Era Of Autonomous AI And Digital SideKicks: What Are AI Agents And Why Should You Care?

The rise of Agentic AI, or AI agents, marks a shift towards autonomous systems capable of solving real-world problems with minimal human intervention

Hyderabad: The public release of ChatGPT in December 2023 marked a turning point for the tech industry as it redirected its attention from blockchain to artificial intelligence, generative AI, and conversational chatbots. In no time, AI became the only thing everyone could talk about as major tech players raced to harness its potential. Since then, AI has continued to evolve, becoming better, more efficient, and expanding its capabilities, which now include an agentic nature.

The Agentic AI or AI agents are sort of like personal assistants that can perform a series of tasks to achieve a specific goal without requiring constant human supervision. They act independently and have kind of an ‘agency’ that allows them to make decisions, solve problems, and adapt their actions based on the environment to achieve the goal.

Abhishek Singh, co-founder and CEO, SecureDApp, considers AI agents as smart software that act as digital sidekicks, handling tasks from ordering groceries to tracking stocks. "Picture them as tireless workers who learn as they go," he explains.

One fine example of Agentic AI is a self-driving car that assesses its surroundings, predicts traffic behaviour, and plans a route to reach the designated area without requiring human assistance. Instead of executing pre-programmed instructions, the Agentic AI works by evaluating and responding to situations in real time to reach its goal.

Rise of Agentic AI

Terms like “Agentic AI” and “AI agent” have been continuously gaining traction for the past year and currently sit at an all-time high on the Google Trends graph for both web searches and YouTube searches. The reason behind this sudden rise could be attributed to the adoption of AI agents by tech giants, such as Google, Microsoft, Anthropic, Salesforce, Amazon, IBM, and more.

Google recently announced its first AI agent, called Project Mariner, which can be used to find flights and hotels, shop for household items, find recipes, and do other tasks. Microsoft’s Copilot Agents automate customer service, administrative workflows, and data analytics following their integration with the Microsoft 365 enterprise suite.

Perplexity, last year, released an AI agent that helps people do their holiday shopping by navigating retail websites, finding products, and clicking the checkout button. OpenAI’s upcoming Operator will autonomously tackle complex workflows like web navigation. Meanwhile, Anthropic’s Claude mimics human computer interactions to automate repetitive desktop tasks, such as navigating software, clicking buttons, and browsing the web.

While OpenAI CEO Sam Altman expects agents to join the workforce in 2025 and Microsoft CEO Satya Nadella predicts agents to replace certain knowledge work. SalesForce CEO Marc Benioff aims to become “the number one provider of digital labour in the world” via the company’s various agentic services.

AI agent: The identity dilemma

While the industry is rushing to compete for AI agents, it is facing some problems in agreeing on a single definition. From interchanging “assistants” and “agents” to defining agents as “automated systems that can independently accomplish tasks on behalf of users”, only to switch it to “LLMs equipped with instructions and tools”, OpenAI alone is struggling to keep the definition of agents uniform.

Microsoft calls agents “new apps” for an AI-powered world that can be tailored to have a particular expertise. Meanwhile, it says that AI assistants are those which help with general tasks like drafting emails.

Anthropic defines agents as something that can be defined in several ways, calling them “fully autonomous systems that operate independently over extended periods” and “prescriptive implementations that follow predefined workflows”. Meanwhile, Salesforce calls them a system that can understand and respond to customer inquiries without human intervention.

The chaos surrounding AI agents arises from their evolving nature and the absence of a standardised definition. Companies like OpenAI, Google, and Perplexity are launching varied "agents," but their capabilities differ significantly. While the lack of a standard definition allows customisation, it also results in varied interpretations of what AI agents should deliver, giving birth to misaligned expectations.

While the industry seems unlikely to settle on a unified definition of AI agents or Agentic AI anytime soon, we can look at them as AI-powered digital workers designed to perform multi-step tasks autonomously by interacting with their environment, such as apps, websites, and data systems. They can have additional functionalities and technologies working behind the scenes, based on the kind of work they are required to do. Singh defines these systems as "self-directed programs that use artificial intelligence to analyse information, learn, and take actions automatically, like a virtual assistant that can think and act on its own".

Technologies behind Agentic AI

Agentic AI systems are powered by a combination of technologies, such as machine learning (ML), deep learning (DL), and reinforcement learning (RL). Based on their use case, Agentic AI can also rely on other technologies like computer vision, sensor integration, natural language processing (NLP), and more. Together, these technologies create intelligent systems that act as digital workers, capable of handling multi-step tasks with minimal supervision. Let’s take a detailed look at these technologies:

Machine Learning (ML) allows systems or AI agents to make decisions based on previous experiences. Using ML, OTT platforms recommend content, smartphone keyboards suggest the next word, and email services filter spam. Deep Learning (DL) is a subset of ML that uses neural networks to mimic the way the human brain processes information. DL can pick out relevant information from the raw data. It is the reason why voice assistants know that “turn up the heat” isn’t about cooking. DL is used for identifying objects in photos (self-driving cars spotting stop signs), speech recognition (turning your voice into text for Siri or Alexa), and other similar tasks.

Reinforcement Learning (RL) is also an ML approach where an agent learns optimal actions by trial and error in an environment to maximise rewards. It teaches systems to make decisions through feedback and rewarding correct actions. A robotic vacuum cleaner uses RL to navigate inside a house by trial and error.

Natural Language Processing (NLP) enables AI agents to understand, interpret, and generate human language. It powers chatbots, translation engines, and voice assistants. Speech Recognition and Generation allows a system to understand spoken language and respond with synthesised speech. Coupled with NLP, it lets AI agents interact with humans via spoken language.

Computer Vision allows them to analyse images and videos for object recognition and other use cases. AI agents use this as their eyes, allowing them to see their environment and adapt to it. Sensor Integration lets systems combine data from cameras, LiDAR, GPS, gyroscopes, and other installed sensors to create a detailed understanding of the environment. Drones using GPS and smart home systems using motion sensors are examples of sensor integration. Based on their function, they can also utilise Knowledge Graphs, APIs, Cloud Computing, and other technologies as per their needs.

Singh explains the working of AI agents based on how they think, process data, and take action. "AI agents rely on three core technologies where AI models act as the brain, databases store memories like past actions, and APIs connect to apps, letting them pull flight prices from Expedia or check your calendar," Singh says.

AI agents and jobs

Real-world examples of Agentic AI include self-driving cars, AI-powered drones, industrial robots used in manufacturing, virtual personal assistants, advanced healthcare systems, and customer support chatbots. Each of these requires a different set of technologies to function the way they are designed to work.

Companies around the world have been going all-in on AI agents. The technology is changing the way machines work and has the potential to revolutionise industries. As artificial intelligence systems become more advanced and ‘intelligent’ in the future, AI agents are expected to become more sophisticated and efficient in what they do– solving real-world problems without constant human intervention.

Agentic AI solutions currently available in the market aim to reduce the human burden by taking care of boring, repetitive tasks. However, there is a possibility that more advanced agentic AI systems could eliminate a lot of such jobs. As tech giants race to become "providers of digital labour", countries known for providing cheap human labour could find themselves in a pinch as AI agents could affect offshoring to cheaper labour markets.

Several countries rely on labor-intensive industries like manufacturing, textiles, and low-skill services (like call centres) as a competitive advantage in the global economy, but the rise of Agentic AI could reduce this advantage. These digital sidekicks can automate a wide range of tasks, from repetitive manual labour to basic cognitive work, which directly competes with the low-cost human labor that such countries provide.

Experts at the Tony Blair Institute for Global Change noted that AI's potential to save time and enhance productivity would vary significantly across individual workers. They explained that most time savings from AI were likely to stem from AI-enabled software capable of performing cognitive tasks, rather than more expensive AI-enabled hardware designed for physical tasks. Consequently, they suggested that occupations and sectors involving complex manual work, such as skilled trades or construction, would be less impacted by AI. In contrast, workers engaged in routine cognitive tasks, especially in administrative roles or data-intensive industries like banking and finance—where it is easier to train new AI models—were predicted to be more exposed to AI.

Meanwhile, Abhishek Singh thinks that such systems won’t kill jobs but reshape them just like the Industrial Revolution did. "Expect three shifts: routine gigs like data entry may fade, jobs like medicine or design get a boost from AI tools, and new roles, think AI trainers or ethicists ensuring fairness, will pop up," he says. Singh thinks that it is hard to picture wildcards like ‘AI agent managers’ yet.