By IT Brew Staff
less than 3 min read
Definition:
A chatbot trained by a large language model outputs an answer to a question. An AI agent takes the extra step of initiating action on a user’s behalf, automating a workflow to accomplish tasks like booking a flight, scheduling an appointment, or initiating a password reset.
The parts
An agentic software program has the following components, according to Nvidia:
- An agent core (the decision-making model containing a “user manual” of available tools)
- A memory module (containing internal logs)
- Tools (third-party APIs to connect to services)
- A planning module (that breaks an action down into substeps)
Hold up
Some pros envision an ecosystem where multiple agents will work together to perform a task. IT practitioners will be tasked with ensuring brake mechanisms to prevent any unexpected decision-making—data scientist Richard Bownes recommends practices like limited autonomy and multiple model uses.
Next steps
Agents have resulted from early 21st century achievements in neural networks—systems that make decisions based on training data. AlexNet, developed in 2012, demonstrated high-speed image recognition. Ten years later, GPT was released, which along with its successors has used the internet’s data sources to quickly output answers to questions. Agents aim to be the next evolution.
“It wouldn’t surprise me if there was a little bit of a lag between the excitement of what this potential is and then just the time it takes to convince customers to implement it,” Dave McCarthy, research VP at market intelligence firm IDC, cautioned about agentic deployments in January 2025.
