Cognizant Thursday introduced its new Neuro AI Multi-Agent Accelerator, a no-code developer framework to rapidly tie AI agents from multiple disparate applications to increase productivity and scale in multi-agent systems.
Agents are all the rage, but clients want to connect them and make them interoperable at scale, said Babak Hodjat, CTO of AI for the Teaneck, N.J.-based solution provider.
“Salesforce has Agentforce, ServiceNow has its agent, SAP and everybody else is coming with their agents,” Hodjat (pictured) told CRN. “Agents are basically large language models equipped with some tools. And if Salesforce is giving you an agent, they’re equipping it with Salesforce-related data and tools and so forth, and it’s really good at that. And if ServiceNow is doing that, it’s good at what ServiceNow does.”
[Related: Cognizant CEO On ‘Becoming The First Partner’ In ServiceNow’s ‘Ground-Breaking Workflow Data Fabric’]
The tools those companies let businesses customize their AI agents for their instances of Salesforce or ServiceNow or SAP, leading to an ever-growing collection of agents, Hodjat said.
Advertisement
“Some of them are [homegrown] or built inside the client for their own specific usage, as was the case for us,” he said. “In our case, we were identifying our intranet for our own employees to use, and adding agents, and we started thinking, ‘Oh, we need to have the HR agent and the finance agent and the other agents that we have?’ It doesn’t make sense for employees to decide which agent to talk to. These agents need to be talking to each other.”
Cognizant eventually ended up with Agentforce for its Salesforce implementation, and ServiceNow agents for its ticketing system, with employees required to talk to multiple agents and repeating themselves, Hodjat said.
“It’s just so obvious that these AI agents need to be connected,” he said. “There are disparate but related tasks you need to do, and now these agents can actually collaborate and make them happen, which means that a lot of the silos that exist in our human-only-driven organizations can be broken now that we have agents that can semi-autonomously coordinate and work on your behalf.”
Services companies like Cognizant are tasked to help customers build specific multi-agent systems to augment their human driven processes or their applications, Hodjat said.
“That’s what we do,” he said. “And in the process of doing that, we realized that off-the-shelf tools for doing this sort of thing are just not there. They’re not scalable, and so we had to write our own tooling. So that’s why we’re calling this an accelerator. We’re not a product company. We’re not out there trying to license IP, but we’re seeing a barrage of requirements from our clients and need to meet it in a scalable way. Our accelerator allows these agents to coordinate with one another, using natural language primarily, but also passing along secrets, by which I mean authorization, like this user can get access to this thing, or cannot get access to this thing.”
Cognizant’s Neuro AI Multi-Agent Accelerator brings new agents into a multi-agent system without having to re-engineer the existing multi agent system, Hodjat said.
“That’s one of the key differentiators of the approach,” he said. “We allow for various different departments in an organization to own various different parts of this multiagent system, which totally makes sense, because if you’re in finance, you don’t want someone else to come in and redo your processes and so forth just so you can plug in HR to let HR and finance talk to each other. These have to be what we call ‘encapsulated’ and semi-independent. But plugging in a new agent should just expand what you can do. And this is some of what this technology allows.”
Businesses could manage multiple agents on their own, but for now are still focused on single agents or agents that are related to a single set of tasks, Hodjat said.
“We’ve been on the forefront of this, and so we’ve seen the need for interoperability, and so we had to build this,” he said. “We just had to do it because nobody else had done it. The other thing is that that inter-agent communication and coordination that is needed to make this scalable is tricky. It’s not as simple. I’m not saying it’s rocket science, but it’s kind of tricky.”
Cognizant Thursday also unveiled its Multi-Agent Services Suite to help businesses create and deploy multi-agent systems using a standardized, proven approach.
Cognizant’s services help organizations to redesign their business processes, develop and deploy intelligent agent systems, and efficiently manage them in production.
Cognizant knows how to build agent networks, Hodjat said.
“You need to understand your process flows, where an agent makes sense and where it doesn’t make sense, and how to connect them,” he said. “Also, how do you make them responsible? Responsibility is important because an agent has some level of autonomy. That just goes with the territory. You defer to it to make certain decisions by itself. So it has to be responsible.”
For these sorts of things, Cognizant has a set process and steps, Hodjat said.
“Trained Cognizant associates come in and help build this for you, and they will use the Accelerator along with the tools of your choice as well,” he said. “We’re announcing the two at the same time because we believe they go together.”