A practice guide for IT and OT professionals to implement AI agents in factories

Why Factories Need AI Agents Now

US Total Factory Productivity has dropped 4.4% in 2008-2023 per LNS Research1 and skilled labor shortage is becoming a “national crisis” per Ford CEO Jim Farley2, and 84% of organizations have been negatively impacted by lost experience1.

AI agents offer a timely response to these pressures.

While industrial automation has existed for several decades, it was primarily rule based and required constant maintenance and upkeep. Now with AI agents that can learn, analyze and action on tasks, wherein we can now automate with intelligence.

We are still in the early days of this transition, but the momentum is unmistakable, with promising use cases rapidly emerging and gaining real traction. 83% of manufacturers believe AI agents will expand capacity, and shop‑floor productivity can improve by up to 20% through generative AI–powered assistance, recommendations, and autonomous systems, according to MIT Technology Review3. Two leading industrial enterprises have both reported that studies they’ve conducted show that 80% of current manual decisions will be replaced by agents6.

Over the past year, we have seen several of these use cases convert from prototypes to measurable impact. One clear example is troubleshooting. Textron Aviation reduced troubleshooting time for complex maintenance tasks of aircraft from 20 minutes to just 1–2 minutes using Azure OpenAI Service4.

Additional high‑value scenarios are emerging quickly. For instance, a “factory coach in your pocket” provides a single, intuitive access point to training materials, OEM documentation and manuals, Standard Operating Procedures (SOPs) and operational guidance. A time study from one OpsMate AI customer reported that skilled engineers spend 20% of their time looking for information and another ~30% of their time performing non-value added administrative tasks, for a total of nearly 50% of their time spent on non-value-added activities that AI agents could readily perform.

Why AI Agents Are Hard to Scale in Factories

Although AI agent adoption is growing rapidly, most manufacturing organizations are still in the early stages. According to McKinsey “The State of AI” report, factory operations remain the least mature area for AI deployment across the enterprise, with more than 90% of manufacturers not yet having experimented with agents at all5.

The primary challenges manufacturers face include:

Inaccuracy and inconsistency of AI outputs.
Cybersecurity risks and the need for secure data handling.
Scalability issues due to the highly customized, “snowflake” nature of each factory.
Jagged intelligence - AI models that excel in some tasks but fall short in others.

Another unique challenge noted by OpsMate AI Advisory Board Members is the lack of empowerment of front-line teams. Traditional tools and solutions require specialized resources to collect front-line requirements and then “configure”, or “code” solutions and these resources are always in short supply and high demand, constraining the ability to democratize problem solving and innovation

Overcoming these challenges requires establishing a strong agent‑scaffolding layer - the system and application architecture that sits on top of raw AI models. This scaffolding makes AI usable, reliable, and deeply integrated into real‑world factory workflows. It includes components such as an AI‑first application stack (memory, entitlements, action space), management and observability tools, and the underlying infrastructure spanning storage, identity, and data platforms.

Practical Guidance for IT and OT Leaders

1. Empower Frontline Teams - Securely and Strategically

As Andrej Karpathy7 recently noted, this wave of adoption is different because technology adoption is being led by end users who experience obvious personal value. In many factories where the absence of approved solutions and governance has not been established, individuals and teams already quietly using public AI tools on their own, often by searching vendor manuals, OSHA standards, uploading images, or troubleshooting production issues.

This unsanctioned behavior is gaining momentum and introduces real risks to data security and accuracy.

A more effective approach is for IT to actively empower frontline teams with approved, secure, and governed AI tools.

Start by giving them access to trusted portfolio of solutions such as Microsoft 365 Copilot for general purpose day‑to‑day information retrieval, summarization, etc.
At the same time, extend these general purpose tools with Microsoft approved domain specific solutions like OpsMate AI that can deliver context-specific, trustworthy and secure manufacturing-specific solutions seamlessly integrated into M365 copilot and the broader enterprise agentic landscape.

This both accelerates value and builds the organization’s confidence in responsible AI adoption.

When to consider OpsMate AI:
When frontline teams need agents that can understand factory context, connect to IT/OT systems, and reason across multiple plants and execute reliable workflows, use OpsMate AI alongside M365 Copilot instead of relying on general-purpose tools alone.

2. Prioritize Use Case and Align on Agent-Scaffolding Requirements

As you look to select AI tools for manufacturing operations, it is critical to prioritize high impact and high feasibility use cases and align on the agent-scaffolding capabilities required to support. The following use cases and requirements reflect the operational realities, aspirational goals, and common desires expressed by modern manufacturers as they begin their AI journey

High-Value Use Cases

Most manufacturing organizations recognize that the highest value starting points for AI include use cases such as:

Equipment Troubleshooting
Quality investigation and root cause analysis
Tribal knowledge capture and curation
Factory “coach in the pockets”
Working standards generation and validation
Generative bill of processes and work instructions
Generative shift handover reports

These use cases typically involve fragmented information, tribal knowledge, multiple systems, reasoning and multi-step workflows - making them ideal candidates for agentic AI.

Common scaffolding capabilities required for factory agents

To support these use cases at scale, manufacturers typically require:

Ingesting and processing large volumes of documents at optimal cost and performance.
Retrieving and interpreting complex engineering and technical Files correctly.
Understanding and reasoning multimedia content stored across multiple systems, repositories, and plants accurately.
Capturing, curating and managing tribal knowledge in the daily workflow, so it becomes an asset, not a liability.
Integrating with and using IT and OT systems for agents to reasoning over and taking actions on.
Organizing the data and knowledge against industrial knowledge graph – plant, equipment, process, people and product, and performing context-aware inferencing.
Fine‑grained access control aligned with organizational structure, job roles, projects, and compliance requirements.
Empowering frontline engineers and technicians to make and continuously teach the AI agent (procedures, operational know-how, IT/OT data, actions) without knowing AI and coding skills.
Defining agent operating procedures in natural language and running long‑running, multi‑step workflows such as root cause analysis, troubleshooting, etc.
Factory appropriate, multimodal user experience that meet frontline team in the environments where work happens, including HMIs, stationary computers, rugged mobile devices, laptops and tablets.
Deterministic guardrails ensure consistency and reliability of agents’ behaviors.
The ability to generate value now, with imperfect data, while incrementally improving and organizing data structures over time - often in conjunction with Microsoft Fabric or Azure Databricks.

Where OpsMate AI fits
OpsMate Ai is designed to address these requirements out of the box for manufacturing operations, allowing teams to focus on adoption, change management, and continuous innovation rather than building and maintaining custom agent scaffolding.

3. Select the Right Technology Stack for Your Needs

Manufacturers pursuing AI adoption recognize the need for agentic capabilities that can ingest large document volumes, understand complex technical files, unify multimodal content, connect IT/OT data, capture tribal knowledge, scale across plants, support front-line natural language configurability, integrate into frontline workflows, and meet enterprise-grade security and governance expectations.

They also increasingly desire a way to capture immediate value using today’s imperfect data - while gradually maturing their data landscape over time through Microsoft data platforms and Azure AI.

Depending on your agent‑scaffolding requirements, plan a technology stack that can scale with your manufacturing environment. Microsoft provides a comprehensive set of tools designed to support both simple and highly complex factory use cases.

Microsoft 365 Copilot for everyday information work - fast retrieval of vendor manuals, summarization, documentation support, and knowledge searches.
Copilot Studio, ideal for extending M365 Copilot and quickly building lightweight prototypes.
Azure AI Foundry, the Agent Framework, GitHub Copilot, and related Azure services to build robust, production‑grade agent systems for advanced, custom scenarios.
Domain-specific Partner Solutions like OpsMate AI for manufacturing operations that provide pre-built manufacturing scaffolding and agents tailored to the high-value use cases described above.

OpsMate AI is built natively on Microsoft Azure AI and data platforms and is designed to snap into your existing Microsoft identity, security, and data investments.

For domain specific use cases, determine your Build vs. Buy strategy early. According to recent MIT research, buying has a twice higher success rate than building from scratch. If AI agents represent a core differentiator for your business, be ready to develop an internal AI product capability that can continuously build, refine, and maintain these systems. Otherwise, accelerate your time to value via Microsoft’s partner ecosystem and consider proven solutions already in market.

Recommended approach for sophisticated scenarios:
Use M365 Copilot and Copilot Studio for general productivity and simple workflows. For complex cross-system factory use cases requiring many of the scaffolding capabilities above, adopt OpsMate AI on top of Azure AI and Microsoft data platforms rather than attempting a purely custom build.

4. Build Data Foundations Early - And Treat Them as a Journey

You do not need a perfect data foundation to begin your AI agent journey. Manufacturers can—and should—start delivering value with AI today. However, it is equally important to recognize that long‑term scale and reliability depend on building the right data foundation over time.

Industry research shows that many organizations struggle with data readiness: inadequate data quality (57%), weak integration (54%), and limited governance (47%) often slow progress as AI initiatives mature3.

The fastest path to value is to meet end users where they are today:

Give engineers, technicians, and frontline teams immediate access to the same data, documents, and tribal knowledge they already rely on for problem solving - without waiting for perfect data or system integration.
Whether troubleshooting a machine, performing root cause analysis, or resolving quality issues, teams today lose valuable time manually stitching together information from multiple systems, spreadsheets, and documents - work that AI agents can streamline immediately.

As users interact with agentic capabilities, the system itself surfaces the highest-priority data gaps, revealing where focused improvements in structure, quality, or integration will unlock the greatest additional value. This creates a virtuous cycle:

Start by delivering value using existing data and knowledge.
Use real-world usage analytics, AI reasoning patterns, and frontline feedback to identify the most impactful data investments.
Incrementally enrich, govern, and unify your data foundation—making agents more intelligent, more consistent, and more autonomous over time.

Platforms such as OpsMate AI, paired with Microsoft data platforms, enable organizations to progress along this journey: delivering measurable value on day one while continuously strengthening the underlying data ecosystem.

Appendix