AI Agents as Digital Employees in Industrial OT
AI Agents as “Digital Employees”
The integration of AI agents into industrial environments represents a fundamental shift in operational technology (OT) management. Unlike traditional automation tools, AI agents function as “digital employees” capable of working 24/7 without fatigue or distraction. This parallel to the human workforce is not just metaphorical but functional — these systems require clear objectives, access to quality data, proper guidance, and secure working environments.
In OT contexts, where mission-critical systems demand consistent performance, AI agents should initially be deployed in non-critical operations.
AI agents represent an emerging class of software systems that combine artificial intelligence capabilities with autonomous operational functions.
- As defined by Google, they are “software systems using AI to pursue goals and complete tasks, with reasoning, planning, memory, and autonomy.”
- AWS describes them as “software programs that interact with the environment, collect data, and perform self-determined tasks to meet goals — rational agents.”
- IBM characterizes them as “systems or programs that autonomously perform tasks, design workflows, and use tools, encompassing decision-making and problem-solving.”
In industrial contexts, these definitions converge to describe AI agents as digital systems capable of independent operation, environment interaction, and goal-directed behavior — functioning much like skilled employees within carefully defined parameters.
According to McKinsey, telecom companies could unlock up to $250 billion in value by 2040 by implementing advanced responsible AI practices, turning exceptional customer service into a strategic business advantage. The key advantage is their ability to combine broad IT knowledge (how to process and analyze data) with domain-specific OT expertise (what data matters and how to interpret it). This combination enables continuous monitoring and analysis that would be impractical for human teams alone.
Requirements for Effective AI Deployment
For AI agents to function effectively as digital employees, organizations must provide detailed instructions and comprehensive data access. Unlike traditional systems that follow rigid programming, AI agents require context-rich guidance to perform effectively. As with human interns, failing to provide sufficient context results in underperformance.
The input context must include:
- Specific operational parameters and their acceptable ranges
- Historical patterns and known anomalies
- Clear explanation of which metrics require immediate attention
- Detailed protocols for different scenarios
Consider a simple example: a misconfigured network depicted in a network topology diagram created with Provize Suite (Figure 1). Three end devices — PLC, Sensor 1, and New Sensor 2 — are connected through three switches. The Belden network was initially configured according to best practices by an external consultant, placing the PLC and Sensor 1 in the same VLAN. However, New Sensor 2 and Switch 3 lack a valid VLAN configuration, preventing the PLC from reaching New Sensor 2 through Switch 3.
Figure 1. Network topology showing a VLAN misconfiguration that prevents PLC-to-sensor communication.
In the existing setup, an operator needs a network admin’s help — external or internal — to configure Switch 3 and New Sensor 2 to fix the connectivity issue, which may take days or weeks.
With a generative AI approach and proper data, this troubleshooting can be offloaded to an AI agent.
An example dialogue between an operator and an AI agent might look like:
Operator: Why can’t PLC connect to New Sensor 2?
AI Agent: The connection issue is due to VLAN misconfigurations. Analysis of the network topology shows that while PLC and Sensor 1 share a properly configured VLAN, Switch 3 and New Sensor 2 have invalid VLAN settings that prevent communication with the PLC.
Operator: Please fix it.
AI Agent: I’ll correct the VLAN configuration to enable communication between PLC and New Sensor 2 by: 1) updating Switch 3’s VLAN settings to match the network standard, 2) configuring New Sensor 2 with the appropriate VLAN ID. Changes have been applied. Please confirm connectivity.
Operator: Confirmed.
When properly configured, these digital employees can monitor critical infrastructure continuously, delivering concise and valuable insights with explanatory chains that reference actual data values. This allows human operators to quickly validate AI recommendations, building trust in the system over time.
The explanatory component is crucial — effective AI agents don’t just flag anomalies but provide reasoning that references specific data points, timestamps, and operational context. This approach mirrors how experienced human operators would communicate findings, making the insights more accessible and actionable for technical teams.
Implementation Framework: AWS Bedrock
The VLAN troubleshooting example is not merely theoretical — it represents a real implementation by Belden using the AWS Bedrock AI Agents framework. AWS Bedrock provides a production-ready framework that implements AI agent orchestration and tooling capabilities.
Figure 2 illustrates the AWS Bedrock implementation process, which follows these key steps:
Figure 2. AWS Bedrock AI agent configuration workflow.
The steps are:
- Agent Creation: Define the agent’s name and description, providing context for developers managing the implementation.
- Permission Configuration: Establish the agent’s role with appropriate permissions for accessing services and data.
- Model Selection: Choose an optimized LLM to orchestrate the agent’s actions — in this case, Claude 3.5 Sonnet.
- Instruction Configuration: Provide detailed plain-English instructions for specific tasks, such as VLAN troubleshooting steps and the exact APIs to call.
- Tool Integration: Define the functions and APIs available to the AI agent, including parameter requirements and capabilities.
- Testing & Deployment: Validate the agent’s behavior immediately after configuration.
This implementation pathway is streamlined through AWS services, leveraging AWS Bedrock, Lambda, and other components for rapid deployment and integration.
Building the Foundation for AI Success
To reach the full potential of AI agents, the industry needs to work through several essential steps that accompany any new technological introduction. We are currently at a stage where there is no standardization, industry discourse has not yet aligned on a commonly accepted view of industrial AI, and businesses move forward with careful skepticism toward the new technology. To utilize AI agents effectively, organizations must inevitably invest in mature and cost-effective data infrastructure. Both digital and human employees need to operate from the same foundation of raw facts to work effectively.
This infrastructure requires:
- Reliable data collection systems with appropriate redundancy
- Standardized data formats and storage solutions
- Clear data governance policies
- Robust cybersecurity measures
- A comprehensive API strategy for seamless system integration
When implemented correctly, this foundation leads to new operational capabilities — most prominently, shifting from reactive to proactive management approaches. With AI agents continuously monitoring complex systems and surfacing only the most relevant information for human review, a multiplier effect emerges, enabling businesses to maintain increasingly complex networks despite challenging industry-wide talent shortages.
In the next article in this series, we explore practical approaches to developing a mature, cost-effective data infrastructure that supports both human operators and AI agents.
Originally published in Computer & Automation magazine.