AI Agents

When the assembly line changed manufacturing in the early 1900s, it wasn’t just about faster production. It reshaped how factories operated, redefined jobs, and made quality and scale more consistent. It was a rethink of how work gets done, not just a new tool.

Today, manufacturing is approaching a similar moment. After years of steady gains from enterprise resource planning (ERP) solutions, manufacturing execution systems (MES), and automation, something new is entering the mix: AI agents. AI agents aren’t just analytics tools. They’re software teammates that can sense what’s going on, decide what to do next, and take action within defined boundaries. Instead of reviewing yesterday’s data, they help teams respond to today’s challenges in real time.

A signal from the field

More manufacturers now see AI as a strategic priority. In metals, adoption is climbing. According to the Crowe “2025 Technology in Metals Survey Report,” 40% of survey respondents reported that their organizations use out-of-the-box AI solutions but only 8% noted strategic integration of AI across the business.

While more than half of survey respondents (56%) planned to use AI in data analysis, only 32% had put that plan into action. Survey respondents reported using AI in sales and marketing, but for engineering, production, and maintenance, AI use was rare. This gap between wanting to use AI and executing its use points to technical and cultural complexities as well as a need to upskill workforces before scaling.

Why the current approach is changing 

Even the best-run plants still rely on manual workarounds. A planner updates a dozen spreadsheets. A supervisor walks the floor to spot issues. Someone chases late purchase orders or unconfirmed shipments. These tasks are important but repetitive. They drain time and are easy to get wrong.

In manufacturing, back offices often modernize faster than customer-facing or coordination systems. That mismatch creates friction right when teams need clarity and speed. The result: scattered tools, inconsistent handoffs, and missed insights.

What AI agents can do 

AI agents can be considered digital teammates that can: 

• Listen by monitoring sensors, logs, orders, emails, and MES events
• Decide by using rules and patterns to pick the next best step, with human review as needed
• Act by triggering workflows such as opening a quality ticket, sending an expedite request, or adjusting a quote

The key difference between AI agents and a chatbot or an application like OpenAI's ChatGPT is that agents don’t just answer questions. They complete tasks and log what they do. On the Microsoft platform, agents can use data from the Microsoft Fabric or Microsoft Dataverse™ solutions to stay grounded in context. They can also act through Microsoft Dynamics 365™, Microsoft Power Automate™, and Microsoft 365™, with guardrails from an organization’s identity and policy settings. 

Where to start: Quick wins for agents 

Many planners spend their days putting out fires, chasing updates, checking exceptions, and digging through spreadsheets. AI agents are built to manage that kind of repetitive work. Here’s where they can deliver early impact:

• Shop floor: Quality and root cause 
  An AI agent can monitor line data and quality events. If scrap spikes, it can check the last changeover, operator notes, and maintenance logs. It can also suggest a likely cause, create a corrective task in the quality management system, attach run data, assign the task, and schedule a quick check. A supervisor reviews and approves, and the issue is handled with a clear trail.
• Supply chain: Exception handling
  If an advance ship notice is late, the AI agent can check current work in progress, due dates, and buffer stock. It can flag risk, draft an expedited request, and propose short-term schedule changes. The team reviews and adjusts, and the agent can take care of the follow-through. 
• Commercial back office: Quote-to-cash 
  Sales teams often input quotes in ERP solutions while managing relationships in customer relationship management (CRM) platforms. AI agents can help tie those worlds together by assembling quote packets, prompting approvals, updating CRM notes, and suggesting next steps. They can also remind account managers about renewals or rising customer activity, which can help optimize integrations between the systems or even eliminate repetitive middleware. 
• Finance: Handling documents at scale 
  AI agents can process invoices, validate prices, reconcile freight charges, and suggest adjustments. They can also extract key data, compare it to ERP records, and route drafts for review. The reviewer sees the full context and can approve or revise. In fact, Microsoft has started releasing these agents natively in Dynamics 365 Finance and Operations, its ERP tool, including an agent designed to help automate month-end account reconciliation.

From manual triage to automatic updates 

Consider this scenario: A North American steel producer implements Dynamics 365 to modernize commercial and operational processes. Its next step is to solve the daily scramble caused by late shipments, so it implements an AI agent that starts tracking shipment updates and open orders, flagging risks, and drafting expedite actions. Planners shift from reactive inbox triage to reviewing a ready-made playbook each morning. AI agent tools like this one demonstrate how new technology can optimize existing and current software.

Building on a foundation

AI agents are only as strong as the data and systems they’re built on. The Microsoft cloud connects Fabric for data, Dynamics 365 for process, and Microsoft 365 for collaboration. Agent capabilities available in Microsoft Azure™ help manage design, security, and monitoring. This connectivity creates an environment in which agents can work across a digital thread, not just in silos.

Crowe builds on that foundation. For example, with Crowe Metals Accelerator, we’ve integrated Dynamics 365 Finance and Operations for metals and built in industry-specific processes. This approach means AI agents can plug into proven workflows from day one, without risky workarounds or custom code.

A realistic road map: From copilot to agent 

Most companies benefit from taking a phased approach. They begin by using copilots to assist with tasks such as taking meeting notes or drafting documents. Next, they move to an advisory stage during which agents can propose actions for human approval. Finally, they progress to allowing agents to act by carrying out low-risk steps automatically, supported by guardrails.

This gradual process helps build trust while demonstrating real results. Teams remain in control, governance scales effectively as adoption grows, and everyone can see the benefits of using AI agents early on.

Guardrails that matter

To keep AI agents both safe and effective, it’s essential to design with governance in mind. Humans should always stay in the loop regarding high-stakes decisions, and teams should use each Microsoft platform's identity and access controls to manage permissions.

Clear data lineage and audit trails help maintain transparency, and aligning AI use with company governance policies helps with accountability and consistency as use expands.

What to expect in the first 60 to 90 days

The first few months of implementing and using AI agents are about learning and measurable progress. To help focus the process, organizations should identify one process area and two metrics that truly matter.

For example, on the shop floor, companies might aim for faster root-cause analysis and automated corrective actions. In the supply chain, success could mean fewer expediting requests and earlier risk warnings. Commercial teams might focus on higher quote follow-through and on-time renewals, while finance teams could measure fewer touchpoints in invoice exceptions and improved audit trails.

By tracking results and sharing early wins internally, teams can build momentum and enthusiasm for broader adoption.

Implementing the first agent

When implementing an AI agent, it’s a good idea to start with a high-volume, repeatable pain point – something that regularly consumes time and resources. Then it’s important to define what the AI agent should do, establish clear boundaries, and connect the agent securely to the organization’s data and systems. Beginning in a test-and-review mode can help build confidence, and then, where it makes sense, automatic execution can be enabled.

As users experience the benefits of using AI agents firsthand, adoption tends to spread organically, creating a ripple effect that fosters openness to new ways of working and a culture ready for AI-driven transformation.

Crowe can help you move forward

Crowe specialists can guide your team through the first setup by mapping out use cases, configuring the tools, and enabling the agent to deliver real work safely. Your team stays in charge. The AI agent simply makes things easier.

Microsoft, Azure, Dataverse, Dynamics 365, Microsoft 365, and Power Automate are trademarks of the Microsoft group of companies.