As manufacturers continue to adopt AI agents, one vendor has launched a platform that, it says, can unify physical systems and agentic AI through one package that also has governance features.  

This summer, AI and technology services provider Cognizant launched its physical AI platform-as-a-service built on its Intelligence Spine, a platform that aims to unify AI agents that companies use in operations. 

See also: Industrial IoT starts with the physical layer: Rethinking cables, conductors and connectivity 

In recent show interviews, Cognizant described its platform as a sovereign institutional PaaS for physical AI, combining physical factory elements—such as sensors, cameras, robots, and AI twins—and the agentic layer that reasons and acts, connecting physical AI systems with agentic AI into a unified system that is designed for enterprise ownership and governance. 

"Think of it like a 'central nervous system,' which pervades across multiple physical assets, and is able to sense, reason, action, and learn as one," said Vijay Narayan, the global head of physical AI and business unit head of manufacturing, logistics, energy and utilities at Cognizant. 

"It's kind of like an abstract layer, which is able to make these physical assets become more intelligent as well," he said in an interview at Cognizant’s Manufacturing Innovation Center at the Digital Manufacturing and Cybersecurity Institute, known as MxD, in Chicago.   

According to Cognizant, the platform can be deployed across eight industries, including manufacturing; oil and gas; utilities; logistics; transportation; aerospace and defense; health care and life sciences; and consumer, retail and consumer packaged goods. 

‘Intelligence in pockets’ 

In manufacturing settings, physical AI systems collect data on the physical environment but do not share that intelligence with each other, Narayan explained, calling this "intelligence in pockets," making it difficult for manufacturers to scale their AI.  

For example, a manufacturer might use different physical AI agents for different functions in a factory setting, all made by different companies.  

The platform was developed, he explained, to solve that problem.

"Unless you actually figure out a way for all of these [agents] to think as one, it's difficult to scale it in an enterprisewide way across plants, across assembly lines, across stores," he said. 

Podcast: With outdated and isolated OT, your AI strategy may be built on a blind spot 

"Everybody is collecting information about the same physical environment, but they don't have a full-fledged view of what is happening. That's a problem which we sought out to solve, and that's what was the genesis of what we created." 

Case studies: Automobiles and semiconductor humanoids 

One example of systems the spine could unify are different physical agents made by different OEMs that operate within the same factory, such as computer vision systems, automated guided vehicles and humanoids.  

Although it has not been deployed for widespread use yet, it is already being implemented with a handful of manufacturers, utilities and automobile companies. Cognizant is in advanced discussions with others, according to Samih Fadli, the company’s chief AI officer and CTO of physical AI autonomy systems. 

For example, one automobile manufacturing company that is not using the spine has fragmented AI across assembly lines, all built by different OEMs.  

Along one assembly line, windshields are assembled by physical AI agents, but the separate inspection line agent does not have awareness of what happened in the assembly line because the agents are built by different companies, Fadli explained.

"What you're going to see here is we're going to move from manual manufacturing to a fully robotic AMR system. Within that system, there are multiple OEMs with different OEM sensors," Fadli said.

The spine can then create "a fully operational, coherent intelligence system, where now all those fragmented OEMs operate as one AI," that is connected by the spine, he added. 

See also: Robotics isn’t having its ChatGPT moment just yet 

There are also separate cases where the platform is deployed in semiconductor manufacturing factory settings with humanoids, which Fadli said can successfully work alongside human workers on factory floors.  

The humanoids can perform safety measures and maintain spatial awareness of humans while monitoring "thousands of sensors across multiple assemblies," he said. The humanoids can advise human workers and intervene in the assembly line at the same time, he said. 

Fadli added: "Once you have all this intelligence [in] the same environment, it could actually create correlations which nobody else thought of." 

Governance features  

The platform has features that work together to unify different AI platforms while allowing enterprises to maintain security, Narayan said.  

For example, the spine has institutional memory to store information from different agents. It also has a context-aware engine, meaning that any action or reasoning doesn't happen outside of the context of the enterprise and creates unified reasoning across platforms.  

These features and others work together to allow enterprises to apply cybersecurity more intelligently, Narayan said. This is important, he added, as manufacturers remain the top targets for ransomware attacks. 

Manufacturers were hit the hardest by ransomware attacks last year, with a 58% year-over-year increase in victims, according to one study.   

Although cyberattacks against manufacturing operations are hardly new, AI is rapidly changing how quickly attackers can identify targets, find vulnerabilities, and exploit them. 

See also: Attack chain glue: How one form of AI is hypercharging cyberattacks on manufacturing 

The spine also allows users to apply the same constitution of ethics across all fragmented agents. 

"We didn't replace cybersecurity. We did something that is going to unlock the potential of cybersecurity," Fadli said.  

The spine has a governance model, which means a user can load up an enterprise's constitution of ethics, then all physical AI models will work with that, including where external intervention is needed and what actions are prohibited.  

"Anybody can create one single policy and it can get propagated across the entire fragmented system," Fadli told Automation World.  

It also maintains audit trails of every decision in a ledger, allowing users to trace back through agent decisions if an incident takes place.  

"Every time a system is reasoning in our infrastructure, a ledger has been built in a blockchain, and if a model tries to do something against the constitution [of the enterprise], the ledger will be able to block it because there is a blockchain integrated there," Fadli said.