By VimalRaj Sampathkumar
Enterprise AI stalls not because the technology fails, but because companies deploy autonomous agents into workflows still designed around humans making every final decision.
Something unusual is happening in the world of enterprise technology. Recent research revealed that nearly eight in ten companies report using generative AI in some form and yet simultaneously, an equally identical number has reported that they’ve seen no meaningful impact on their end result.
So if the industry doesn’t get honest about why this is the case, the next wave of AI budgets and investments will end up as badly as the last.
The issue isn’t with the technology in itself. It’s the role that enterprises have assigned for it to play.
So much AI investment, so little impact
The hardest pill to swallow here is that enterprise-grade AI wasn’t initially built to push out huge outcomes. It was built to make humans more efficient at getting the desired results out by themselves.
The first wave of AI at least, was meant to be simply assistive on a fundamental level. It’s place was next to already existing workflows, just to offer opinions and key information whenever needed. The human in the loop still made the final call. At the end of the day, AI was merely meant to be a mere catalyst, just a faster way to get things done.
But now, AI adoption is widespread, while meaningful impact is not. That gap isn’t due to the lack of training, or employees who do not want to upskill themselves. It boils down to how it was positioned in the first place.
What actually makes agentic AI different?
The distinction between copilots and AI agents doesn’t actually seem like a lot, but it’s quite crucial that we comprehend it.
Copilots are designed around the basic assumption that a human will always be in the loop, making the final call. However, agents are designed around the complete opposite assumption, with humans stepping in only when something genuinely needs their intervention.
This shift is already playing out across our IT service management, endpoint management, observability and security operations, where autonomous execution is replacing the assisted model entirely.
Beyond monitoring: AI Infrastructure that acts
IT and SecOps are where the gap between AI’s promise and its eventual delivery has been the highest. These teams have been plagued with issues for years: understaffed, and burdened with ever-increasing volume of alerts.
The data justifies this as well. ManageEngine’s latest research into the operational resilience of UK and European businesses in 2026 found that 60% of IT and security professionals reported increased pressure on their teams over the past 12 months. Nearly half of the respondents cited a skills gap as their single biggest challenge, and 43% flagged AI-powered attacks as their top risk for the year ahead.
There’s a deeper irony that becomes apparent in the last sentence. The AI being weaponised by attackers is the same AI that could be defending against them – and the difference is just a question of how it’s being deployed.
In more mature environments, this is playing out differently already. Infrastructure that doesn’t wait for a human to notice something is wrong, systems that catch the anomaly, run the patch, and confirm the fix – often before a human has even checked the alert, let alone act on it.
Your AI pilot worked. So why hasn’t anything changed?
The organisations that get the most out of this paradigm shift won’t be the ones that deploy the most agents, or made the most expensive AI investments. They will be the ones that actually restructured based on the resources deployed, rather than just investing in and dropping agents into workflows that were never meant for them in the first place.
Plugging an agent into an existing workflow is like putting a Formula 1 car on a road full of speed bumps. The car isn’t the problem — the road is. Most organisations keep tinkering with the car. The ones seeing results actually rebuilt the road.
In theory, getting it right tends to come down to the following basic factors:
- Be clear about where human involvement comes into play. Agents work best when it is decided upfront on what they are supposed to handle, and what triggers may mean it’s time for human involvement. The boundaries should be visible, well documented, and customisable according to the demands of the workflow.
- Connect agents to systems where they can actually act. An agent’s ROI multiplies with its access to real data and actions. Agents connected to service desks, endpoint management tools, identity directories and cloud environments don’t even need a human in the loop in most cases. The ability to extend that connectivity further — through an open ecosystem of partner-built integrations and agents — is what separates platforms that scale, from ones that fail at the pilot stage itself.
- Observability should be given top priority. Autonomous systems need more transparency. Every action and decision should be done with full traceability. When something goes haywire, teams need to understand exactly what the agent did, in what order and why. That same visibility is also what lets agent performance improve over time.
Beyond operational visibility, this is also where AI governance enters the picture. As agentic AI matures across most markets, the conversation has shifted toward accountability: who is responsible when an autonomous system makes a consequential decision, and what approval trails exist to support that. Platforms that build governance in from the start, rather than bolting it on after the fact, will be the ones that earn the institutional trust needed to scale.
Security teams are drowning. The lifeline? AI Agents.
Security operations are where the probability of getting this wrong is the highest — and where the case for agen.tic AI is hardest to argue against. With 43% of UK organisations identifying AI-powered attacks as their leading risk, the threat landscape is moving faster than teams built around human response times were ever designed to keep up with.
Security teams aren’t dealing with just a handful of threats a day, they are dealing with hundreds, sometimes thousands of signals, most of which turn out to be noise. The time spent figuring out what’s real, what’s probably building up to something bigger, and what can be safely ignored is so much precious time not spent on the things that actually matter.
Agentic AI doesn’t solve this issue by being smarter than a security analyst. It solves it by handling the volume that was never a good use of an analyst’s time in the first place. Agents sort through the noise, connect the dots across systems, and surface the things that genuinely need a human decision. The analyst stops being merely a problem solver and starts to be what they were hired to be — someone who investigates, interprets, and makes long-term decisions that a system alone can’t.
The bottom line is – the goal isn’t to replace security expertise. It is to stop wasting it.
The lesson the first wave of AI left for us
The disappointment caused by AI these past few years wasn’t exactly inevitable. It was the apparent result of looking to fit a new kind of technology alongside the same old ways of working.
The organisations that figure this out first won’t run leaner. They will end up being harder to disrupt, faster to recover, and less dependent on heroic individual effort or leadership involvement when things go wrong. And that’s the real enterprise resilience story.
About the Author
VimalRaj Sampathkumar, is a Technical Head – UK & Ireland, at ManageEngine. He is Presales and Strategic Accounts Manager with 15+ years of experience in Technical Sales, Account Management and Customer Success. He has deep technical expertise in consulting and implementing ITSM, ITOM, SIEM, End-point Management, CRM, ATS, and HCM/HRIS applications globally. His expertise has been to drive revenue and market share increase by consistently delivering customer-focused solutions, demonstrating product value, and building the foundation for loyal, long-term customer relationships. He enjoys playing cricket, reading, and travelling in his spare time.
