Agentic AI Is Coming to Manufacturers but Not in 2025

Agentic Artificial Intelligence (AI) is a step beyond Generative Artificial Intelligence (Gen AI) and copilots, with the differentiating feature being the ability to autonomously act and change processes. Much like general AI and Gen AI, agentic AI must be trained on viable datasets and integrated within the manufacturing suite of applications currently used by an organization to provide beneficial actions and Return on Investment (ROI). Before agentic AI can permeate the manufacturing sector, building AI agents must be democratized to provide ease of customized training, setup, and validation of answers. In doing so, agentic AI will be able to support real-time fixes to production issues such as bottleneck prevention without worker intervention.

Leading the charge in building AI agents is NVIDIA, which showcased its new NVIDIA AI Blueprints at CES 2025. NVIDIA AI Blueprints, developed with the aid of AI software providers CrewAI, Daily, LangChain, LlamaIndex, and Weights & Biases, are a set of building blocks to develop agentic AI models that can be integrated into existing software on the NVIDIA AI Enterprise platform. Microsoft has released agentic AI models with Semantic Kernal for ready to use AI agents, and AutoGen for manufacturers that require multi-agent solutions. In February 2025, Salesforce will release Agentforce 2.0 that provides enhanced reasoning, integration, and customization for all industries; however, it has not taken steps toward deploying agentic AI outside of the Salesforce environment.

With the hype of new technology still strongly in the grasp of Gen AI, it is imperative to understand the fundamental changes between Gen AI and agentic AI. While often seen as an extension of Gen AI, the core difference comes from decision-making and autonomous execution. Gen AI is a powerful tool to aid manufacturers; however, it is stifled by the inability to enact a resolution allocated to fix a problem. Generative AI requires human intervention to fulfill the action produced by Gen AI insight; however, agentic AI can reason, decide, and then execute a sequence of events in a human-out-of-the-loop system.

With the core advancement of agentic AI being the autonomous execution of operations, vendors that are looking to deploy agentic AI will be focusing on the predominate use cases of:

• Data Transfer and Storage Location: Agentic AI can be used to decide in real time whether data produced on the factory floor need to be stored on the edge for quick access and lower latency, in the cloud in the form of data lakes, or in historians for future use.
• Initiate the Analysis of a Machine: Gen AI can instruct a frontline worker about an upcoming machine error and suggest preventative actions for an employee to undertake. Agentic AI will take this a step further and autonomously conduct machine analysis when data suggest imperfection, and then resolve the concern by implementing changes such as correcting bugs in machine code.
• Autonomous Bottleneck Prevention: With multi-agent solutions, agentic AI can alter operations on the production line to solve bottleneck issues such as restricting a belt feeder, increasing the output speed of lagging machines, or halting all production in extreme circumstances.

The development and talk around agentic AI is nearly identical to how software providers were speaking about Gen AI 18 months ago. Software providers are touting the autonomous capabilities of agentic AI and how it will be a game changer for manufacturers across the board with use cases that redefine current operations. Similar to Gen AI, many new agentic AI startups have formed, including but not limited to, Agency, Adept AI, Beam AI, and Moveworks. Following the trajectory of Gen AI, it would not come as a surprise if more software and AI providers start releasing AI agents and showcasing the new technology at trade shows such as Hannover Messe and IMTS in 2025. Although agentic AI will steal the headlines in press releases and trade shows, the implementation onto the factory floor will not happen in 2025 in any meaningful fashion. Vendors still need to flesh out solutions, democratize the deployment of agentic AI for connection to disparate manufacturing software, and run pilot tests with willing manufacturers.

ABI Research recently released its annual survey of manufacturers in the United States and found that the most pressing concerns being addressed by U.S. manufacturers were improving network infrastructure and security. Improving a company’s understanding of AI applications ranked 7 out of 9. The low ranking for AI-based technology primarily comes from two avenues: the dire need to protect against digital threats and a lack of prerequisite technologies that aids in deploying agentic AI. Before deploying agentic AI, U.S. manufacturers across all industries need to resolve the core issue of network resiliency, followed by taking incremental steps toward Industry 4.0. First and foremost, U.S. manufacturers need to explore options such as deploying more Internet of Things (IoT) sensors, edge computing, Software-as-a-Service (SaaS) transitions, digital twin utilization, and forming a connected digital thread before agentic AI becomes a viable new technology adoption. Once digitization of the factory is at a sufficient level, U.S. manufacturers can look toward agentic AI for specific use case adoptions, with the caveat of exploring what large manufacturers in early adopting industries such as automotive and aerospace have already done.