Stories of AI adoption: Wolfspeed all-in with 22 agents across key company teams

Editor's note: Is AI replacing manufacturing workers, assisting them or something in between? This is the first in an occasional series of company-specific profiles that Smart Industry hopes will help bring answers to this question. We begin with this Part 1 of 2 on Wolfspeed, which has deployed agentic AI extenstively. 

As companies across manufacturing verticals continue to adopt AI agents with varying degrees of success, semiconductor manufacturer Wolfspeed found some by efficiently deploying AI across numerous company teams.  

Durham, North Carolina-based Wolfspeed is a developer and manufacturer of wide-bandgap semiconductors and specializes in silicon carbide materials and devices for applications in transportation, power supplies, power inverters, and wireless systems.

Wolfspeed is deploying agents during a time when AI is being adopted by manufacturers at a large scale, as physical (the kind that helps operate robots, vehicles, and machinery in real-time) and agentic AI become increasingly useful, and perhaps necessary, for factory operations.

See also: Podcast: Why data collection is worth the time, effort and expense 

But unlike many other manufacturers that experience high failure rates when integrating these virtual agents, Wolfspeed has about 22 of them deployed widely across teams such as finance, HR, sales, manufacturing and more. 

Data organization came first at Wolfspeed 

Company leaders credit their success in part to the company’s data repository—an ecosystem from where AI agents receive data—that holds information from manufacturing systems, operational documentation, troubleshooting logs, and engineering discussions within one hub. 

That ecosystem is a combination of structured data—such information from tool reports and factory floor settings—and unstructured data—such as institutional knowledge, information from company presentations and emails, among more. 

Wolfspeed uses architectural foundations for these agents provided by AI agent software service Snowflake Intelligence, which lent the architecture to unify Wolfspeed’s structured and unstructured data. 

Both types of data—according to leaders at Wolfspeed that Smart Industry interviewed—were integral in developing useful AI agents that employees across teams could use. However, understanding how to process and use company data to inform AI agents is a continuous process that requires trial and error. 

“A lot of data extraction for AI [involves] making sure that you have not only the right data, but you have the right governance across that data and the right architecture,” said Priya Almelkar, who is chief information officer at Wolfspeed. 

See also: AI is exposing a massive data problem in the supply chain 

Almelkar said there are “three pillars” to successfully developing and deploying AI agents, especially in her experience in a manufacturing setting.  

“Three things that I've come across in my learning is to ensure that you have the right data, the right governance and the right architecture. Because if you don't have either one of those, the use cases will not be successful,” she said. 

Three pillars of data plus an enterprise AI hub  

Those three pillars—data, governance, architecture—are part of what helped Wolfspeed be successful in AI agent deployment, compared to other places that have had higher rates of failure, she said.  

Almelkar said those three pillars are key to having a 100% success rate in deploying AI agents, a percentage that is a requirement before the company deploys any agent. 

She said that this enterprise AI hub allows departments to be collected quickly and allows for "continuous learning" for agents, which also contributes to successful deployment.

She gave an example of an “AI solution” in tool troubleshooting that involves multiple agents collecting both structured and unstructured data for employees to use at the start of each shift.  

With an agent used in tool troubleshooting, it first collects data from the factory-deployed computerized maintenance management system—which centralizes maintenance information and tracks operations—and other web applications that collect tool state history. 

See also: Physical AI in manufacturing: Assistant, replacement or something in between? 

On top of that, there is another agent deployed on top of Snowflake that is able to look up both structured and unstructured data in the repository.  

The “solution” is a combination of looking at the tool state and looking at data from web applications run within the prior 24 hours.  

The solution then creates a tool state report to present to the team including tool state, anomalies in tools and overall conditions.  

Agent supplies reports from multiple places 

Wolfspeed’s AI agents also take all comments and reports from the previous shift's system and operator, summarizes them into actions, and then sends them to department heads and is used in meetings at the start of a new shift. 

Therefore, as the team is looking at the tool state information for that particular day and how the tools are going to run in a factory setting, the AI agent is able to provide them with necessary information both from factory floor data and the repository of tool state history and institutional knowledge.  

“The end decision is still being made by humans, so it's still not an autonomous solution, but it is saving many, many hours for the team,” Almelkar said. 

See also: The living document strategy: Why your CMMS is more than just a digital file cabinet 

Although these agents are functioning properly now, the process at Wolfspeed of determining how to integrate the required data from the enterprise hub to department-specific agents took trial and error.  

Unni Velayudhan, Wolfspeed’s senior director of data and automation, said he views the data that is drawn from each department as a “data pond, not a data lake.”  

Through trial and error, the data team developed a system for getting data from teams and applying it to the agents, he said. 

First, there is an enterprise data hub. The data then passes through an AI agent that classifies it based on department. Data is then organized in each department, which has an AI “business champion” who is tasked with bringing in relevant data and serving as a filter when integrating data, so it isn’t “dumped” into agents, Velayudhan said.

It is then deployed to agents for use.  

“You don’t want to mix department data,” he said, citing concerns of agent effectiveness and security breaches. 

Three possible strategy outcomes of agents  

He explained that when developing these agents, there are three strategy outcomes: “First, the solution should help in improving our efficiency, because when you improve the efficiency, obviously there will be cost benefit associated with that. So, improving the overall efficiency of every department in the company is one of the objective.”  

See also: How OEMs are reimagining vehicle inspection—and AI’s role in this transformation 

“Second, how can it help us with the new revenue generation? For example, how can you use market signals and convert that into some sort of insights to our sales and marketing team or to our business units team? The goal here is that using AI technology as a value stream for new revenue generation market opportunities,” Velayudhan added, noting that Wolfspeed prioritizing understanding signals and insights from both existing and potential customers from the internet and are fed into internal departments.  

Finally, "how can we use this technology to accelerate our innovation? How can new technology introduction, or new production introduction, can happen using the technology?”  

At Wolfspeed, an ecosystem with continuous data flow 

Both Almelkar and Velayudhan emphasized that it was important to take data from specific teams and take data continuously, not just once. This was part of their trial-and-error process, in which they discovered that continuous learning for the agents is necessary for successful deployment. 

Velayudhan said that the overall goal is to be able to integrate this technology effectively into every department. The enterprise AI hub, he said, is not a use-by-use case and should be viewed holistically. 

See also: Podcast: Stories of real AI adoption in manufacturing maintenance 

But that ecosystem, and the holistic view of its role, is what makes Wolfspeed successful at AI agent deployment in comparison to other manufacturers, he said. 

“If you build your enterprise AI hub properly with all the relevant data, you can create any number of agents,” he said. “So, building an AI knowledge hub and having a constant flow of data into the AI hub is very important for any organization to be successful. That's why we were able to manage to roll out these agents at a short span.”