How AI Agents Usher in Smarter Automation for Enterprises

Much of today’s business operations are already automated. Think of processing payroll, software updates and data backup and recovery, among many others. These fall under Robotic Process Automation, or RPA. It is software that automates repetitive tasks based on predefined if-this-then-that rules. RPA is like a highly skilled but inflexible assembly line of workers.

An artificial intelligence (AI) agent is different. It is software that also automates tasks, but acts more like an adaptable knowledge worker. Instead of merely following predefined tasks under preset rules, it can understand context, make decisions and adjust its approach based on changing circumstances. Importantly, it can autonomously carry out actions on behalf of users.

When RPA faces a problem that doesn’t match its template, it needs to be reprogrammed for changes in process. When an AI agent faces the same situation, it can reason through the problem and come up with a fitting response. It can learn and adjust to new situations. RPA uses structured inputs and logic; AI agents use unstructured data and reasoning.

For example, an RPA-powered chatbot on a website can answer pre-programmed questions for the most commonly asked questions by customers. But deviate from it, and the chatbot provides incorrect or irrelevant answers. Now imagine an AI agent trained on GPT-4, for example. Its response is more natural, and it can better answer the customer because it can understand the question and dynamically find the answer.

While AI agents are a smarter form of automation than legacy RPA, they will complement rather than replace it. Each handles tasks best suited for them. For example, it is unnecessary to use an AI agent when a simple RPA bot can do the job more economically. RPA’s actions also do not deviate, which makes it predictable and auditable — better in highly regulated industries.

An AI agent is a software program that can autonomously accomplish tasks to meet goals. At the center of AI agents are large language models (LLMs). AI agents combine the power of generative AI models with tools that let them access and interact with real-world data sources.

While a human user sets goals, an AI agent finds the best route to take to accomplish that goal. For example, a customer service AI agent wants to answer a customer’s question or solve a problem. It would autonomously get more details from the customer, look up information from internal documents and offer a solution. It would also determine if this customer needs to talk to a human customer service representative.

Unlike generative AI models, AI agents can reason, plan and execute tasks on their own that mimic human-like problem-solving. They can perform tasks without human intervention, although humans can still approve the final outcome. For example, an AI agent can look at your calendar and notice you have an off-site meeting coming up. It can book an Uber ride for you, but you will be able to approve the reservation.

The AI agent is self-learning, adapting to the user’s expectations over time. Its ability to store past interactions in memory and plan future actions can lead to personalized experiences and comprehensive responses. For example, AI agents can tap external datasets, do web searches and collaborate with or even supervise other agents.

Here’s an example of how it could work, according to IBM.

A traveler wants to plan a vacation and asks an AI agent to predict which week in the coming year would be most likely to have the best weather for a surfing trip in Greece. Since LLMs do not specialize in weather patterns, the AI agent has to retrieve the data from an external database of daily weather reports for Greece in past years. But the AI agent still doesn’t have enough data for a prediction.

So it creates a subtask: Communicate with an external agent that specializes in surfing. In doing so, the AI agent discovers that high tides, sunny weather and little to no rain are optimal conditions for surfing. The AI agent combines all the data it has gathered to predict which week in the coming year will have these attributes.

After presenting its findings to the traveler, the AI agent stores this information along with the traveler’s feedback to improve performance and adjust to the user’s preferences. To avoid repeating the same mistakes, the AI agent can keep data about prior solutions in a knowledge database.

AI agent enterprise solutions in the market include the following: Microsoft offers Copilot users the ability to create, manage or deploy AI agents in its Copilot Studio, while its Azure cloud division provides a fully managed AI Agent Service. Google Cloud offers a portfolio of pre-built AI agents for common use cases, while AWS is working with clients to build AI agents on its Bedrock generative AI platform. Nvidia offers AI agent blueprints that can be customized by industry. Salesforce offers Agentforce, which lets users build and customize AI agents.

To be sure, many applications of AI agents cut across industries. For example, in customer service, AI-powered virtual assistants or chatbots with AI agent capabilities could provide personalized, proactive support by anticipating customer needs and resolving issues autonomously.

The following is a sample list of industry-specific applications:

Healthcare

• AI agents can analyze patient data to recommend individualized care plans, monitor progress and ensure the level of care adjusts to the patient’s needs to drive better outcomes.

Retail

AI agents can act as virtual personal shoppers, analyzing customer preferences and tastes and providing customized recommendations, creating a tailored shopping journey.

Smart manufacturing

• AI agents could optimize factory production lines through autonomous monitoring and real-time adjustments. They can predict coming maintenance needs, reducing downtime and costs.

Financial services

• Financial institutions could use AI agents to manage portfolios, detect fraud and optimize trading strategies by responding dynamically to market conditions, deciphering trading signals to adjust strategies and mitigates risks in real time.

Automotive

• AI agents offer real-time, data-driven insights to enhance driving safety. They can continuously monitor driver performance and provide timely interventions to reduce the number of hazardous incidents.

The post How AI Agents Usher in Smarter Automation for Enterprises appeared first on PYMNTS.com.