As AI agents multiply, IT becomes the new HR department

Agentic AI’s role in connected enterprises goes deeper than helper apps. AI agents are fast becoming the emerging power behind the microservices that form the fabric of enterprise systems. Plus, as these agents proliferate, information technology departments will become virtual “human resources” departments, acquiring, onboarding, and guiding AI-powered assistants in parallel with HR’s role in human capital management. 

These are the insights from a panel hosted by Deloitte at the recent Mobile World Congress, exploring the emerging role of AI agents within enterprises. Panelists said agentic architecture resembles the advent of microservices architectures, breaking down monolithic applications into flexible, independent, bite-sized morsels. 

Also: Crawl, then walk, before you run with AI agents, experts recommend

“Agentic AI is the next step in breaking apart and solving problems,” said Bryan Thompson, vice president for GreenLake product management at HPE. With agentic AI in particular, there are opportunities to “leverage these types of models and break them into almost like a microservice type of approach to tackle them — breaking them apart into specialized services.”

Agentic AI enables the stitching together of enterprise workflows, agreed Fred Devoir, global head of solution architecture for telco at Nvidia. “We take componentry and put it together into a RESTful architecture. Nvidia was able to optimize those with our microservices, and then bring together those microservices into blueprints to give a very quick time to value or time to first results.” 

Of course, agentic AI brings capabilities way beyond what traditional microservice architectures could ever produce. “Until now, we’ve never had a technology that could ideate, or execute independently,” said Abdi Goodarzi, head of gen AI products, innovations and new businesses for Deloitte. “Just think about that statement, and any other software package solution you’ve ever dealt with. None of them could independently execute any of it. That’s really the power of AI.”   

The agentic AI services take on many of the onerous tasks of humans — essentially, a parallel workforce forms, but onboarded and managed by IT instead of HR. “Human capital management and agentic AI capital management are the same thing, right?” Devoir said. “But the difference is instead of an HR for humans, now you have an IT department that’s acting as HR for all these agents.” The IT department also takes on the roles of “curating, guardrailing, training, and fine-tuning AI agents to do specific tasks and interact with human workflows. This is no small feat. There’s a lot of effort that goes into this. It is just like HR at a much, much deeper technical level.”

Also: AI agents may soon surpass people as primary application users

This also means sweeping changes across organizations as well. “Humans have emotions. Agents don’t have emotions,” said Goodarzi. “How do you incorporate the emotions that will be part of the execution of the work? When the work gets down in a different way, the culture has to be shifted, talent strategies have to be shifted, and how humans and machines work together have got to be changed.”

Getting to an agentic AI-powered enterprise has its challenges, however — especially when it comes to data, trustworthiness, and talent. Regarding data, “enterprises have been spending so much investment in getting their structured data under control,” Goodzari said. “Building ERP systems. Building systems of record. Systems of action.” All these applications or systems end up with separate data silos. 

Agentic AI may help address this, enabling deploying the agent where the data resides. “Instead of having to bring all your data to the AI, you’re taking the AI to the data,” said Devior. “When you make a service call, it actually asks all those data agents for a response — and collates that data into a model.”  

Also: How businesses are accelerating time to agentic AI value

Then there’s the issue of the trustworthiness of agents. “You need to think about whether you’re actually dealing with the right data,” Goodzari pointed out. “Am I dealing with the right results? All the other previous technologies were designed around transactional activity. Agentic AI is designed around probabilistic technologies. So you get the best probable answer because you have trained agents with a lot of knowledge on how to digest the data and make a decision and make a recommendation.” 

Then, the issue of trust kicks in: “Can I trust this agent? Can this data be right? Am I dealing with the right data? That has to be solved as well.” 

Overall, “these are new concepts for enterprises,” Goodzari emphasized. “That’s why it has slowed things down in terms of adoption. But the capabilities are real. The technology is advanced enough to leverage within enterprise production systems. And I do believe this is the year that it’s going to take off.”