The “Automation Blindspot": Is flawed edge data weaponising enterprise AI against the network?

The conversation around artificial intelligence in the enterprise is shifting from "how do we adopt it?" to "why is it breaking our systems?"

 

As businesses hand more responsibility for IT operations to autonomous AI agents, they’re discovering a harsh truth: agentic AI doesn’t fix chaotic infrastructure; it automates the collapse. Unlike generative AI, which still relies on people to decide what happens next, agentic AI is designed to act. It investigates problems, decides what to do next, and independently intervenes. That’s exactly what gives it the potential to transform IT, but also what makes poor operational data so much more dangerous.

 

Too much of the conversation focuses on how capable autonomous AI has become, and not nearly enough on the quality of the information it’s acting on. If that data is fragmented, intermittent, or missing critical context from the network edge, automation doesn’t strengthen resilience. It simply allows poor decisions to happen faster.

 

That’s why I believe enterprise networks are approaching a turning point. We’re asking autonomous systems to make increasingly complex operational decisions, yet many organisations still don’t have a complete picture of what’s happening at the edge of the network. 

 

Before we hand over the keys, we need to make sure we’re giving those systems the information they need to make the right decisions. To prevent autonomous agents from turning into network nightmares, organisations need what I think of as a "noise-cancelling" data strategy for AI, one that filters out operational noise and gives autonomous systems trusted, high-fidelity telemetry to work from.

 

 

The Toxic Loop of Autonomous Remediation

One of the biggest misconceptions about agentic AI is that better models automatically produce better outcomes. In my experience, they don’t. The quality of the data guiding those decisions is just as important. 

 

The difficulty is that most monitoring tools still rely on snapshots of network health. That’s enough to identify broad trends, but it often misses the short-lived events that shape an employee’s actual experience: brief packet loss, momentary latency spikes or device-level issues that disappear before the next polling cycle. An autonomous agent doesn’t know what it can’t see. It simply reaches the best conclusion it can from the information available.

 

For me, this is where the conversation shifts from experience management to experience engineering. Rather than waiting for employees to report problems through the service desk, organisations need to engineer environments where autonomous systems have reliable operational signals from the outset. If we expect AI to make decisions on our behalf, we have to give it a complete and accurate picture of the environment it’s operating in.

 

Consider a regional office where users suddenly begin experiencing application lag. A traditional monitoring platform reports normal bandwidth utilisation because traffic has been averaged over five-minute intervals. The autonomous agent assumes the network is the problem and redirects traffic through another gateway or changes router configurations to improve performance.

 

In reality, the problem sits somewhere entirely different. A corrupted driver on a batch of laptops is causing memory leaks and degrading performance locally. The network changes do nothing to fix the underlying issue, but they do introduce unnecessary disruption across the wider environment. The agent detects performance falling even further, concludes another intervention is needed and makes additional changes. Before long, a relatively minor endpoint issue has escalated into a much larger network incident.

 

 

Embedding SRE into End-User Computing

Relying solely on centralised infrastructure monitoring is a little like trying to steer a ship through fog using only a lighthouse. You know where you’re heading, but you have very little visibility of what’s directly in front of you.

 

That’s why telemetry from an end-user’s device matters so much. It fills the visibility gap that traditional infrastructure monitoring often misses, giving autonomous systems context before a minor issue turns into a major outage.

 

For years, we’ve treated the data centre and the cloud as engineering disciplines. Site Reliability Engineering (SRE) has helped organisations balance innovation with resilience by replacing reactive operations with measurable objectives, automation and continuous improvement. End-user computing has remained far more reactive, shaped by support tickets, manual intervention and unpredictable desktop environments.

 

I don’t think that divide can survive the arrival of agentic AI. If autonomous systems are going to make operational decisions on our behalf, they need the same engineering discipline at the endpoint that we’ve spent years applying to cloud infrastructure. 

 

Industry analyst Gartner predicts that, by 2028, 80 per cent of enterprises will adopt SRE practices as IT leaders look to reduce the burden of "keeping the lights on" and create more capacity for AI initiatives. I think that’s exactly the right ambition. Where I believe the industry is getting ahead of itself is in assuming autonomous systems can compensate for poor operational data. 

 

Without meaningful baselines, autonomous systems have no reliable way of distinguishing genuine degradation from normal variation, reducing the risk of unnecessary or even damaging automated interventions.

 

 

The Clean Blueprint for AI Success

Fragmented, unmapped IT environments amplify autonomous AI failures. An agent cannot safely optimise systems it doesn’t fully understand. If hidden dependencies, ageing infrastructure and inconsistent telemetry are left unresolved, AI simply inherits those weaknesses.

 

Organisations spend enormous time evaluating models and autonomous workflows while paying far less attention to the operational data those systems depend on every second they’re running.

 

Clean, structured telemetry is the blueprint that makes safe autonomy possible. It gives autonomous systems a continuous understanding of what’s happening across the environment instead of forcing them to infer the state of the network from partial information.

 

Ultimately, the biggest challenge isn’t building smarter autonomous systems. It’s building environments that deserve to be automated. Agentic AI has enormous potential to transform enterprise IT, but if we want it to strengthen resilience rather than undermine it, we have to start by giving it a clear, continuous and trustworthy view of the environment it’s expected to manage.

 

---

 

Dan Salinas is COO at Lakeside Software 

 

Main image courtesy of iStockPhoto.com and nopparit