Agent AI: What is this, why are people talking about it?

According to the AI ​​overlord, this is a year to proxy AI.

You may have seen Google announce its “age of agents” through web browsing research assistants and AI robots called Nail Salons and Mechanics. Openai leaders talked about agent AI being the “big topic of 2025” and have introduced a preview of the operator’s research, an agent that can perform tasks on your behalf, as well as in-depth research, “doing multi-step research on the internet to complete the Internet for complex tasks.” Microsoft has just launched Microsoft Discover, an enterprise-agent AI tool for scientists. Your next smartphone can have proxy capabilities that can send custom messages, create calendar events, or get information from different applications.

If you nod and smile every time you mention Agentic AI, don't feel embarrassed. This is a new entry in the AI ​​vocabulary, but it can never be ignored.

So what exactly is proxy AI?

“Agent AI refers to a class of artificial intelligence systems designed to operate autonomously, perceive its environment, set goals, plan actions to achieve these goals and execute these plans without continuous human intervention. These systems can learn and adapt over time based on feedback and new information.”

That's based – what else? – Google's AI chatbot Gemini.

Unlike generating AI, it is essentially a tool to create some kind of output – code, text, audio, images, video – Agent AI can perform tasks independently on behalf of the user. This is an improvement over the standard AI chatbot experience. Rather than responding based on training materials, agent AI takes other steps, such as conducting an internet search and analyzing results, consulting other resources, or completing tasks in another application or software.

You may have heard of this term used interchangeably with AI agents, but Agesic AI is a broader term that covers technologies that may not be completely Autonomous, but with some proxy-like features.

Therefore, OpenAI thinks that the operator is an AI agent because it is context-conscious and can perform tasks for you, such as sending text messages. Its in-depth research tool is proxy AI, because it can crawl autonomously and compile reports for users, although its functionality is almost stopped for now.

Agentic AI is powered by more advanced inference models such as Chatgpt O3 and Gemini 2.5 Pro Preview, which can decompose complex tasks and make inferences. This brings models like large language models closer to mimicking the way the human brain works. Unless you keep re-exploring generative AI models with new information, it won’t be able to learn new things, says Karen Panetta, an IEEE researcher and professor of engineering at Tufts University. “Another AI can learn from looking at other examples and can be more autonomous with more exploratory or return information in activities that break down tasks and help you with more target-driven types.”

When combined with computer vision, this is what allows the model to “see” the user's computer screen, and we get proxy AI that everyone is very excited about.

Why is everyone talking about proxy AI?

Proxy AI is not completely new. Both self-driving cars and robot vacuums can be regarded as early examples of proxy AI. They are technologies with autonomous performance that rely on advanced sensors and cameras to understand their environment and respond accordingly.

But there are several reasons for the moment of Agesic AI. It is crucial that the latest models get better and easier for users (although sometimes also Friendly). As people began to rely on AI chatbots like Chatgpt, using these tools to automate daily tasks, such as responding to emails, has attracted people's interest. With Agesic AI, you don't need to be a computer programmer to automate with ChatGpt. You can simply tell the chatbot what to do and execute your instructions in simple English. At least, that's the idea.

Companies like OpenAI, Google, and Anthropic are all relying on AgentIC AI because it has the potential to move technology beyond the novel chatbot experience. With Agesic AI, tools like Chatgpt are essential for businesses and individuals. Agent AI tools can order groceries online, browse and buy the best rated espresso machines for you, and even research and book holidays. In fact, Google has taken a step in this direction with its new AI shopping experience.

In the business world, companies are seeking Agesic AI to solve customer service queries and adjust their stock trading strategies in real time.

What might be wrong?

This tweet is currently unavailable. It may be loading or being deleted.

Is it risky to release automated robots in the wild? Why, yes.

There is always the risk of an agent operating on behalf of your agent that may send sensitive emails to the wrong person or accidentally make a large purchase. Then there is the issue of responsibility. “I'm going to let my agent do something because I'm going?” Panetta wanted to know. “Saying I’m an official for something, I use an AI agent to make decisions, help us with plans, and then you lose money from the organization.”

Major AI players have put in place safeguards to prevent AI agents from continuing to rogue, such as requiring human supervision or approval of sensitive tasks. Operators do not take screenshots when in human coverage mode and currently do not allow their agents to conduct bank transactions, Openai said.

But when did the technology become more common? As we become more satisfied with proxy AI, will we become more passive and relaxed with supervision? Early in this article, we used Google Gemini to help define Agentic AI. If we rely on AI tools or even simple learning, will humans become stupid?

Then we have to provide extensive data access. Of course, Chatgpt will be convenient to automatically filter, sort, or even delete emails. But do you want to give AI companies full access to every email you send or receive?

What about bad actors who don’t have such safeguards? Panetta warns that using proxy AI is increasingly complex in cyber attacks.

“Because it’s so cheap to have access to powerful computing right now, it means bad actors can access it,” she said. “They can run simulations and be able to come up with complex schemes to break into your system, or use it to raise this stock loan.”

AI has always been a double-edged sword with equally powerful damage and benefits. As proxy AI prepares for prime time deployment, the bets are getting higher and higher.

Disclosure: Mashable's parent company Ziff Davis filed a lawsuit against Openai in April, accusing it of infringing on Ziff Davis' copyright in training and operating its AI systems.