The Work That Goes Behind AI Agents

AI agents need training, too

getty

AI agents – or agentic AI if you want to keep it formal – have been the at the top of the tech hype wave the past year. Are they finally beginning to deliver as promised?

It’s the next logical stage of evolution from genAI, says Adam Famularo, chief executive officer at WorkFusion and former CEO of erwin (now part of Quest). “While genAI excels at generating content and responses, it can’t take action,” he explained. With hallucinations and lack of real-time performance, genAI alone doesn’t cut it, he added.

At the same time, creating and deploying AI agents aren’t exactly overnight tasks. AI agents still rely on a genAI foundation, as well as “machine learning, statistical analysis, robotic process automation, intelligent document processing, and more,” he pointed out.

Also, an agentic AI architecture requires “proper controls and guardrails,” Famularo cautioned. “AI agents are automations that can execute complex, multistep workflows, across digital environments and apply reasoning. These agents don’t just respond to prompts; they act with intent, execute multi-step tasks, learn from outcomes, and make decisions in context. They leverage other forms of AI, automation approaches, and people – the human in the loop.”

Indeed, agentic AI’s “implications stretch far beyond automation or productivity gains,” a report out of Harvard Business Review Analytic Services, underwritten by Wipro, confirms. Along with the technology integration required, “preparing an organization and its people to adopt agentic AI can be challenging as leadership may face disinterest, skepticism, or resistance. Critical thinking is an important human skill that will always be needed, regardless of how automated the world becomes.”

The report cited estimates out of Gartner, which predicted that by 2028, 33% of all enterprise software applications will include agentic AI, up from less than 1% in 2024.

“The shift to agentic AI requires a more sophisticated infrastructure and mindset,” agreed Nitesh Bansal, CEO of R Systems. At this time, only cutting-edge innovators are leading the way. “Many organizations lack AI literacy, specialized skills, and market ready tools and infrastructure.”

Such preparation is essential to the success of agentic AI, since it “can’t decide what services you need to provide or what’s needed in the market.” the HBR report noted. ”Critical thinking is an important human skill that will always be needed, regardless of how automated the world becomes.” Agentic AI will serve to augment human skills.

Another development is also shaping agentic AI’s value to organizations. This is taking the form of an interconnected “internet of agents” — in which agents collaborate, even across organizational boundaries. “This is one of the most exciting possibilities for agentic AI down the line,” said Famularo. Agents need to be able to collaborate with each other, and be built to be transparent and auditable, able to plug into collaborative ecosystems when the time comes.

“In the financial world, especially in compliance, there’s growing interest around secure, cross-organizational intelligence sharing – under strict privacy and regulatory controls,” said Famularo. “Imagine AI agents at different institutions collaborating to identify emerging fraud patterns or sanction risks using anonymized data.”

“We’re already seeing the early signs: increased openness to data-sharing frameworks, evolving regulatory support, and growing trust in agentic systems,” said Famularo.

To successfully adopt agentic AI, businesses should consider:

Get comfortable with AI. “A key thing is just getting comfortable with AI itself,” said Famularo. “The ChatGPTs and Gemini’s of the world are available to use – use them and try to understand how they can help your day-to-day at work but also at home. But keep in mind that those represent only a fraction of the broader AI offerings.”

Keep employee skills refreshed and current. “Equip teams with the necessary skills through training programs focused on AI technologies and customer service applications,” said Bansal. “This will help bridge the knowledge gap and foster a culture of innovation.”

Build use cases. “Organizations need to look for opportunities to embed AI across the organization, for big and small use cases that can achieve value and success early on,” said Famularo. “One approach is AI Agents with a pre-defined purpose, which avoids overly long development cycles – a way to start small if you will.”

Keep humans in the loop. “Adapt to unknown unknowns – those anomalies no one’s seen before,” said Famularo. “For people to make a decision they often need to look at a lot of data. And if the AI agent is confused on the decision to make, being able to provide the data available back to people is still a big win on its own, saving minutes or hours, depending on the use case.”

Treat AI agents as teammates. Organizations “may be entering a new era where human teams and AI agents operate in tandem, which will demand fresh approaches to processes, management, governance, and workforce planning,” the HBR report agreed.

“Create environments where AI agents can operate like true digital teammates,” said Famularo. “Moderna announced that it was merging tech and HR. We’ve been envisioning this hybrid workforce for years. You will start to see more and more of these fusion teams of AI agents and humans. It makes perfect sense for IT and HR to work more closely together in filling job roles.”

Open the agentic AI development process to a broad base. “A collaborative approach can help reduce bias in AI and make the output of its analyses or decision making more useful to organizations and their customers,” according to the HBR report.

Rethink the processes AI agents are touching. “AI agents aren’t just automating tasks; they’re making decisions, learning from outcome, and working alongside your people,” said Famularo. “That means the entire process has to support autonomy, collaboration, and real-time adaptability.”