At the beginning of the year, AI agents were widely tipped to be the defining breakthrough of 2025. Six months in, that prediction appears to be gradually taking shape as OpenAI, Anthropic, Microsoft, Google, and several other tech companies have launched products and features based on AI agents that are designed to autonomously complete tasks under minimal supervision by humans.
Besides building AI tools that go beyond text and image generators, several companies are also adopting agentic artificial intelligence (AI) to automate enterprise workflows. A recent survey by Big Four accounting firm EY found that out of 500 tech industry executives in the US, nearly 48 per cent of them are already deploying AI agents within their companies.
Half of the respondents said that more than 50 per cent of AI deployment will be autonomous in their company in the next two years, as per the report. As agentic AI continues to gain momentum, here is what you need to know to help make sense of what is ahead.
What is an AI agent? How is it different from an AI chatbot?
An AI chatbot like OpenAI’s ChatGPT is used to generate text and images. It can communicate with users based on prompts submitted by them. These chatbots are powered by large language models (LLMs) that are trained on vast amounts of data to generate text, images and videos.
On the other hand, an AI agent is more flexible than a chatbot as it can interpret complex commands and trigger various actions on its own. The foundational LLMs help decide what actions the AI agent should take. These actions depend on the type of AI agent as well. While a web browsing agent can handle tasks such as searching the internet, booking travel tickets, and making online purchases, a coding agent is designed to navigate codebases, retrieve snippets of code, and even generate or debug code.
These agents are developed in a way that they do not need human input at every step. A human can input a command or prompt and just sit back. But they will likely need to stay in the loop and monitor the AI agent’s actions in order to be able to intervene if needed.
Tech companies developing AI agents claim that they will enable humans to be hyper-productive.
Story continues below this ad
Who are the main players building AI agents?
Companies ranging from big tech giants such as Nvidia, Microsoft, Amazon, and Google to well-funded startups like OpenAI and Anthropic are focused on building AI agents.
Earlier this month, Amazon announced that it is creating a new unit within its hardware research-and-development unit that will focus on developing an agentic AI framework to be integrated into the e-commerce major’s robots or physical AI systems.
The Jeff Bezos-founded company is expected to have more of an edge in developing AI agents because it has access to specialised data that can be used to train these agents in how humans navigate and shop on Amazon’s website.
Startups like Cursor are also developing AI coding agents. In fact, over 70 startups incubated by San Francisco-based startup accelerator Y Combinator as part of its spring 2025 batch are reportedly focused on agentic AI.
Story continues below this ad
The rise of agentic AI has made its way to India as well. Ola founder Bhavish Aggarwal’s AI venture, Krutrim, recently launched a new agentic AI app called Kruti which is capable of autonomously booking cabs and ordering food on the Ola platform. It further has plans to enable ride bookings and food delivery orders on rival platforms such as Uber, Zomato, Swiggy, etc.
How are AI agents being used today? Will their role evolve in the future?
One of the biggest use cases for AI agents is said to be customer service. According to a report by market research firm Gartner, over 80 per cent of common customer service queries will be resolved by AI agents in the next four years.
Currently, AI agents being deployed by tech companies are capable of surfing the web, calling a restaurant to make reservations, or fulfilling routine tasks in a Microsoft Office environment. AI agents also have immense potential in the field of software development which has, in turn, sparked fears that they could lead to the elimination of coding-related jobs. They could also handle a lot of repetitive back-office work such as filing invoices.
As AI moves toward a major platform shift into hardware, it would be interesting to see the role that AI agents play when integrated with physical devices. The idea that AI agents could potentially end up automating a lot of mundane chores may also bring us one step closer to achieving artificial general intelligence (AGI).
Story continues below this ad
What are the limitations of AI agents?
Despite their advanced capabilities, AI agents are just as prone to hallucinations and misaligned, unpredictable behaviour as chatbots since they are essentially powered by the same LLM technology.
The high cost of computing power needed for AI agents to operate autonomously is another drawback. For instance, OpenAI could charge up to $20,000 per month for access to its specialised AI agents that will supposedly be able to perform tasks with the same level of expertise as a PhD graduate, according to a report by The Information.
Given their autonomy to take actions within a user’s system, AI agents also pose a new security risk. If compromised, AI agents could be used by attackers to steal information and carry out other malicious activities.
