After the novelty of generative artificial intelligence (AI) disappeared, many people asked an important question – yes, it’s cool, but how does it make an impact in the real world? This is a valid question. While AI chatbots can be considered as a one-stop service to quickly find information, engage in improvisational conversations, make them write papers, generate images or videos, their roles are largely limited to the fact that human users must often use them. The system commands it gets the output and supervises the results.
Even if its capabilities cannot be disbanded, and does have a significant impact on improving workers’ productivity in certain areas, it lacks a key factor that prevents it from becoming a loyal assistant who can handle and truly automate tasks-decision. Today’s Generative AI can help you with certain aspects of your work, but cannot perform tasks.
For example, you could ask it to email it to a customer to let them know of unexpected delays, but it can’t send that message or process the angry replies they sent. Similarly, you can ask for “the best smartphone to shoot videos” using Gemini or Chatgpt, and you can recommend the latest iPhone 16 Pro Max or Samsung Galaxy S24 Ultra. But it won’t be able to search the web to find your best deal and make a purchase.
Realizing this gap, tech companies working in large language models (LLM) began to use the term AI proxy. Researchers believe that AI agents can take knowledge-based AI systems and transform them into action systems that can perform end-to-end tasks without human intervention.
The term gained prominence in the second half of 2024 and is currently seen as a panacea for all work-related issues. Despite some truth, is it really a transformative technology of this potential? The answer may be a bit complicated, but we will do our best to break it down and highlight all the different aspects you should know. Let’s dig deeper.
What is an AI proxy?
Since the technology is still in its nascent stage, there is no unified definition of what constitutes an AI agent. IBM defines it as a “system that can perform tasks autonomously on behalf of users” by designing workflows and using tools. Similarly, Google announced last year its first AI agent known as the Mariner Project Mariner project, calling it a system that is like human assistants and helps them complete tasks.
Amazon gives a more comprehensive definition that describes it as a software program that can interact with the environment, collect data and use the data to perform custom tasks to achieve predetermined goals. Humans set goals, but AI agents People independently choose the best actions needed to achieve these goals.”
In short, AI agents can be understood as AI systems that can take action, rather than just telling users about their actions.
Decompose AI Agents
A typical AI agent will have a large language model (LLM) as the brain. But it will also include other elements that enable it to use that intelligence in action. Most commonly, these additional parts are different sensors, mechanical parts, encoders or integrated into other software.
These sensors enable AI agents to collect data in different formats. These can be visual, sound, temperature or electronic signals. Mechanical parts are often used in embodied AI or robots that require actual operations such as lifting an object or moving from one position to another. Encoders are used to convert different types of signals into information that can be processed by the LLM. Finally, software integration enables the ability to perform tasks.
At this point, it is also important to emphasize another key difference between the AI model and the AI agent. The AI model contains a pre-trained database that forms the basis of its knowledge. Nothing that does not belong to the database will generate output. A good example is an earlier version of Chatgpt, which is not connected to the Internet and has a knowledge deadline. If it is prompted to answer questions about current affairs, it will not be answered.
Conversely, when integrated with related systems, AI agents can independently collect new data to solve the problem that is impossible based on their existing databases. For example, Google’s project Mariner can interact with the browser to find the best deals on a smartwatch.
Another aspect of AI agents is the ability to handle complex tasks. AI agents are able to perform advanced inference, so complex tasks can be broken down into multiple easier tasks and then completed them one by one. The contextual understanding of the problem and the ability to know how to decompose the problem are the basic functions of AI agents.
A good example is Gemini’s recently added deep research tool. Users can ask it to explain technology or niche topics. AI will then develop a multi-step research plan that breaks the topic into smaller sections, finds relevant research papers and articles on the topic, executes plans, conducts research and analyzes the collected data to create detailed reports.
Application of AI Agent
AI companies have been touting AI agents as tools that can be used in the industry and in different situations. It can be used as a voice assistant for devices that can perform device-specific tasks (such as taking photos or playing music). It can be added to an application or software and perform tasks in it (purchasing products through a browser-based agent). It can also be added to the enterprise system, which can detect fraud or find ways to optimize different processes.
In addition to this, AI agents are said to be able to perform transformative tasks in certain industries. In healthcare, it can be used for diagnosis, treatment recommendations, and drug discovery. In the automotive industry, it can be used to create self-driving cars. AI agents are also said to be able to fly drones in disaster areas to collect and analyze data and provide actionable insights into rescue operations.
It also develops personalized learning plans and uses test paper ratings in the manufacturing industry through AI-powered robots, game developers or non-playing game characters (NPCs) in games and in the education field as well as in the education field. Programs and are used as game developers (NPCs) in the gaming industry. A kind of human fashion.
However, it is important to note that while tech companies are selling AI agents as the entire goal of various end-to-end intelligent automation, current technologies limit their use cases to basically specific task-based roles rather than general roles -Use the tool.
AI Agents in 2025
That being said, it is important to understand our expectations and understand what we actually expect from AI agents back then. AI agents are unlikely to enter the workforce in any critical sector, such as manufacturing, automobiles, healthcare or education.
However, this year should mark the entry of AI agents in consumer electronics, mobile and desktop applications, as well as websites and platforms. For example, Google’s Project Mariner can integrate with Google Chrome and help users purchase and find files from the web by the end of the year.
OpenAI is rumored to have launched an AI proxy this year, which can further enhance Chatgpt’s capabilities and allow it to perform certain operations on users’ devices and the Internet. Anthropic’s computer usage tools are expected to make global versions, and assist users in completing their daily tasks on the device.
Ultimately, we should also see a change where AI agents can mimic keystrokes, mouse actions and clicks, and do more on the device. For example, by the end of the year, more proxy tools such as coding agents might be to write code end-to-end, test it, find and fix failures, and deploy them without human intervention. However, it is very optimistic to include it in the 2025 itinerary.
On the enterprise side, AI agents can play a greater role in completing some organizational tasks, such as monitoring large amounts of data, preparing analytical reports, and providing advice and course corrections. It can also be used in certain cybersecurity roles. It is worth noting that Meta says it has used AI to ensure the guidelines are followed. YouTube also uses AI to monitor copyright infringement.
However, we do not want AI agents to enter any critical working function this year, as the technology is largely untested and its reliability will be questioned. Businesses, especially public or large investors-backed businesses, are often risk-averse and unlikely to provide access to sensitive data.
AI Agent Problems
Since AI is a current trend in the technology field and has the potential to disrupt a large number of industries, it is understandable why it is so excited about AI agents. But besides rose-colored glasses, there are several issues with AI agents that need to be addressed before technology sees mass adoption. On the other hand, if not inspected, the technology can pose several risks.
One of the main problems with AI agents is bias and discrimination from training data and can lead to discriminatory results. This also highlights another transparency issue in AI proxy. With complex algorithms and architectures, most AI agents are complex and opaque systems, and it is difficult to understand how and why certain decisions are made.
There are also security and privacy issues. From a security perspective, AI agents can be vulnerable to adversarial attacks, with malicious actors manipulating input data to deceive the system. Additionally, AI agents also pose privacy risks as they need to connect to multiple systems and collect large amounts of data to perform tasks.
Facing many challenges, AI companies will be working hard to convince businesses and individuals to the benefits of technology while reassuring their adverse aspects. In any case, it is undeniable that AI agents will form a large part of the AI announcement in 2025.
