Generative AI has become a regular workplace feature, with employees using prompts to do everything from summarizing data to writing press releases, but most people are only scratching the surface of what’s possible.
Increasingly, people are using artificial intelligence to handle complex tasks semi-autonomously, says Harvard Business School Professor Tsedal Neeley and Expedia Group's Ritcha Ranjan. So-called agentic AI are autonomous AI systems that plan, reason, and act to complete tasks. The systems can carry out entire workflows with minimal human oversight.
I can’t see a future without every individual using AI to dramatically improve their work, their relationships, and their collaborations.
Most businesses surveyed by McKinsey recently have applied AI to at least one function, and 39 percent have begun to experiment with AI agents. More organizations are likely to follow, argue Neeley and Ranjan, who's senior vice president of product at Expedia and a tech expert in AI and product management. They offer a vision for the potential of agentic AI in the technical note, “Generative and Agentic AI as Strategic Partners for Leaders,” released in November.
“I can’t see a future without every individual using AI to dramatically improve their work, their relationships, and their collaborations,” says Neeley, the Naylor Fitzhugh Professor of Business Administration and chair of the MBA program. “Individuals and companies will be able to do so much more than they did before.”
“People talking about using AI to write an email or draft a document is just the starting point—what we’re doing is taking it to a higher order,” Ranjan says. “It’s the type of thinking that can unlock a 10x improvement in productivity.”
Agentic AI can even monitor and respond to external developments, using deep research capabilities to create ongoing summaries of information on collaborators or competitors—acting as a strategic press secretary.
It could really help leaders stay ahead of emerging developments and get the insights to act.
“The instruction for the agent could be to scan the internet every morning and give me everything written about a company or product by the time I start my day,” Neeley says. Far beyond a simple Google News alert, the agent could synthesize and interpret the information, translating it into action items related to your organization’s strategic goals. “It could really help leaders stay ahead of emerging developments and get the insights to act.”
“That’s a hard thing to do mentally,” says Ranjan. “Even if you sat down and started categorizing things in Excel, it would take forever.”
An AI agent could also evaluate meeting transcripts to determine what was actually discussed, rather than what was on the agenda. “Maybe my calendar looks like I was spending time on subject X, but we actually talked about subjects 1, 2, and 3,” says Ranjan.
Beyond categorizing the data, an agent could find patterns and compare them with organizational priorities to recommend how leaders could use their time better. “It effectively becomes a chief of staff, saying ‘here’s where you have been focused, and here’s how that compares with the highest priorities you’ve set,” Ranjan says.
The agent could then create a working profile of that individual, enabling the user to seek feedback ahead of an important presentation, for example. It could also provide tips on how to respond to concerns the person is likely to raise.
“One of the hardest things, especially for a junior employee, is thinking through every angle or question you are going to be asked,” Ranjan notes. AI could surface those likely questions and provide key insights to help them prepare more effectively.
Plan. What data does the underlying model need to perform tasks in a given workflow? To create a person for a particular stakeholder, users might need to provide examples of correspondence, for example.
Execute. Whether ChatGPT, Gemini, or Claude, every major AI program now has an agent mode through which you can set up an ongoing task; a deep research feature that can integrate large amounts of information based on connected data sources; and a scheduler to automate tasks, the authors say.
Learn. As the model begins to operate, it will collect new information and refine its approach.As with all AI, quality requires user vigilance. People should not only periodically check the agent’s work, but also review its output at crucial junctures, such as before making a major decision.
“As we move towards more agentic workflows, that human-in-the-loop moment is going to be critical,” says Ranjan.
For Neeley and Ranjan, the real promise of agentic AI lies in its ability to work independently, pulling together information from multiple sources and synthesizing it in ways that go far beyond prompt-based interactions. As these systems mature, they argue, agentic AI could function as a digital support resource for leaders, helping them make sense of complexity, focus their attention, and perform at a higher level.
First Published: Mar 12, 2026, 13:50
Subscribe Now