Enterprises have invested big in AI agents with 40% of enterprise applications this year expected to feature task specific AI agents, Gartner predicts. At this level, agents perform complex, end-to-end tasks, such as implementing a new feature in a software application—no human needed.
But for agents to take enterprise results not only from 0 to 60, but from 60 to 100, something more is needed. That thing is context, specifically context that is unique to each enterprise. Back to the software example: that agent would need to know whether the new function already exists elsewhere in the enterprise. It must adhere to specific coding standards and know which code libraries are okay to use within that enterprise. If not, the agents’ actions cannot be trusted for important work.
This kind of context is generated by org-native solutions that connect an enterprise’s code repositories, private data, documentation, and policies so the agents’ suggestions or actions are context-aware and company-specific, just as they would be if coming from a highly trained developer. This kind of knowledge layer learns from an enterprise’s data to arm agents with contextual understanding. With it, they achieve better results while protecting an enterprise’s data from everyone else.
New Computing Shift: New Foundational Layer
The need for a new foundational layer in the agentic era is completely predictable. Every major computing shift introduced a new foundational layer. Without databases, data remains locked in spread sheets or manilla file folders. Without the cloud, computing stays rigid and non elastic. Without virtualization, infrastructure is highly inflexible.
Now, generative AI and agentic tools in the enterprise are hitting the same wall. Last year’s MIT study took the veil off, revealing that 95% of enterprise AI initiatives returned zero in terms of ROI. The primary reason: “Most GenAI systems do not retain feedback, adapt to context, or improve over time,” the researchers found, adding “model quality fails without context.”
New research also reveals the limits of generic AI tools, finding that three of four (76%) of workers say the AI tools they like best lack access to company data or work context, “the information needed to handle business-specific tasks,” research from Salesforce and YouGov reports. At the same time, 60% of workers said “giving AI tools secure access to company data would improve their work quality, while nearly as many point to faster task completion (59%) and less time spent searching for information (62%).”
For enterprises to trust and rely on AI agents —especially teams of agents working autonomously— context will become a central factor that defines whether AI can actually deliver and be trusted to scale. For enterprises, context, along with security and governance, are must-haves for agents to reach their full potential in the enterprise.
No Longer Just the Best Model
No doubt, the models that enterprises use to inform AI agents are important. But LLMs behind genAI are largely based on massive datasets from public data, like the internet. These are static, sometimes outdated, and lacking domain expertise needed to serve specific companies. To counter this, we prompt the AI, and front load with data and context to inform better outputs. That’s too slow for enterprise teams. More than 6 in 10 experts in the MIT study said manually inputting context was a key barrier to preventing use of ChatGPT for mission-critical work.
With true enterprise context, however, an AI agent will act like an experienced employee. It will understand an enterprise’s systems, including entities, relationships, and dependencies, enabling agents to reason about architecture, workflows, and consequences. It will know where to go to get the data it needs to make a decision and all the sources that it needs to access. As a result, the agent will be more accurate and efficient.
Traditional approaches like basic vector RAG were designed to retrieve text, not understand systems. They can find where a term appears, but they cannot reason about architecture, dependencies, ownership, or the downstream consequences of change. In complex environments, “context” is not a document lookup problem. It is a graph problem. Without understanding how services connect, how teams own components, and how changes propagate across systems, AI can retrieve relevant information yet still miss what actually matters.
Software development is a perfect example of why context matters. Everyone wants AI to accelerate development and generate more code, faster. But automation without context isn’t the same as intelligence. The real value comes when AI understands the system it’s working in. With the right context, AI can reason about architecture, understand dependencies across repositories and services, anticipate the impact of changes, and support developers in building software that fits the system as a whole.
Trusted AI: the Next Era
Models provide intelligence. Agents provide action. But enterprises also need understanding. That’s the last-mile problem for AI. Without context, automation breaks in complex systems as every change introduces risk. That’s why context will be a linchpin factor in the next evolution of AI in the enterprise while lack of context will remain an unforgiving roadblock.
