Industrial Agentic AI Success Centers on the Evolving Role of People, Not Their Replacement

Industrial giants Siemens, Schneider Electric, and Rockwell Automation have ambitious plans to industrialize Generative Artificial Intelligence (Gen AI) innovation with the introduction of new modalities that extend beyond text and speech to include product, process, and factory data. The prevalence of such domain-specific initiatives alongside major investments in Artificial Intelligence (AI) by Amazon Web Services (AWS), Microsoft, and NVIDIA underpin the need for a framework to guide technology suppliers, implementers, and end users in their use of both Gen AI and the agentic workflows on the back end.

Gen AI is most often associated with copilots, which are the taxonomical front end with which users engage. AI agents, or Agentic AI, are different; these are the back-end pieces of logic that operate autonomously to perform specific tasks.

For example, consider manufacturing a refrigerator. Mechanical and electrical engineers may mockup the initial Computer-Aided Design (CAD) designs and perform Computer-Aided Manufacturing (CAM) programming for the machining or production stage, while a different cohort of engineers is responsible for setting up and optimizing the production process once underway. In this scenario, a company may have an agent for topology optimization to improve the surface finish of the appliance, an agent to help with material selection based on supplier availability, and an agent to help with structural and thermodynamic analyses.

An engineer would interact with a copilot to engage or direct agentic workflows, though the agents may also interact autonomously with one another on the back end to optimize engineering workflows, improve quality, and speed up New Product Introduction (NPI).

Gen AI and agentic workflow implementations must be human-centric to minimize errors and improve time to market for closed-loop automation.

At the first, most basic level (Level 0), where there is no automation, a human is the loop; for example, a bread maker manually mixing ingredients and putting bread into an oven.

At the next level (Level 1)—or the first level of agentic automation—a human would be in the loop (human-in-the-loop) to validate each step of the process. Here, a human is an integral part of automation checks and balances by forcing an approval step for automation to occur.

Human-on-the-loop is the next stage (Level 2), where an adaptable, automation feedback system is set up and the role of employees is to monitor—rather than validate—activities, stepping in only when correction is required.

The final stage (Level 3) is to have humans fully out-of-the-loop, meaning the automated system takes real-time inputs and feeds them back into the process in a closed-loop manner to enable adaptive adjustments/actions on the fly. Here, the role of humans is to intervene only if something unexpected occurs.

Embracing these four steps, with an emphasis on the transition from human-in-the-loop to human-on-the-loop is essential to making the most of agentic initiatives that improve the efficacy of Gen AI, while enabling higher degrees of autonomy with the use of agents. Suppliers, implementers, and end users all serve to benefit from this framework.