New research from Enterprise Management Associates (EMA) suggests that the first wave of AI adoption—centered on chatbots and virtual assistants—is succumbing to an AI agent-driven model.
In a survey of 458 IT professionals actively using AI in network operations, EMA found that just 15% prefer traditional chatbot-style interfaces. The same respondents also reported the lowest levels of success with their AI initiatives. Organizations leaning more toward agentic environments in which systems continuously analyze conditions, recommend actions, and collaborate with human operators, are seeing strong results, according to the EMA data.
“One of the things that we discovered is that the era of chatbots is over. Only 15% said they’re focused on one-on-one interactions with virtual assistants, where you ask a question that gives you information. The people who said that this was their preference are the ones who were getting the least amount of value out of AI,” said Shamus McGillicuddy, research director for network management at EMA, during a recent webinar. “A little more than a third (33.6%) of them said they wanted AI-enabled collaborative workspaces. There’s an agentic environment that’s sort of telling you what it’s seeing and what it thinks is happening, and what it suggests to do about it, and then the whole team can chat about it too and chat with the agent.”
From answers to action
The evolution from chatbots to agentic AI is about taking action. Chatbots operate reactively, replying to questions. Agentic AI proactively delivers insights that it garners from being directly embedded into workflows, according to EMA.
As one-third of respondents indicated, AI-enabled collaborative workspaces enable operators and AI agents to interact in real time. Another 19% favor proactive systems that flag emerging issues and suggest remediation steps before human operators begin troubleshooting. This type of capability closely aligns with the ultimate goal of predictive network operations, according to EMA.
“We want to get to a point where AI is telling us there’s something wrong that we should look at. We’d also like to suggest playbooks that it can run to fix those issues. That is very much not a chatbot. That is an agent, an agentic environment,” McGillicuddy said.
Among the expected business benefits of AI-driven network management are:
- Faster resolution of network problems: 54.1%
- Improve network performance/experience: 51.3%
- Reduced security risk: 48.7%
- Cost optimization: 47.8%
- Proactive problem prevention: 45.9%
- More time available for strategic projects: 41.9%
- Responsiveness to change: 37.8%
- Mitigation of network team’s skills/personnel gaps: 33%
“I think AI is going to help us respond to incidents quicker,” said a network infrastructure and operations manager with a Fortune 500 energy company in the EMA report. “It will help us diagnose yellow flags before they turn into red flags. And it will help us reduce our self-inflicted outages.”
Not ready for fully autonomous operations
Still, EMA found that only 35% of enterprises are completely successful with applying AI to network management. Organizations relying on simple interfaces or loosely integrated features are seeing less impact than those embedding AI deeply into workflows and decision-making processes. And just 31% of IT professionals say they fully trust the outputs of their AI tools.
The research also found that human oversight remains critical. In related research from EMA, 63% of organizations said they require human approval for AI-driven automation, which highlights continued reliance on “human-in-the-loop” models, according to McGillicuddy.
“No one’s quite ready for autonomous operations, but human in the loop, definitely, maybe human out of the loop in the future for certain things. These are the top barriers to autonomous agentic IT ops: layering humans and systems and processes together, developing policies and guardrails for compliance and data security, overcoming mistrust and fear, resource gaps and budget and skills gaps, and then establishing roles and responsibility for introducing into production,” McGillicuddy said.