Why agentic AI’s next challenge is making systems work together

Ask any IT team how they usually spend their week and you’ll likely get the same answer – firefighting.  

It’s not the urgent, high-stakes level of putting out fires, but the slow-burn of everyday operational fires of triaging an influx of alerts, tracking down duplicate tickets and moving information from A to B because they can’t talk to each other. The work inevitably piles up, creating a constant heat that pulls teams away from the strategic work - and potentially larger fires to put out. 

Agentic AI is already changing that, helping teams handle the repetitive operational work that takes up the bulk of their daily bandwidth, spotting problems early and figuring out what they mean in context before acting on them without the usual manual oversight needed. 

But we need to recognise that this is the easy part. The harder question is what happens when teams want to move beyond one agent doing one thing to expand across different workflows and business functions. The issue isn’t whether any single agent can deliver. It’s whether multiple agents, built by potentially different departments or vendors,  can operate coherently together to plan, act and embed automation directly into how an organisation runs. 

As agentic adoption climbs, so does the complexity 

The shift towards agentic systems is no longer confined to IT. Increasingly, functions including finance, customer support and operations, sales, marketing, risk and compliance are seeing these systems shape the order in which work happens and the decisions that follow.

Investment patterns for agentic systems in the US reinforce this, with leaders carving out budgets (around 25%) dedicated for agentic capabilities rather than treating them as a secondary investment or sub-feature of ‘general AI’.  The same direction of travel is visible in the UK, with AI adoption moving into the mainstream for a sizable portion of businesses, with over a quarter reporting they were already using some form of AI in 2025. 

As that baseline becomes normal, the focus shifts naturally towards the next layer of maturity: how agent-driven systems can operate reliably in environments that are already complex. 

Scaling steadily and navigating the obstacles  

This is where the real test begins for IT leaders. What works well in a small, neatly defined deployment can start to buckle once agents are asked to coordinate across platforms and teams. In other words, the question is no longer whether an agent can complete a task, it’s whether a growing number of agents can make compatible decisions when they share responsibility for the same outcomes. 

The problem is rarely that agents fail outright. It is that their decisions begin to overlap and override one another. Each may act sensibly based on its local context but without a shared view, the combined outcome can be messy, creating duplication, contradictions and new issues that teams have to spend even more time untangling. Frictions created by these conflicting processes are not a result of agents being unable to perform, its because they weren’t given the right guardrails on how to connect and work off one another.

Interoperability should be seen as the foundation

As agentic systems are plugged into core platforms such as CRM, ERP and service operations, those knock-on effects become unavoidable. Actions taken in one area ripple into another. At that point, simply adding more intelligence does not improve performance and can even make it worse by increasing the volume of decisions without improving alignment. 

Agentic systems don’t behave like traditional software, where the same input reliably produces the same output. They take in new information, adjust what matters most, and change course as conditions shift. That’s the point, but it also makes it harder to keep multiple systems aligned. 

When scaling, progress depends on shared context. Agents need enough awareness of one another’s actions and intent to avoid working at cross purposes. Without it, activity can increase without much material improvements. 

The organisations that are progressing at pace, bake interoperability in from day one. They decide up front, which agents are allowed to take action and how information moves between systems instead of hoping it sorts itself out over time. As agents are trusted with more critical workflows, end-to-end visibility stops being a nice-to-have. 

What does this mean for IT leaders? 

IT leaders will see the strongest ROI when agentic systems are allowed to work across platforms rather than being boxed into a single workflow; value shows up when agents can connect data and decisions end-to-end so the organisation can respond as one system instead of a set of disconnected teams. 

That puts integration and visibility at the top of the agenda. Research repeatedly points to the same constraints: fragmented environments and capability gaps that make it difficult to take AI beyond pilots. As agentic adoption matures, interoperability becomes less of a technical preference and more of an operating requirement. 

For executives, the difference between experimentation and impact is measurable and will be clearly seen. Faster incident resolution and real productivity gains make it obvious where these systems are improving performance and where they are introducing new complexity that still needs tightening. 

Agentic AI will continue to advance, that’s a given. But its trajectory and benefit inside the enterprise will depend on how well organisations adapt. The next wave of impact won’t come from smarter individual agents, it will come from smarter ways of getting those agents to work together efficiently.