Can AI Agents Ease Workloads for Enterprises?

The Components of an AI Agent

According to experts, there are three key components of an AI agent:

1. Large language models: LLMs allow agents to perform specific tasks by responding to feedback, according to Andersen.

LLMs gather context using retrieval augmented generation. A general-purpose model can connect to RAG, and a RAG application can connect an LLM to specific information a business needs to know, Andersen explains.
RAG also allows AI agents to retrieve proprietary sensitive patient data in healthcare, and customer data in customer service, Saunders says.

2. Memory: AI agents can remember a customer’s buying history, which is particularly beneficial in retail. For example, Andersen says, an AI agent could suggest the type of water bottle to buy that would fit into the cupholder of a car the customer bought.

“It provides a level of depth in terms of helping somebody make selections, because it understands what you've done in the past to a degree that you don’t see in the simpler types of e-commerce applications today,” Andersen says.

3. Tools: The AI agent performs tool-calling to send out queries to task-driven models that can do more specific tasks, says Saunders. The LLM understands how to communicate with these tools through application programming interfaces.

Another tool consists of the AI agent’s workflow, which dictates the steps for a job to be completed. Companies such as Salesforce and ServiceNow offer AI agents to help with basic workflows, Andersen notes.

DISCOVER: The power of AI and data-driven decision-making.

What Are the Benefits of AI Agents?

A key benefit that AI agents provide is handling repetitive tasks, something that can be a shortcoming for people at times. Microsoft Copilot Studio allows organizations to build agents to respond to “autonomous triggers” for business tasks.

Google launched an AI agent in May, called Astra, which lets users experiment using audio and video, and NVIDIA offers an API catalog of blueprints for AI agents.

“What's great about these is they can be designed and trained on your brand, and you can have a look and feel and a style that carries your company's experience into that AI agent that your customers are working with,” NVIDIA’s Saunders says.

Another type of agent, called a reflex agent, can respond to single-step instructions. Reflex agents handle concrete steps such as writing a LaMDA handler, calling a certain function and generating a block of text for that function, says Anu Sharma, the director of Amazon Bedrock Experiences and Tools at AWS.

Amazon Q Developer serves as a contextually aware agent that could help HR professionals understand the context for company policies and carry out a benefits eligibility workflow to identify which policies apply to certain employees, Sharma says.

Users can review changes to code that AWS Q Developer suggests and can fix errors in an agent’s recommendations.

Q Developer has been able to free software engineers for other tasks while upgrading about 30,000 Java applications, according to Sharma.

“We are allowing our engineers to have more time to solve problems and invent on behalf of our customers, so they're doing more problem-solving work rather than undifferentiated work while still delivering at the speed and quality they want to maintain for their software,” Sharma says.

What Are Some Applications of AI Agents in Real Life?

A network admin could use an AI agent to roll out a change to a router, then gather information about the router and create an API to interact with it, Andersen explains.

“Instead of forcing that human to go back to the internet and look at the settings, the AI agent can do it itself,” Andersen says. “It is able to say, I know you're this model router. I know these things about you. I can now actually take the ability to go online, look up that information about you, and begin to build an API around it and put that in place.”

AI agents can also perform predictive maintenance to anticipate hardware failures or network issues. For predictive maintenance applications (which have certain thresholds, such as the temperature of machines), you can have the AI agent issue a security alert to avoid a safety issue, Andersen suggests.

Agents can also assist IT professionals with handling internal help desk requests and resetting passwords automatically. They can also help customer service agents triage questions from customers that may come in at peak times.

“Not only does it sometimes make a better experience for the person asking the question because they're getting a much faster response, but it also frees up employees’ time to be able to answer the more complicated questions or to address those that need that human interaction,” Saunders says.

DIG DEEPER: Why is customer service the focus of most digital transformation projects?

In healthcare, AI agents can help physicians with patient monitoring and collect data on patient conditions, Andersen says.  “The doctor can walk out of that room and go care for the next patient, and from listening to the meeting, the agent could then take those action items discussed in the meeting and go perform them,” Andersen says.

In financial services, teams can use Amazon Q Developer to improve customer conversion rates and reduce churn. The agent prefills the occupation of a purchaser to offer better matches for products and improve the company’s conversion ratio, Sharma says.

In addition, online tools such as Asana allow businesses to embed AI agents as part of their teams. Asana AI Studio lets organizations build these AI agents without code as part of their workflows. With low-value busywork taking up 53% of employees’ time, according to Asana’s 2024 State of Work Innovation report, AI agents can ease this burden for employees.

How Will AI Agents Change the Way We Work?

Going forward, agents will become part of an end-to-end process that will involve more than just managing code, according to Andersen. They will also perform more tasks related to code automatically, including documentation and automating and delivering tests, he says. 

“Developers are going to become more powerful, and they're going to be able to develop a higher quality and quantity of applications because of the agents,” Andersen says. “It's like going from a handsaw to a chainsaw.”