Agentic AI: The next frontier of AI power

On 30 November 2022, OpenAI's release of Chat-GPT popularised Generative AI and provoked a storm of enthusiastic acclaim. Today, we are seeing an echo of that excitement in another related field with the rise of Agentic AI. And, although it has not met with the same degree of hype, Agentic AI may well be an even more powerful phenomenon. 

Some consider Agentic AI as stage three of five stages in the development of AI, sandwiched between conversational and reasoning as stages one and two and innovating and organisational AI as stages four and five. Looked at another way, Generative AI changed the way we create content, but the real excitement surrounding Agentic AI is its ability to automate complex processing tasks via the ability of multiple agents to autonomously learn, adapt, and make decisions to accelerate and automate decision-making and problem-solving.

At its core, Agentic AI embodies the ability to act within an environment and receive feedback—fundamental properties of an agent. This is achieved by leveraging Large Language Models (LLMs) as powerful reasoning engines that can send commands to existing tools via APIs, enabling Agentic AI to perform actions and learn through trial and error. Unlike traditional reinforcement learning, which starts from scratch without prior knowledge or reasoning, Agentic AI can bypass much of the extensive learning phase due to its inherent reasoning capabilities. When tasks become complex, multiple LLMs can collaborate as specialised agents in a role-playing dynamic, enhancing reasoning and leading to more accurate decisions. The long-term impacts of Agentic AI are broader and more significant than those of Generative AI, and when combined, they create sophisticated systems with vast potential for efficiency, precise information parsing, and smarter decision-making. Imagine using Agentic AI to make fast, auditable decisions and then utilising its own LLM to explain and communicate these outcomes effectively to partners, stakeholders, and employees.

Agentic AI will go everywhere

Agentic AI is transforming business operations by autonomously making decisions and taking actions across various applications.

More than just a tool for automating repetitive tasks, agentic AI unlocks new avenues for innovation and creativity. By shouldering operational burdens, it empowers employees to focus on strategic problem-solving, driving innovation, and delivering exceptional customer experiences. A prime example is CodeVista Liby FPT, an AI-powered platform that boosts software development productivity by 48%. It provides intelligent code suggestions, identifies patterns, and enhances coding standards, empowering developers to work smarter and deliver results faster.

Its ability to process vast datasets in real-time provides actionable insights faster than any human team, optimising critical functions such as pricing strategies, inventory management, supply chain logistics, and customer-facing interactions with unparalleled precision.

This impact is already evident across industries. For instance, FPT recently partnered with a German chemicals manufacturer to create multi-agent solutions that automated multiple processes within their business. For instance, the AI Agents automate IT ticket resolution, processing over 1,000 tickets daily and cutting response times. These agents transformed supply chain management, reducing purchase order response times from weeks to minutes. It also streamlines the HR process for onboarding and other admin tasks, as well as supports data retrieval and process assisting to speed up approval and resolution timeline. Additionally, this multi-AI agents solution helped a Japan-based multinational corporation automate data analysis, extract actionable insights, and enable seamless cross-cultural communication via automated language translation, driving strategic growth and informed decision-making.

Agentic AI's transformative power extends beyond back-end operations into a critical line of defence, detecting risks, responding instantly, and automating recovery to protect data and minimise downtime. For industries reliant on machinery, it predicts failures, reduces maintenance costs, boosts uptime, and extends asset lifespans. Agentic AI is reshaping businesses to be smarter, faster, and more resilient.

Perhaps a more pertinent question than 'what can Agentic AI do?' is 'what can't it do?' Certainly, as with AI generally, it's less well suited to tasks requiring human empathy, creativity, physical dexterity, and complex ethical judgements, but the applications we see even in the near term are vast.

What happens next?

Agentic AI is poised to revolutionise the digital realm by exchanging data, augmenting human intelligence, and automating increasingly complex tasks. Initially, it will excel in digital environments, leveraging its reasoning capabilities to handle sophisticated operations. Over time, Agentic AI will gradually transition into the physical world, empowering robotics to make sound judgments and operate effectively in complex physical settings.

The infrastructure supporting Agentic AI will evolve in two directions. On the one hand, it will scale up, consuming substantial GPU resources to manage extensive computations and data processing. On the other hand, it will become more specialised and efficient, enabling lean models with domain-specific knowledge to run on smaller devices. Low-level hardware optimisations, including AI-specific silicon designed for the unique computational demands of Agentic AI, will be crucial for accelerating its development and enabling large-scale deployment. Additionally, advancements in in-memory technologies and interconnects will support the massive data processing and communication needs of Agentic AI systems, ensuring optimal performance across diverse platforms.

It's also important to understand that there is a long road ahead for Agentic AI to sand off the rough edges that inevitably accompany any new technology. Improving Agentic AI requires addressing familiar AI ethical concerns like bias mitigation, transparency, and privacy. Technical advancements in learning algorithms, perception and hardware will be crucial. Additionally, fostering effective human-AI collaboration, including skill augmentation and continuous learning, is essential. By combining these elements, we can gain confidence in the safe and beneficial use of Agentic AI.

Building trust in Agentic AI is crucial. Transparency, robust security, human oversight, and user empowerment are key. Explainable AI (where nonspecialists can understand what AI is doing), clear data practices, and strong cybersecurity are essential. By prioritising these aspects, we can harness the power of Agentic AI responsibly and ethically.

Expect surprises…

AI is rapidly evolving through the five levels of AI, and one of the most intriguing aspects of Agentic AI is its potential for emergent behaviour. As AI systems become more complex, they can exhibit unexpected behaviours. This raises concerns about control and safety, making understanding and managing emergent behaviour a significant challenge in AI development.

It's also important to look at some of the more prosaic business adoption challenges of Agentic AI. Deploying it will be no free ride and early adopters can expect to have to justify large bills as they build necessary infrastructure and train models.

Finally, while we've explored the potential and challenges of Agentic AI, it's important to acknowledge that it's a rapidly evolving field with many unknowns. As AI continues to advance, we may encounter unforeseen ethical dilemmas and societal implications. As with any new and fast-moving technology, a thorough ability to look around the corner and anticipate issues and opportunities will be needed. But it's also important to recognise that the opportunities before us are very real and merit attention and action.