have always been a big fan of suspense movies, especially spy thrillers in which the lead character is an agent from a secret service undertaking an almost impossible task of saving the world from a catastrophe while chasing a group of bad people. Agents Ethan Hunt from the Mission Impossible series and James Bond 007 are physically strong and always equipped with high-tech gadgets, with firm determination to accomplish the job. The agents climb walls, intrude on security parameters, breach safety nets and hack systems. Recently, we have been witnessing the emergence of another kind of agent known as AI agents. These agents are trained to undertake a variety of tasks using resources and means that were unavailable to previous AI models. Like the agents in the movies, these agents are task-oriented. They perceive, reason, act and learn.

AI is becoming part of almost every system and workplace. The agents are rapidly making their way to undertake these jobs. Since the emergence of ChatGPT in November 2022, advancements in AI have been revealed at lightning speed. The world is witnessing a race to dominance among the players, mostly originating from the USA and China. The agents are powered with the latest models and algorithms, such as ChatGPT 5.4, Gemini 3.1 pro, or Opus 4.6. Agentic AI is gaining ground. The agents get verbal or written commands (prompts). The user is then free to indulge in other activities while the agent continues working on the assignment. After completing it, it sends a notification on the mobile device informing the user that the job is done. The agents are your team; your wish is their command.

The AI agents exercise great autonomy. They aim to complete the assignment using online resources with minimal human input. These agents gather data from user chats, web searches and application programming interfaces (APIs). These interfaces help them connect back-end systems to the tasks one intends to accomplish. The agents draw their reasoning power from large language models (LLMs). They break down the tasks, plan the course of action and choose the tools and APIs to use to accomplish the tasks. They also (sometimes) tap into external knowledge sources to retrieve information. The technique employed is known as retrieval-augmented generation (RAG). The RAG makes models smarter and more accurate. Once the agents acquire sufficient information, it is time to run like a movie agent (think of Ethan Hunt running in a Mission Impossible movie). They execute the steps by making API calls, writing code and running commands. In the process, if something fails, it finds another way to accomplish the task. As it encounters various scenarios, the agent stores all the information. Over time, it gets better at handling similar situations. The agents are learning all the time. They are starting to get to know you and your work more closely. They are goal-driven and proactive, unlike traditional AI systems, which are reactive and wait for the user’s commands. The agents act, adapt and learn. AI companies are providing their agents more sophisticated tools to handle complex tasks. These agents can analyse information, generate reports and create videos. They are capable of analysing and criticising even their own outputs. Now is indeed an exciting time for technology enthusiasts.

Although agentic AI requires substantial resources such as LLMs, memory systems and tools, companies are now beginning to build agents capable of accessing data on personal computers and performing tasks with minimal resources. They can act like the owner and administrator of the computer. The agents can now scan your emails, folders and files, apparently with your permission, and accomplish tasks such as replying to emails, booking hotels and ordering your groceries. There are plenty of options for consumers to employ. The agents and companies know this well.

OpenClaw recently made headlines when it announced an agent capable of controlling a personal computer and completing assigned tasks via social networking apps like WhatsApp or Line. OpenClaw runs directly on the computer or server and connects advanced LLMs (like Claude or GPT) to local files, apps and web browsers, enabling it to perform tasks rather than just answer questions. Installing the programme requires expertise. Third-party services like SetupClaw charge thousands of dollars to install and set up the agent on the computer. Obviously, there is a price to be paid for an autonomous personal assistant. It is believed that easy-to-install agentic models are just around the corner. Those will enable everyone to connect them to personal computers and devices. Agents are being trained aggressively for the tasks.

Companies are also providing customers with the chance to make customised agents for specific jobs. It’s like making your own Ethan Hunt or James Bond. As part of my academic work, I have explored various agentic models currently offered by companies. There are options for using general-purpose agents for academic activities such as searching the literature and compiling reports, but more customised models are also available with ample pre-training. These agents are now widely used for research and writing in the academic world. SciSpace currently offers over 800 agents exclusively for academic tasks, ranging from literature search to thesis writing.

The rise of agents is a cause for concern as well. In movies, we have seen agents go rogue. Since they know a lot about one party, a rogue agent is even more dangerous than the actual adversary. These double agents are hard to identify. They operate under layers. They share vital information with the other party and it may be used in dangerous way. Security may be breached and data transferred without consent, with no trace. In an agentic AI model, this is a clear and present danger. An agent working for you can also work for others - a precarious situation indeed. Recently, there was a report of an AI agent formatting the user’s hard drive, erasing all data. As agents fetch data for you on your instructions, your own data (private and personal) may also be in danger. Agents require your permission to use your computer’s data. Once it is granted, they scan everything on it. The implications for privacy and security are serious. Agents run their routines and algorithms, largely unknown to most users. Once you permit an agent to accomplish the task, they make sure to do it by any means necessary, exactly like the agents in the movies. Considering the potential of Agentic AI, this is a big caveat. One must be conscious that these AI agents are driven by capable algorithms that help accomplish the task. At the same time, they can be your most dangerous enemies. So, agents must be employed with utmost care and diligence. When working with AI agents, you grant access to the repositories (folders and files), then you must ensure that personal data is kept in separate folders. You may keep a separate mailbox for your personal correspondence, as rogue agents may sneak into your mailbox and write emails to your friends and family without your permission. You may develop a habit of keeping sensitive information on a separate hard disk, which should be plugged in only when needed. Assigning a task to an AI agents requires great clarity. There are plenty of blog posts and YouTube tutorials on how and when to use the agents. One must adopt and learn good practices. Always remember that you are dealing with powerful agents powered by complex algorithms and that you have granted them access to your personal devices.

My favorite spy thriller movie series has come to an end with its last sequel, The Final Reckoning. We may not see Agent Ethan Hunt climbing walls and breaching data vaults, but I am sure there will be some new, more capable agents coming to screens soon. As it happens, we will be working with AI agents even more often than before, as they become part of our personal lives. The future looks promising but also precarious due to the possibility of encountering rogue agents. The biggest question is whether we are ready for it?

The writer is a professor at the National University of Sciences and Technology.