Beena Ammanath – Global Deloitte AI Institute Leader, Founder of Humans For AI and Author of “Trustworthy AI” and “Zero Latency Leadership.”
In this era of humans working with machines, being an effective leader with artificial intelligence (AI) takes a range of skills and activities. In this series, I provide an incisive roadmap for leadership in the age of AI.
The Difference Between Automation And AI Agents
AI innovations are advancing rapidly, bringing a deluge of new terms that can lead to confusion due to their complex nature. One of the most exciting developments in the field is the emergence of AI agents or multi-agent systems. But a common question is, how are these agents different from the AI automations that are already widely used?
AI automation typically refers to a specific task or function performed by a machine. AI can fuel many types of automation, and the focus of research to date has been on increasing the volume of activities that AI can process. However, there is a limit. Robotic process automation (RPA) has carried us a long way toward an AI-fueled world, but static, rule-based systems struggle with complexity. Unstructured data, shifting conditions and an inability to reason limit RPA’s capacity to function over workflow sequences.
An AI agent is a combination of machine learning, natural language processing, Generative AI and cognitive reasoning. It can reason, adapt and act independently, performing complex work without human intervention. That work may include engaging with data, systems and people to understand context, plan workflows, connect to external systems and make decisions. AI agents don’t need to be told how to accomplish a task; they determine that autonomously. Multi-agent systems take this a step further, combining multiple role-specific AI agents that can break down a goal and achieve it collectively.
Consider An Analogy
A restaurant owner decides she wants to add a new dish to the menu—chicken alfredo—and she sends an email to inform staff that she wants to begin serving it the next day. The restaurant swings into action: A sous chef checks the inventory to ensure they have the right ingredients, and he calls up the supplier for a fresh shipment of chicken. The head chef creates the dish so the cooks understand the recipe and the servers learn how to promote it to customers. A senior manager determines the optimal price for the dish, and a junior worker edits the menu and prints the updated offerings.
The restaurant is now ready to serve chicken alfredo, and the restaurant owner did not perform any of the necessary actions. She simply said what she wanted, and the restaurant personnel acted on her behalf, including all the reasoning and forethought necessary to do so.
Each employee role can be compared to an AI agent, and the restaurant as a whole is not unlike a multi-agent system. A human user tells the system what to do but not how to do it. This is a significant departure from traditional automation, and it is enticing because it may be the technological advancement needed to extract more enterprise value from AI.
AI Agents Have Limitless Potential
Where will these autonomous capabilities lead us? In the near term, agents may supplement (and, in some cases, replace) existing RPA deployments to improve flexibility and decision making. Later, as the technology improves, agents may become more fully autonomous managers (rather than task-oriented assistants) and eventually generalist systems with reasoning capabilities across multiple domains.
Some adoption is already underway, with early movers using RPA as a foundation for AI agents. These automated workflows become building blocks for greater autonomy as AI agents improve in context awareness and adaptive reasoning. Their maturing capabilities prompt a change in thinking about where and how AI can create value. Rather than seeking discrete, routine tasks suited for RPA, organizations can consider higher-order strategic decision making and identify which complex workflows could be supplemented with agentic capabilities or automated entirely.
The potential value is significant, and there are steps organizations can begin taking today to prepare for agentic AI adoption:
• Compile tasks and workflows that may be suited to agentic capabilities and optimize them for AI agents, perhaps by identifying and removing unnecessary process steps.
• Enhance data governance and cybersecurity to protect the enterprise data and systems to which AI agents are granted access.
• Take a granular approach to balancing risk and value when determining the degree of autonomy AI agents should be granted.
There is more work ahead if agents are to become common tools for enterprise value. Yet, as agentic AI improves and organizations identify where it can contribute value, the opportunities will be limited not by technology but by our imagination.
Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?
Read the full article here
