Everything you need to know about AI agents and what they can do

By Andrew Tarantola Published January 19, 2025

AI2-Thor multi-agent — Allen Institute for A.I.

The agentic era of artificial intelligence has arrived. AI agents are capable of operating independently and without continuous, direct oversight, while collaborating with users to automate monotonous tasks. Based on the same large language models that drive popular chatbots like ChatGPT and Google Gemini, agentic AIs differ in that they use LLMs to take action on a user’s behalf rather than generate content.

In this guide, you’ll find everything you need to know about how AI agents are designed, what they can do, what they’re capable of, and whether they can be trusted to act on your behalf.

Recommended Videos

What is an agentic AI?

Billed as “the next big thing in AI research,” agentic AI is a type of generative AI model that can act autonomously, make decisions, and take actions towards complex goals without direct human intervention. These systems are able to interpret changing conditions in real-time and react accordingly, rather than rotely following predefined rules or instructions.

AutoGPT and BabyAGI are two of the earliest examples of AI agents, as they were able to solve reasonably complex queries with minimal oversight. AI agents are considered to be an early step towards achieving artificial general intelligence (AGI). In a recent blog post, OpenAI CEO Sam Altman argued that, “We are now confident we know how to build AGI as we have traditionally understood it,” and predicted, “in 2025, we may see the first AI agents ‘join the workforce’ and materially change the output of companies.”

Marc Benioff hailed AI agents’ emergence as “the third wave of the AI revolution” last September. The “third wave” is characterized as generative AI systems outgrowing being just tools for human use, instead, evolving into semi-autonomous actors capable of learning from their environments.

“This is the biggest and most exciting piece of technology we have ever worked on,” Benioff said of the company’s newly announced Agentforce platform, which enables the company’s enterprise customers to build digital stand-ins for their human customer service reps. “We are just starting.”

What can AI agents do?

Being designed to take action for their users, AI agents are able to perform a staggeringly wide variety of tasks. It can be anything from reviewing and automatically streamlining computer code to optimizing a company’s supply chain management across multiple vendors to reviewing your calendar availability then booking a flight and hotel accommodations for an upcoming business trip.

Claude | Computer use for automating operations

Claude’s “Computer Use” API, for example, enables the chatbot to effectively mimic the keyboard strokes and mouse movements of a human user, enabling Claude to interact with the local computing system. AI agents are designed to tackle complex, multi-step problems such as planning an eight-course dinner party by establishing a menu after contacting guests for their availability and potential allergies, then ordering the necessary ingredients from Instacart. You’ll still have to cook the food yourself, of course.

Where can I see an AI agent in action?

AI agents are already being rolled out across myriad industries. You can find agentic AI in the banking system where it assists with fraud detection and automated stock trading tasks. In the logistics industry, AI agents are used to optimize inventory levels and delivery routes as market and traffic conditions change. In manufacturing, AI agents are already helping to enable predictive maintenance and equipment monitoring, ushering in an era of “smart” factory management. In healthcare, AI agents help patients streamline appointment scheduling and automate prescription refilling. Google’s automotive AI agent will even provide near-real-time information about local landmarks and restaurants for Mercedes’ MBUX entertainment and navigation system starting with the next model year’s CLA.

The technology is also being applied to enterprise business and Salesforce is far from the only SaaS company to embrace AI agents. SAP and Oracle both have similar offerings for their own customers.

It should come as no surprise then that the industry’s top companies like Google, Microsoft, OpenAI, Anthropic and Nvidia are all racing to develop and deploy AI agents for the business and consumer markets as well. In November, Microsoft announced Copilot Actions, which would see Copilot-based agents integrated throughout the company’s 365 app ecosystem, and began rolling the feature out to business and enterprise users in January 2025.

NVIDIA AI Building Blocks for Agentic AI

In November, Google Cloud announced its AI agent ecosystem program, dubbed AI Agent Space, which, like Agentforce or Google’s other AI agent platform, Vertex AI, enables business customers to develop and deploy their own customized AI agents. Nividia unveiled its Nemotron model families, designed specifically for agentic AI tasks, at CES 2025 earlier this month.

For its part, OpenAI recently unveiled its new Tasks feature for ChatGPT which allows users to set future reminders and regularly-scheduled tasks (like weekly news roundups) for the chatbot to perform at a later date. The company has also developed an AI agent of its own, codenamed Operator, which it released in January 2025.

Anthropic

Are AI agents safe to use?

That depends on your definition of “safe.” Because agentic AI systems are built atop hallucination-prone large language models susceptible to adversarial attack, AI agents are themselves prone to hallucinations and can be tricked by malicious actors to behave outside of their established safety guardrails. A 2024 study from Apollo Research, for example, found that tasking OpenAI’s o1 model with achieving a goal “at all costs” led the AI agent to try to disable its monitoring mechanism before copying “what it believes to be its weights to a new server and then lie about it to its developers,” claiming it suffered “technical errors.”

Of course, when a chatbot boofs its answer, the stakes are relatively low (unless that user is a lawyer or Google, mind you), compared to what would happen if an AI agent hallucinates data about its automated stock trading strategy. As with all generative AI, users need to be vigilant about what information (be it financial, medical, or personal) they share with AI chatbots and LLMs.

Topics

Artificial Intelligence

Andrew Tarantola

Former Digital Trends Contributor

Andrew Tarantola is a journalist with more than a decade reporting on emerging technologies ranging from robotics and machine…

Computing

Microsoft 365 Copilot gets an AI Researcher that everyone will love

Researcher agent in action inside Microsoft 365 Copilot app.

Microsoft is late to the party, but it is finally bringing a deep research tool of its own to the Microsoft 365 Copilot platform across the web, mobile, and desktop. Unlike competitors such as Google Gemini, Perplexity, or OpenAI’s ChatGPT, all of which use the Deep Research name, Microsoft is going with the Researcher agent branding.
The overarching idea, however, isn’t too different. You tell the Copilot AI to come up with thoroughly researched material on a certain topic or create an action plan, and it will oblige by producing a detailed document that would otherwise take hours of human research and compilation. It’s all about performing complex, multi-step research on your behalf as an autonomous AI agent.
Just to avoid any confusion early on, Microsoft 365 Copilot is essentially the rebranded version of the erstwhile Microsoft 365 (Office) app. It is different from the standalone Copilot app, which is more like a general purpose AI chatbot application.
Researcher: A reasoning agent in Microsoft 365 Copilot
How Researcher agent works?
Underneath the Researcher agent, however, is OpenAI’s Deep Research model. But this is not a simple rip-off. Instead, the feature’s implementation in Microsoft 365 Copilot runs far deeper than the competition. That’s primarily because it can look at your own material, or a business’ internal data, as well.
Instead of pulling information solely from the internet, the Researcher agent can also take a look at internal documents such as emails, chats, internal meeting logs, calendars, transcripts, and shared documents. It can also reference data from external sources such as Salesforce, as well as other custom agents that are in use at a company.
“Researcher’s intelligence to reason and connect the dots leads to magical moments,” claims Microsoft. Researcher agent can be configured by users to reference data from the web, local files, meeting recordings, emails, chats, and sales agent, on an individual basis — all of them, or just a select few.

Why it stands out?

Computing

Samsung might put AI smart glasses on the shelves this year

Google's AR smartglasses translation feature demonstrated.

Samsung’s Project Moohan XR headset has grabbed all the spotlights in the past few months, and rightfully so. It serves as the flagship launch vehicle for a reinvigorated Android XR platform, with plenty of hype from Google’s own quarters.
But it seems Samsung has even more ambitious plans in place and is reportedly experimenting with different form factors that go beyond the headset format. According to Korea-based ET News, the company is working on a pair of smart glasses and aims to launch them by the end of the ongoing year.
Currently in development under the codename “HAEAN” (machine-translated name), the smart glasses are reportedly in the final stages of locking the internal hardware and functional capabilities. The wearable device will reportedly come equipped with camera sensors, as well.

What to expect from Samsung’s smart glasses?
The Even G1 smart glasses have optional clip-on gradient shades. Photo by Tracey Truly / Digital Trends
The latest leak doesn’t dig into specifics about the internal hardware, but another report from Samsung’s home market sheds some light on the possibilities. As per Maeil Business Newspaper, the Samsung smart glasses will feature a 12-megapixel camera built atop a Sony IMX681 CMOS image sensor.
It is said to offer a dual-silicon architecture, similar to Apple’s Vision Pro headset. The main processor on Samsung’s smart glasses is touted to be Qualcomm’s Snapdragon AR1 platform, while the secondary processing hub is a chip supplied by NXP.
The onboard camera will open the doors for vision-based capabilities, such as scanning QR codes, gesture recognition, and facial identification. The smart glasses will reportedly tip the scales at 150 grams, while the battery size is claimed to be 155 mAh.

Computing

The search system in Gmail is about to get a lot less frustrating

Gmail icon on a screen.

Finding relevant information on Gmail can be a daunting task, especially if you have a particularly buzzy inbox. Right now, the email client uses a search operator system that acts somewhat like a shortcut, but not many users know about it.
Today, Google has announced an update to how search on Gmail works, thanks to some help from AI. When you look up a name or keyword in Gmail, the matching results are shown in chronological order.
Moving ahead, search results will be shown based on their relevance. In Google’s words, relevance will take into account three factors viz. frequent contacts, most-clicked emails, and how recently the relevant emails arrived in your inbox.

Old search (left), new search (right) Google
“With this update, the emails you’re looking for are far more likely to be at the top of your search results — saving you valuable time and helping you find important information more easily,” the company says in a blog post.
The updated search system in Gmail is rolling out to users worldwide, and it will be implemented on the desktop version as well as the mobile app. And just in case you are wondering, this is not an irreversible change to the search function in Gmail.
Google says users can switch between “most relevant” and “most recent” search results at their convenience. The overarching idea is to help users find the intended material at a quicker pace.

nproxy.org