testable, deterministic code.”
This is a profound insight for any tech leader. The safety of your AI systems doesn’t just live in the probabilistic world of the model; it’s anchored in the deterministic world of your own coded interfaces and APIs.
The People Problem: This is a Culture Shift, Not a Software Update
You can have the best strategy and the safest guardrails, but if you don’t bring your people along, your agentic AI initiative is doomed. Tolga and Dilip both spent significant time on the human element, acknowledging the very real fear and organizational resistance that comes with this level of automation.
“Most of the fatigue, resistance, etc., around this particular change comes from a personal, ‘is my job at risk?’ kind of feeling,” Tolga observed.
The key to overcoming this is in the framing. This isn’t about replacing people. It’s about augmenting them.
- Frame AI as a Copilot: The most successful messaging reinforces that AI isn’t here to take your job; it’s here to take your worst work—the tedious, the repetitive, the low-value tasks that lead to burnout. “An AI-powered human is going to be replacing the job of a non-AI-powered human,” Tolga clarified. It’s a powerful distinction.
- Start with Empowerment: Find the tasks your team already complains about. The delays, the duplicative effort, the mind-numbing data entry. When you apply an AI agent to that pain point, you aren’t just seen as driving efficiency; you’re seen as making their lives better.
- Co-Create the Solution: Don’t build an AI agent for a team. Build it with them. Involve them in the design process. They know the nuances and edge cases you don’t. This not only builds a better tool but also transforms the team from potential resistors into proud owners.
Dilip shared a fantastic real-world story of this in action. A company used an AI agent to automate a scheduling process that previously took one person four weeks of grueling work. That time shrank to just four hours of governance per week. But the person wasn’t fired. She was upskilled. She now manages the AI, handles the exceptions, and focuses on higher-level problems.
Who Do You Trust? Navigating the AI Vendor Jungle
Finally, the million-dollar question- who do you trust when the market is flooded with vendors all claiming to have the ultimate agentic AI solution? It’s a classic case of “shiny object syndrome,” and it’s incredibly difficult to know who to trust.
Here, our panelists offered some tough love and crucial advice for any CTO or CIO.
Tolga warned against going all-in on niche, all-in-one platforms that promise to solve everything. “You get potentially embedded in an ecosystem that’s very niche,” he said, “and so what happens is you end up with 30 of those things.”
His recommendation? Own the integration layer. Your core AI strategy, your data, and your primary execution engine should be built on an open, neutral architecture that you control. Then, you can experiment with specialized SaaS products and AI agents as “leaves on your tree.” If one vendor goes away or a better tool comes along, you can simply swap it out without ripping up your core infrastructure. It requires more thought upfront than just dumping your data into a vendor’s system, but it pays massive dividends in flexibility.
Dilip added another critical layer to this: own your IP. He cautioned, “All the vendors are learning and building their IP on your data.” For non-core functions, that might be fine. But for the processes that define your competitive advantage, that learning and the resulting intellectual property must remain yours. “This data and the knowledge hidden in the data for your core systems is your future asset,” he stated. “This asset may be more valuable than your whole company today.” That’s a statement that should make every executive pause and reread their vendor contracts.
The Road Ahead
If there was one unifying message from the panel, it was this: agentic AI is not a far-off concept from a sci-fi movie. It is the practical, and tangible next step in enterprise automation. The journey from piloting your first AI agents to running a truly autonomous organization is a marathon, not a sprint. It will demand a thoughtful strategy, robust governance, a people-first approach to change, and a discerning eye for partners.
The era of just talking to our computers is ending. The era of giving them a to-do list has begun. The real question is, what’s the first thing you’ll ask your AI agent to do?
At KMS Technology, we specialize in helping organizations navigate these complex technological shifts. If you’re ready to move from conversation to action, we’re offering a complimentary 2-hour ideation sprint to help you identify high-value use cases for agentic AI in your business. Let’s build the future together.
TAGS