Agentic AI promises systems that can reason, plan, and act autonomously. But adoption hinges on whether an organization’s IT environment can truly support that level of autonomy. In this article, I explain why companies must first close fundamental gaps in observability, root cause analysis, workflows, and data trust before enabling agentic AI. Without that foundation, even the most advanced AI will struggle to produce consistent, reliable outcomes, backed by insights from professionals across the industry.
Today, the moment agentic AI systems go live in production, they all hit the same wall: AI cannot make reliable decisions because it has no understanding of the root causes of why things actually go wrong in the real world.
A device might slow down for unclear reasons, an app may crash only during specific workflows, or a patch could impact certain users without a clear explanation from existing monitoring tools.
This struggle to identify root causes is not about a lack of tools. If anything, IT teams are drowning in tools, from dashboards and alerts to machine-generated insights. Instead, the issue is fragmentation, as these tools are often scattered across systems that were not designed to work together. Endpoint tools, SaaS analytics, network telemetry, and the IT service management platform all tell different stories. By the time the IT team needs to act, critical context has already been lost across the gaps.
In practice, vital information is lost as incidents pass through the workflow. Level 1 agents must guess solutions. Level 2 engineers repeat diagnostic work that has already been done. Level 3 teams receive escalated cases with only symptoms, not explanations. AI built on this structure inherits these limitations and magnifies them.
IBM’s Jennifer Kady identified the core issue: AI agents only work when backed by detailed first-party endpoint context. But fragmentation kills speed. Many organizations implement AI-driven monitoring and automation expecting quick wins, then spend weeks or months integrating data, tuning models, and bridging visibility gaps. The promise of quick benefits fades as teams realise their IT environments cannot deliver the consistent, high-quality evidence that AI demands.
Without that evidence, autonomous capabilities stall at the proof-of-concept stage. Service providers observe this issue on a large scale.
Preparing for agentic AI requires tackling the root of these issues. It starts with understanding that true root-cause clarity relies on continuous, high-quality, context-aware telemetry at the points where problems truly start: the endpoint. Many digital disruptions begin long before a ticket is raised, stemming from minor changes in resource usage, interactions between background processes, or differences in environmental conditions across devices. Snapshots fail to capture these dynamics. Sampling overlooks crucial moments. When telemetry is divided across multiple tools, the story becomes fragmented.
To move toward readiness, organizations require continuous visibility into how devices and applications behave over time, rather than just whether they meet a specific threshold at any given moment. This ongoing observation serves as the foundation for accurate diagnostics. It allows IT to trace issues from symptoms back to their source, rather than relying solely on correlations. It also prevents duplicate workflows and unnecessary escalations by ensuring that each diagnostic step is documented and not repeated.
At a global level, the importance of contextual continuity becomes even clearer.
Preserving context as data moves across the estate is equally important. When information flows through IT service management systems, maintaining the full depth of telemetry data is critical. A rich telemetry signal from an endpoint needs to retain its surrounding context, such as user activity, network conditions, patch history, and changes to configuration, as it moves between teams and tools. Without this contextual information agentic AI cannot function successfully – it is being asked to operate in an information void.
Trust is non-negotiable, as autonomous systems must justify every decision. When an AI recommends a solution or identifies a root cause, engineers need to understand the reasoning and confirm the evidence. AI that cannot explain itself will not gain trust. Organizations preparing for agentic AI must focus on transparency and auditability. Every insight and action must be supported by clear evidence.
The real shift is not about technology, but about understanding, where digital experience is not viewed as something to review only after the fact. It is the system where agentic AI either thrives or falters. Tougher questions need to be asked, such as: can we trace problems back to their source without unnecessary work; does the context for diagnostics carry over during transitions; when AI takes action, can we assess its reasoning; and when actions do not succeed, do we understand the reasons?
Here is what truly matters: agentic AI will not come from larger models or more automation alone. It needs environments that can clarify themselves. Organizations that are building that foundation now, improving root-cause analysis, linking telemetry, removing workflow obstacles, and maintaining context, will not just adopt agentic AI more quickly. They will implement it with confidence, gain the trust of their teams, and turn the idea of autonomy into a solid reality.
The author
Dan Salinas is Chief Operating Officer, Lakeside Software
