Imagine a team in your office where each member is a specialist. One person is an expert at scheduling, another excels at data entry, and a third can draft clear and comprehensive reports. Now, imagine they’re AI. This is the core idea behind an agentic AI system: a team of specialized agents working together to automate complex tasks. By handling repetitive, step-by-step work, the AI system is designed to free up people to focus more on creativity, strategy, and complex problem-solving – the things humans do best. Companies across all industries are in a constant race to become more efficient, and I believe the winners will be those who thoughtfully embrace this change.

While the idea of agentic workflows isn’t new, recent breakthroughs in AI have made building these digital workforces a potentially easier and cheaper proposition. You don’t necessarily need to be an experienced programmer to build these AI teams. Thanks to LLMs and new “no-code” platforms, creating an AI agent can be as simple as describing what you want the agent to do in plain, everyday language, while the technical details are handled automatically in the background.

However, this new type of workplace likely requires a focus on new skills. While software creation and technological expertise will always be highly valued, management skills and a deep understanding of business processes will also be essential for designing the most effective automated systems. A critical job for humans will be to act as the managers and architects of the AI workforce, perhaps in a similar way to how software architects work today. This agentic architect role will likely have to clearly define the workflow steps, data requirements and the exceptions to the automated rules, among other things, to create a system that executes business processes with as few unforeseen hiccups as possible.

At an organizational level, governance of these systems, as well as scalable and adaptable frameworks, will be paramount. Effective entitlement administration is now more critical than ever.  Companies will need to consider screening protocols for prompt vulnerabilities and assess the suitability of language models (large and small). Important in that is functional ownership. As powerful as agentic systems are, human accountability can’t be delegated.

The opportunities for creative use of these AI agents are potentially limitless. At the moment, an AI agent built on one company’s platform can’t easily communicate with an agent built on another. They are stuck within their own digital walls. The future vision is a world where these walls come down. I can imagine an “internet of agents,” where AI from any company, anywhere in the world, can be discovered, connected, and coordinated to work together seamlessly. A company could hire a specialized AI agent from a vendor for a few hours to complete a specific task, in some ways similar to engaging a freelancer.

Getting there won’t be easy. It will require major tech companies and industry leaders to cooperate and agree on open standards for agent-to-agent communications. Just as we have standards that allow any computer to connect to the internet, we will need them for AI agents to connect to each other. This future, which will also require applicable oversight, training, governance and other protocols to be in place, may take time and compromise to build, but I believe the potential economic benefits will be worth it.