AI agents are having a moment.

Every second vendor pitch we hear includes the word “agentic”, every vendor roadmap seems to have AI baked in (at least in theory). And every leadership team we speak to is asking some version of the same question: “Should we start using AI agents?”

Gartner’s answer is… sometimes. And for many organisations, not yet.

We would go one step further.

AI agents are not a strategy. They are a delivery mechanism that only makes sense when your system is ready for them.

Let’s unpack what Gartner actually says, and how to decide whether AI agents belong in your business right now. We’ll keep it grounded in the operational realities of everyday businesses and easily accessible for non-technical leaders.

First, let’s clear up the language 

Gartner opens their research with a blunt observation: the market is confused. And honestly, they are being polite about it.

“Agentic AI” is being used to describe everything from basic chatbots to highly autonomous multi-agent systems. Vendors are “agent-washing” (rebranding existing tools), while buyers assume anything with an LLM or a workflow is an AI agent.

Gartner draws clear distinctions:

• AI assistants: support with individual tasks and rely on human input (prompt -> response); 
• AI agents: act semi-autonomously toward goals, often across platforms, typically in pre-determined workflows; 
• Agentic AI: an architectural approach using one or more agents, able to exercise independent decision-making, to achieve an outcome.

Some may call this semantics, but we see it time and again: expectations lead to disappointment when reality falls short of the sales deck.

We also see a lot of confusion over these terms:

• Automation: the technology (software or hardware) used to perform repetitive steps automatically, often based on triggers/conditions; 
• Workflow: the sequence of tasks/steps/activities necessary to achieve an outcome. Can be manual, automated, AI or a mix of all.

Via’s translation:
If your so-called “AI agent” cannot perceive context, take action, adapt and operate with some independence, it is probably not an agent. 

The capability spectrum most leaders don’t ask about

One of the most useful contributions in the Gartner paper is the AI Agent Capability Spectrum. It shows that agency is not on or off. It is a gradient.

Agents vary across six capabilities:

• Perception
• Decisioning
• Actioning
• Agency
• Adaptability
• Knowledge

Most tools in the market today sit at basic or emerging levels, even when they are sold as autonomous.

This might lead buyers to design business-critical workflows that assume advanced agent autonomy, then deploy tools capable only of deterministic or situational responses. The result is brittle systems, human workarounds, and disillusionment with the technology.

At Via, we see this pattern constantly: the technology is blamed, while the real issue is a mismatch between system design and capability reality.

When AI agents are overkill (stick with good old programming) or downright too risky

Gartner explicitly warns against using AI agents when requirements are too low, and we couldn’t agree more. If your process is:

• Repetitive
• Rules-based
• Allows for low (or no) variance

Then AI agents add unnecessary complexity. Workflow automations will be cheaper, faster, and more reliable.

Too many teams reach for agents because they feel behind, not because the use case demands it. 

There is also the opposite problem: some use cases are simply too complex or too risky for today’s agents.

Red flags include:

• High-stakes decisions requiring judgment and empathy (we call it discernment)
• Regulated environments requiring explainability (think highly regulated industries e.g. healthcare)
• Scenarios demanding real-time data processing with zero tolerance for delays

Imagine yourself explaining to an auditor that “AI chose this”... not a conversation we’d want to have.

Add to this organisational readiness (or more accurately, lack of). Skills, governance, data quality and change management are often the real blockers, not the model.

Via perspective:
If your team does not trust your data, does not understand your processes, and does not have decision clarity, AI agents will not save you. They will just surface those cracks faster.

When AI agents actually make sense

Gartner is very clear: AI agents shine in a specific sweet spot.

They are valuable when:

• The environment is dynamic and context-rich
• Goals can shift or conflict
• Execution is variable and not fully predictable
• Traditional automation becomes too brittle
• The impact of occasional errors is manageable

Think coordination-heavy work, not precision-critical work.

AI agents do well when they are embedded into a well-understood system that already has:

• Clear decision rights
• Defined boundaries
• Known escalation paths
• Strong human oversight

This aligns perfectly with our Via principle: Don’t automate chaos.

Without a clear, solid system, AI simply amplifies the mess rather than removing it. Don’t expect AI to magically fix your operational chaos and lack of systems foundations; it won’t.

So what should leaders actually do?

Here is the Via lens, informed by Gartner and grounded in the operational reality of everyday businesses and non-technical leaders.

Before asking “Can we use AI agents?”, ask:

1. Do we understand the System this work belongs to?
2. Are decision rights clear?
3. How important is it to always get the same result?
4. What happens when the agent is wrong? What risks are we willing to tolerate?
5. Who owns the outcome (not the tool)?

AI agents are powerful; Gartner does not deny that. But used without intention, they are just adding to the noise our human brain has to manage and orchestrate.

The businesses that win with agentic AI in 2026 will not be the ones who move first. They will be the ones who design first.

If you are feeling the pressure to adopt agentic AI, take the time to step back and design Strategic Business Systems intentionally, and only then decide where (and if) AI fits into them.

If you want help doing that, you know where to find us.