I keep trying to polish my education on all things AI in ServiceNow. There’s no shortage of material on Now Learning — in fact, there are a LOT of short 10 mn courses dedicated to AI.

One that caught my attention recently was “Real World Use of AI Agents.” It’s a quick course, but it triggered some bigger reflections for me/

My First Impressions

When I first heard “AI agent,” I pictured a digital persona that could work on my behalf. A kind of clone that would know how I operate and make decisions the way I do.

In my head, it went even further: an “Innie/Outie” setup, like in Severance (minus the drama). My “agent” would grind through the boring tasks while I focused on meaningful things.

And then some questions came: if I left my company, would they keep my digital persona? Would they snapshot my skills and let me go once the clone was ready? Why keep the employee if the agent is just as good?

BACK TO EARTH — we’re nowhere near that.

Where We Actually Are

In the course, ServiceNow is transparent about it: agentic AI is not that mature yet. The training material shows potential more than reality.

The truth today is that AI agents are more like smart orchestrators. They don’t invent new capabilities. They bundle what’s already in the platform — GenAI, flows, integrations, notifications, classification — and add a layer that can take instructions in plain language and keep going even when things don’t go as expected.

That resilience is new. But the building blocks are not.

Use Cases (and the Reality Check)

Here are some of the examples presented:

• ITSM: diagnose a server issue, suggest a fix, ask approval, execute, notify. This is the zero-touch IT story — but we’ve had flows and automations doing most of that before.
• HR: approve time-off, guide onboarding, generate knowledge articles. Again, that’s Virtual Agent, onboarding workflows, and GenAI in action.
• CSM: confirm refund eligibility, send notifications, escalate. Same story — chatbots and flows already handle it.
• Security: autogenerate closure notes, post-incident summaries, answer analyst questions. That’s GenAI summarization, not a new agent-only trick.
• Dev & Finance: syntax checks, claim summaries. Useful, yes — but not an Agentic AI revolution.

The pattern is clear: many “agentic AI” examples are existing capabilities that indeed can be wrapped in an agent. The difference is in how they’re packaged and orchestrated.

What’s Actually Different

So, does it actually work?

• You give it a goal in plain English.
• You outline steps and limits in plain English (“don’t notify the user more than once”).
• you define specialized agent and you give each of them a skill (flows, APIs, spokes, notifications).
• When ran, It will keep trying to achieve the goal, adapt when blocked, and report progress.

That’s where the “agent” becomes more than a static flow. Where a flow breaks at unexpected data, an agent can try another approach or ask clarifying questions.

The Risks and Open Questions

As human we are not always good at explaining ourselves, and leave a lot unsaid because it seems obvious. This is where my doubts kick in

• If a use case is too complex for us to model as a flow, how do we trust the agent to do better?
• Can we make sure the agents don’t spam an end user with endless questions?
• Can we prevent inefficient retries — like rebooting a server ten times in a row?

Like with a person, sometimes, checking that the agent did the work correctly takes longer than doing it yourself. So the value is clearer in mass, repetitive tasks where validation is simple.

The Bigger Picture

The industry loves to put every capability under an “AI” label. That’s not a ServiceNow thing specifically — it’s a market-wide reflex. And while it creates buzz, it can also blur the line between what’s genuinely new and what’s a repackaged feature.

For me, the right way to see agentic AI is this:

• It’s not a digital clone.
• It’s an orchestrator that handles fuzzier plain text instructions, navigates uncertainty, and ties together capabilities we already know.

That’s useful — but let’s not oversell it.

Closing Thought

The dream of digital coworkers is powerful. But today, AI agents are closer to persistent assistants that don’t get stuck when flows may. They can be valuable, especially for scale, but they’re not yet the “living” personas I once imagined.

The hype makes them sound like more than they are — and that’s fine, as long as we stay clear-eyed about what’s actually happening under the hood.