Over the past several years, AI models have become increasingly capable of applying advanced reasoning to enormous amounts of information. But most organizations’ data stores have not kept pace, and enterprise data environments are often still characterized by fragmentation and inconsistency.
Even when AI tools have broad access to enterprise data, they frequently lack the business context needed to make sense of this information.
As enterprises move from generative AI tools that answer questions to agents that take actions, that missing context layer is becoming a significant constraint. In May, Gartner projected that organizations that prioritize semantics in AI-ready data will increase agentic AI accuracy by up to 80% and reduce costs by up to 60%.
"Agentic AI outcomes depend on context," said Rita Sallam, distinguished vice president analyst at Gartner, who defined context as a clear understanding of the specific relationships and rules within an organization's data. "Without context, AI agents cannot operate accurately and are far more likely to hallucinate, introduce bias and produce unreliable results."
The challenge surfaced repeatedly at Splunk's September .conf26 conference in Denver, where the word "context" echoed from the keynote stage to panel discussions, press briefings and one-on-one conversations.
"Models now provide the reasoning, and GPUs provide a lot of the processing, but what they are lacking is the operational context," said Mangesh Pimpalkhare, senior vice president and general manager of Splunk platform for Cisco, during a keynote about product advancements designed to turn machine data into operational context for AI.
"Agents need the context and intelligence to drive predictable actions, not just actions," he said.
During a theater session in the conference pavilion, Kevin Carroll, director of next-generation data services for the global tech solutions integrator Insight, said many organizations operate with "more information and less clarity."
"There's not a shortage of data," Carroll said. "There's a shortage of context."
What Blocks AI Agents From Acting on Enterprise Data?
Silos cause more trouble over time. Telemetry piles up across dashboards and tools, and no single system stitches it together for an agent to read.
Ronak Desai, co-founder and CEO of the AI site reliability engineering (SRE) startup Ciroos, said traditional incident investigation amounts to "click ops," with engineers jumping between dashboards for the frontend, backend, cloud, network, data center and other environments.
Many enterprises, he said, use between 15 and 20 monitoring and operations tools. When something breaks, organizations set up "war rooms" to gather specialists from different silos.
"There's no single person, or there is not a single system, where you have all the data and insight," Desai told The Forecast.
"We have a ton of signals," said Niall Murphy, distinguished engineer at Ciroos. "We're awash in data, and we also know roughly the actions we have to take to fix things. But we have this missing layer of insights in the middle."
The problem, Desai said, is not that enterprises lack telemetry. It is that this telemetry lives in eight to 10 different tools, with no one responsible for consolidating it. Infrastructure and application signals sit across cloud providers, log managers and APM platforms, plus CI/CD, Git, ticketing and chat environments.
"Enterprises don't have a data problem," Desai said. "They have a reasoning and understanding problem."
Desai said enterprises need a distributed reasoning layer that operates across systems and platforms, bringing together the information that information agents need. Ciroos leaves IT observability data where it resides but "brings the intelligence there," using multiple AI agents that collaborate to analyze alerts, determine root causes and recommend fixes.
"I spent years at Cisco building the model where you bring everything into one big data lake," Desai said.
"The friction that created is exactly why I started Ciroos. Data in the enterprise is distributed, and it's going to stay that way. So, reason across it where it sits, and let the customer keep their tool choice."
Murphy said observability tools hold a wealth of data, and professionals can quickly see the results of their efforts, making IT observability a proving ground for the context-rich, agentic workflows leaders hope to deploy across business units. "We have faster feedback loops in observability, and that's why we see more traction," he said.
How Are Enterprises Building Context for Agentic AI?
During a panel discussion titled "Building Agentic Advantage with Your Data Foundation," several experts and practitioners outlined real-world examples of organizations overcoming context constraints with full-production agentic implementations.
Rajashekar Ramdass, senior vice president of technology, transformation and managed services at the Indian multinational technology-services and consulting company Wipro, offered the example of a consumer-goods client using agents for invoice processing and accounts payable. That customer reached up to 70% autonomous capability, he said, although getting there took six to nine months.
That sort of autonomy, Ramdass said, requires more than feeding an agent the data associated with a task. Agents also need enough context to understand how an event affects the wider business.
"When we talk about root-cause analysis, the context fabric becomes very important for how we bring that data together," he said.
During a panel discussion at .con26, Karun Subramanian, senior director of AI for infrastructure and operations at UnitedHealth Group, highlighted the company's homegrown root-cause-analysis agent. The most significant obstacle to standing up that solution, he said, was providing the agent the information it needs from disparate observability, IT service management and other systems.
"The biggest challenge we had when it comes to implementing a root-cause-analysis agent in production is access to the right data, at the right latency," Subramanian said. The agent began at roughly 10% to 15% accuracy, and although its performance is improving, it has not yet reached the 80% to 90% level the organization would like.
Full autonomy, Subramanian said, will require not only context-rich data but also reliable controls. "Really, to get to 100% autonomy for agents, guardrails need to be 100% accurate," he said.
Hanlin Fang, vice president of product and head of Cisco Data Fabric at Splunk, said organizations ultimately need to connect technical signals with business impact, risk and policy considerations, and continually reassess that context as conditions change.
"Moving forward with agentic AI is not just about a technical solution, or just business outcome," Fang said. "It has to be always the risk, business outcome and technical ability altogether. It's a never-ending process."
Related:
Calvin Hennick is a contributing writer. His work appears in BizTech, Engineering Inc., The Boston Globe Magazine and elsewhere. He is also the author of Once More to the Rodeo: A Memoir. Find him on LinkedIn.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
