What is agentic AI?

Capacity breaks down AI’s newest buzzword, what it means and its impact on the telecom sector, and beyond.

Nvidia describes agentic AI as:

“Us[ing] sophisticated reasoning and iterative planning to autonomously solve complex, multi-step problems.”

Red Hat defines agentic AI as:

“Agentic AI is a software system designed to interact with data and tools in a way that requires minimal human intervention. With an emphasis on goal-oriented behaviour, agentic AI (also known as AI agents) can accomplish tasks by creating a list of steps and performing them autonomously.”

Put simply, agentic AI is a fancy way of saying automation of a particular application by an AI system tasked with performing a specific task.

Agentic AI should not be construed with so-called ‘artificial general intelligence’ or AGI, where an AI can think for itself — a concept straight out of science fiction.

Agentic AI, rather, is the concept of an agent-based AI system performing assigned tasks on behalf of a user or another system.

Agentic AI emerged as a concept in 2024. This new paradigm sees a shift from AI systems that simply offer recommendations, such as assistant tools, to those that can autonomously make decisions and take action.

Agentic AI isn’t just about large language models, instead combining different AI techniques, models, and approaches to create autonomous agents capable of analysing data, setting goals, and taking action with minimal human supervision.

It’s quickly forced its way to the front of conversations about AI. Sam Altman, CEO of OpenAI, suggested that agentic AI will be integrated into our daily lives by 2025.

Meanwhile, analyst firm Forrester named AI agents as one of its top 10 emerging technologies, predicting that it will deliver sizable benefits within the next two to five years. Gartner also listed agentic AI as a top strategic technology trend for 2025.

Agentic AI is now everywhere — Both AMD CEO Lisa Su and Nvidia CEO Jensen Huang have talked at length about agentic AI during keynote speeches this year, while Andrew Ng, the machine learning pioneer and Google Brain co-founder, said agentic AI was the technical trend he’s “most excited about” and the “most important AI technology to pay attention to”.

Agentic AI was even front and centre of the reveal of Google’s new flagship AI model: Gemini 2, with the search giant stating the model was specifically designed “for this new agentic era”.

Agentic AI differs from traditional AI in its autonomy, goal-setting, and interaction with the environment.

Capacity breaks down the distinctions:

Traditional AI: Typically follows pre-programmed rules or optimises within a fixed framework. It’s task-specific and relies on predefined parameters or direct human instructions.

Agentic AI: Operates with a degree of independence and is capable of pursuing long-term objectives that may not require constant human oversight. Agentic AI can decide how to achieve its goals, adapting dynamically to new situations without straying beyond set behaviours.

Traditional AI: Functions more passively, processing inputs and providing outputs within static preset boundaries. It doesn't independently influence or adapt to its environment in the same way.

Agentic AI: Acts as what is effectively an autonomous agent in its environment, capable of perceiving, learning, and taking actions. It can experiment and adjust its behaviour based on feedback or changes in its surroundings.

Traditional AI: Excels at single, narrowly defined tasks, for example, image recognition or text generation and translation but typically struggles with integrating multiple objectives or handling unforeseen complexities.

Agentic AI: Can manage complex, multi-step tasks by breaking them down, prioritising, and sequencing them efficiently, more akin to human thinking — just don’t view it as AGI.

Traditional AI: Learning is usually confined to training phases before deployment, and updates often require manual intervention through a new underlying model or dataset revision.

Agentic AI: Often incorporates reinforcement learning or other techniques to improve its performance over time without explicit reprogramming, instead learning from its actions and the resulting consequences.

Traditional AI: Suited for repetitive, highly structured applications such as data analysis, recommendation systems, or automated customer service.

Agentic AI: Ideal for dynamic systems such as robotics, autonomous vehicles, and advanced simulation environments, where adaptability and self-direction are critical.

As Agentic AI is suited for more dynamic applications, use cases for agentic AI include:

Inventory management: Forecasting demand, automating restocking, and adjusting supply chain operations in response to disruptions like shipping delays or demand spikes.

Personalised shopping experiences: Agentic AI tailors recommendations and promotions by autonomously analysing customer behaviour and preferences.

Smart Factories: Agentic AI coordinates robotics systems to adapt to changing production requirements, ensuring efficiency and reducing waste.

Quality Control: It monitors production lines in real-time, identifying defects and making adjustments to minimise errors.

Medical Diagnostics: Agentic AI assists doctors by autonomously analysing patient data, identifying potential issues, and suggesting personalised treatment plans.

Hospital Operations: It optimises patient flow, staffing, and resource allocation in real time to improve efficiency and care quality.

Autonomous Vehicles: Agentic AI enables vehicles to navigate complex environments, adapt to traffic, and make real-time decisions for safer travel.

Route Optimisation: It plans and adjusts delivery routes dynamically to minimise delays and fuel costs.

Smart grids: Agentic AI balances energy supply and demand, integrating renewable energy sources while reducing waste.

Predictive maintenance: It ensures critical infrastructure, such as pipelines or power plants, operates efficiently by pre-emptively addressing potential issues.

Fraud detection: Agentic AI identifies unusual patterns in transactions, autonomously investigating and flagging potential fraud.

Portfolio management: It dynamically adjusts investment portfolios based on market trends and risk tolerance.

Vinay Samuel, CEO and Founder of Zetaris, said agentic AI is a “major game changer for any industry”.

Samuel explained: “It is the extension of generative AI that executes tasks, has a feedback loop to learn, and can interact with humans and mimic human behaviour. Front-line complex decision support and processing roles are prime candidates for disruption. We are also working on projects that deploy agentic AI from the board level into the leadership team.

“With agentic AI a board can have a virtual board member that can access, analyse and interpret thousands of documents in real-time. This may result in a golden age of compliance and corporate governance.”

Agentic AI can revolutionise telcos across a variety of ways:

Network optimisation: Agentic AI can analyse network data, identify bottlenecks, and dynamically allocate resources to ensure optimal performance.

Predictive maintenance: AI agents can monitor network equipment, predict potential failures, and proactively schedule maintenance, reducing downtime and costs.

Fraud detection: AI agents can analyse patterns, detect anomalies, and flag potential fraud, improving security and protecting revenue.

Resource allocation: Agentic AI can optimise resource allocation, such as bandwidth, to meet varying demands and ensure service quality.

Sumeet Arora, CDO of ThoughtSpot, told Capacity that in the telecom and cloud markets, agentic AI is helping drive advanced capabilities in customer-facing roles, such as AI-powered customer service assistants capable of handling customer inquiries autonomously and customer service optimisation.

“A telecom customer uses ThoughtSpot to identify fraudulent activities, this is another common use case for agentic AI in telecom and cloud services, " Arora explained.

Agentic AI sits front and centre of many AI industry players' minds. With the likes of Google focusing its next-gen flagship AI model on agent-based AI, expect it to be a major topic going into 2025, and beyond.

“Looking ahead, agentic AI will evolve from a tool that supports decision-making to

becoming a primary driver of actionable insights and autonomous operations,” Arora said. “The true potential of agentic AI lies in its ability to not only interpret data but also act upon it in real-time, driving business outcomes and operational efficiencies.

“With AI acting as both a strategic and operational partner, organisations will have the capability to make faster, smarter decisions with less reliance on human intervention. Rather than replacing workers, agentic AI will empower them - augmenting their decision-making abilities and enabling them to focus on creative problem-solving and innovation. This will result in the vision of autonomous business, where AI acts as a primary catalyst for growth and transformation.”

Nvidia: What Is Agentic AI? 

IBM: Agentic AI: 4 reasons why it’s the next big thing in AI research

Andrew Ng & AutoGen: AI Agentic Design Patterns with AutoGen

Datacloud trends to watch in 2025: What's next in the digital landscape

Salesforce partners with Google Cloud to launch autonomous AI agents

AI startup gold rush: Inside the high-stakes world of AI investments

Networks must transform to support AI era, says Huawei exec