The way we define the “future” of artificial intelligence (AI) is constantly changing. Go back just a few years, and the conversation was centered on the burgeoning potential of large language models (LLMs) and generative chatbots. Before that, it was machine learning (ML) models for anomaly detection. Jump ahead, and you end up including artificial general intelligence (AGI) that can match or pass human capabilities. As it stands in early 2026, the next tangible evolution of these applications across the enterprise landscape, including the industrial sector, is agentic AI. In fact, IDC forecasts that nearly half (45%) of organizations will orchestrate AI agents at scale by 2030, with manufacturing identified as a priority sector.
Even so, market excitement around agentic solutions does not always map to on-the-ground realities. According to McKinsey research, while 23% of organizations report scaling an agentic AI system somewhere in their enterprise, and another 39% have begun experimenting with these solutions. No more than 10% of companies are actually deploying AI agents at scale in any individual business function.
This reality is reflected in the development and deployment of Industrial AI agents. While promising in theory, these agents can be incredibly difficult to incorporate into legacy infrastructure. Actual implementation goes beyond just researching, building and deploying agents; it requires addressing data quality, structure and context challenges that too often plague industrial environments. Ultimately, AI agents will only deliver value in the industrial sector if manufacturers first assess and mature the data architecture around them.
How agentic AI changes the industrial landscape
To begin addressing these data challenges, manufacturers first need to understand how agentic AI solutions differ from their legacy systems.
Agents are much more than just analytics dashboards in industrial environments like manufacturing plants, storage and distribution warehouses and utilities facilities. They are not monolithic platforms, nor do they operate like traditional manufacturing execution systems (MES), data historians and quality systems. Instead, AI agents are task-specific, autonomous (or semi-autonomous) applications that can execute specific operations across both operational technology (OT) and information technology (IT) systems. They’re applied to specific real-world use cases and used not just to analyze data, but to actually act on it. These agents are assigned roles that require cross-system data orchestration and context-aware decision support for engineers and operators. In practice, maximizing the value of these solutions involves creating hyper-specific agents for hyper-specific tasks. For example, a single work cell might include specific agents for quality, maintenance, scheduling and supply chain management. Each of these agents requires data from multiple systems and equipment with different contextual information, delivered by model context protocol (MCP) tools.
Multiply this across work cells, production lines and facilities at an enterprise scale, and the challenge of supplying the correct data to each agent becomes incredibly daunting with current infrastructure.
The data maturity gap holding agents back
The current state of industrial data across most modern manufacturers is siloed OT and IT systems, inconsistent naming conventions and schemas, missing context and a heavy reliance on undocumented or tribal knowledge. This makes it difficult for industrial AI agents to find and leverage the right data.
Figure 1: The current state of industrial data across most modern manufacturers is siloed OT and IT systems.
Similar challenges were manageable in the past because humans could consume partial information from multiple systems, validate the information and make decisions based on experience. But as AI agents multiply exponentially across industrial data infrastructure while operating at machine speed, they’ll quickly expose the fragility of the status quo by turning small data issues into large-scale systemic failures. Any error in agentic functions — such as predictive maintenance agents acting on data from the wrong asset or using the incorrect telemetry values, or quality agents misinterpreting sensor data without process context — can easily grow into a larger issue that gridlocks workflows, halts operations, ships faulty products and degrades the trust in the entire manufacturing plant.
This makes improving the overall health and quality of a facility’s data infrastructure an operational requirement, not simply an IT project. Only by strengthening their overall data maturity can any manufacturer support the fast, safe deployment of automation and agentic workflows.
What 'AI-ready' industrial data looks like
Not all industrial data is needed to drive these solutions. In fact, that kind of volume would quickly overload the systems and agents. However, the data that is made available to the agents must be usable and reliable. Data prepared for agentic use should be contextualized, governed at scale and optimized for use-case tasks rather than broad use reports. To ensure this level of actionable quality, manufacturers should prioritize the adoption of the following:
- Versatile open protocols, such as the MCP, are built specifically to support the data requirements of AI agents. Coupling protocols such as MCP with industrial DataOps solutions can aggregate and contextualize data from various sources and expose it as tools for agent discovery and use. The MCP server provides data to the AI agents, uniquely curated for them and their needs.
- Robust industrial DataOps, with data pipelines to source and curate data for the MCP tools provide the required data to the AI Agents. This also includes pipeline monitoring, the detection and resolution of data quality issues and ensuring that agents operate seamlessly with the high quality, reliable tooling. By observing pipelines and connections, teams can better assess how data is transformed throughout the process and what changes are needed.
- Strong data governance enables clear ownership, accountability and standardized definitions across OT and IT data. This will help ensure that hyper-specific agents operate as intended, taking safe and appropriate action and avoiding the hallucinations and errors that can come with overexposure to data sources and additional tools.
Strengthening data maturity is not a “rip and replace” process. Instead, it should prioritize incremental progress that modernizes architectures atop and alongside existing systems and enables agents to operate within defined boundaries. By taking this deliberate approach, manufacturers can build a foundational data infrastructure that enables them to adopt agents faster and with less outright risk.
A foundation for agentic success
For those who prepare accordingly, agentic AI will reshape industrial automation. Early investment in the admittedly less-than-glamorous work of strengthening data infrastructure and improving overall data maturity will lay the foundations needed to support agentic operations well into the future. Readiness and maturity will ultimately be rewarded with reliable success.
