Agentic AI - A leap forward in the evolution of AI

"AI is no longer merely a tool but is becoming an active agent capable of making decisions and taking actions. As a result of LLMs [...] AI no longer shows up just as a tool. It shows up as an agent."

This quote is from Dan Kraemer, professor at the Northwestern University - Kellogg School of Management , during his opening keynote on Generative AI and the importance of Agentic systems at the last Innovation Roundtable? summit in Copenhagen.

In this article, we will talk about Agentic AI, which represents a significant leap forward in the evolution of artificial intelligence, promising to revolutionise both personal and professional landscapes. This shift from Artificial Intelligence as a passive instrument to Artificial Intelligence as an Autonomous actor signifies a profound transformation in the way we interact with and leverage technology.

The Historical Trajectory of Agentic AI

Let's start with a bit of historical context. The development of Agentic AI has been a gradual process, spanning several decades and marked by distinct stages.

• 1950s: Rule-Based Agents. These early agents were designed to follow a set of predefined rules and logic. They operated in a very rigid manner, executing tasks based on specific instructions given by programmers. These 'agents' were not capable of learning or adapting to new situations; they could only perform tasks within the scope of their programming. Examples include early expert systems used in medical diagnosis and simple decision-making processes. However, during this time, the idea of 'agents' was still largely theoretical.
• 1980s: Predictive Agents. Predictive agents emerged with the ability to leverage historical data and probabilistic models. These agents focused on forecasting future trends and outcomes based on past information. They used statistical techniques to predict events such as market trends, weather patterns, and consumer behavior. This era saw the rise of more sophisticated algorithms that could handle larger datasets and provide more accurate predictions.
• 1990s: Analytical Agents. Analytical agents were developed to process both structured and unstructured data. They aimed at extracting valuable insights from large datasets, which included text, images, and other forms of data. These agents utilized advanced data mining and machine learning techniques to analyze data, identify patterns, and generate actionable insights. They were widely used in fields like business intelligence, healthcare, and finance.
• 2000s: Autonomous Agents. Autonomous agents were designed to operate with predefined goals and real-time input. These agents could make decisions and take actions in dynamic environments without human intervention. They were capable of learning from their experiences and adapting to new situations. Applications included autonomous vehicles, robotic systems, and intelligent personal assistants. These agents marked a significant step towards more independent and self-sufficient AI systems.
• 2014-Present: Generative Agents. Generative agents represent the current cutting edge of agentic AI. They leverage advanced technologies such as Large Language Models (LLMs), Generative Adversarial Networks (GANs), and Latent Action Models (LAMs) to create new content and strategies. These agents can generate text, images, music, and even complex strategies for various applications. They are used in creative industries, content generation, and strategic planning, showcasing the potential of AI to not only analyze and predict but also to create and innovate.

This historical overview shows how Agentic AI has evolved from simple rule-based systems to today's sophisticated generative agents. Early AI was limited to performing basic tasks, but advancements in machine learning and deep learning have led to the development of more adaptive and intelligent systems. Today, AI can engage in natural conversations, generate realistic content, and provide personalized recommendations, showcasing the remarkable progress in AI technology.

Generative AI: The driving force behind current Agentic AI Advancements

The current excitement surrounding Agentic AI is largely attributed to the emergence of Generative AI, powered by technologies like LLMs, GANs, and LAMs. This has led to a surge in investment in Generative AI, as well as the development of numerous AI tools.

Generative AI has the potential to revolutionise various aspects of our personal and professional lives. Generative AI agents are transforming the way we interact with technology, enabling more personalised and engaging experiences. Examples include AI-powered video generation, which can create custom video content for marketing or entertainment; personalised shopping recommendations based on individual preferences and browsing history; and task automation agents that streamline daily activities by managing schedules, sending reminders, and handling routine tasks like email sorting. Additionally, AI-driven content creation can produce articles, reports, and social media posts, while virtual assistants offer real-time support and information.

Generative AI is also poised to revolutionise businesses across numerous sectors. For example, AI agents can enhance productivity in areas such as sales by analysing customer data to generate insights and predict trends; in recruiting, where AI screens resumes and conducts initial interviews; and in customer service, where chatbots and virtual assistants handle inquiries and provide support. In healthcare, AI can assist in diagnosing diseases and personalising treatment plans, while in finance, AI-driven analytics optimize investment strategies and detect fraud.

The potential applications of Generative AI are vast and continually expanding.

Navigating the Potential and Limitations of Agentic AI

While the potential of Agentic AI is vast, it is essential to recognise the limitations of current AI capabilities. Artificial Intelligence excels in tasks within its competency range, but can underperform if users over-rely on it for tasks that exceed its current capabilities.

A Harvard Business School study titled "Navigating The Jagged Technological Frontier" sheds light on this dichotomy. The study identifies tasks where AI can significantly boost productivity and quality, such as idea generation, writing and persuasion, analytical thinking, and structured processes. These tasks generally fall "within the frontier" of current AI capabilities. However, the study also cautions against over-reliance on AI for tasks that go "beyond the frontier" of its current abilities. These tasks, which often involve complex problem-solving, subtle insights, judgment-based recommendations, and multifaceted data analysis, can suffer from over-dependence on AI.

Dan Kraemer suggests that "Agentic systems, by combining multiple AI agents with diverse capabilities, might offer a pathway to overcome these limitations". The collaboration and interplay between these agents could potentially enable the tackling of more complex and nuanced problems, thus pushing the boundaries of AI's capabilities even further.

The shift toward an Agentic Future: Implications for Consumers and Businesses

Let’s imagine a future where Agentic AI plays a central role in our lives, from being a health coach to offering a personalised travel assistant. This shift toward an Agentic future would have profound implications for both consumers and businesses:

1. Consumers will experience a more personalised and streamlined internet. Instead of actively searching for information and making decisions, they will rely on their personal AI agents to curate and manage their online experiences. This vision is exemplified by Deutsche Telekom AG CEO Tim H?ttges , who imagines a future where apps and websites become obsolete, replaced by a personalised, agent-driven experience (full video here).
2. Brands will need to adapt to a world where they interact with customers primarily through AI agents. In this future, the brand experience will be shaped by the capabilities and personality of the brand's AI agents. This will require businesses to rethink their digital strategies, focusing on designing compelling and effective AI agents that can represent their brand values and deliver exceptional customer experiences.
3. Businesses will need to adopt "AI-in-the-loop" workflows. This involves integrating AI agents into their internal processes to optimise operations, enhance productivity, and drive innovation. Following this principle developed by Dan Kraemer, businesses should start by identifying their most critical and recurring decisions and then explore how AI agents can be incorporated into the workflows that support these decisions. (see template here)

Now let’s imagine a future with Agentic AI systems in the loop.

Example: The Future of Personalised Journeys

Let’s imagine the future of travel with Agentic AI systems. Gone are the days of scouring countless travel websites, juggling flight options, and coordinating hotel bookings.

Instead, imagine having a personal AI travel agent who understands your needs, preferences, and budget. This agent could seamlessly collaborate with a network of specialized AI agents representing airlines, hotels, and transportation services, crafting a personalized itinerary tailored to your desires.

Imagine:

• Whispering your travel aspirations to your personal AI agent. Maybe you yearn for a culinary adventure through Southeast Asia or a tranquil escape amidst Ireland's Connemara lakes.
• Your AI agent then springs into action, consulting with other agents to gather information, compare prices, and secure reservations.

This intricate dance between agents unfolds behind the scenes, presenting you with a cohesive and optimised travel plan.

Think about the power of real-time responsiveness in these agentic systems. Your AI travel agent would not just create a static itinerary; it would continuously adapt to your needs, preferences, and external factors. For instance:

• If your flight is delayed, your AI agent could proactively rebook connecting transportation, adjust hotel check-in times, and even notify your dinner reservation about the change.
• It could suggest alternative activities or destinations based on real-time weather conditions or local events.

This dynamic responsiveness would ensure a smooth and stress-free journey, allowing you to focus on fully experiencing your chosen destination.

As per Kraemer’s recommendation four ?crucial aspects need to be considered to design agent’s workflows and ensure the effectiveness of such agentic systems.

• Reflection: Enabling travel agents to evaluate their performance and identify areas for improvement is essential. This could involve analyzing user feedback, comparing travel plans with alternatives, and continuously learning from experiences. Automated performance assessments could help identify patterns, allowing agents to refine their decision-making and improve the accuracy of future recommendations.

• Tool Use: Providing travel agents with access to a diverse set of tools for gathering data, processing information, and taking action is vital. In our example, these tools could include APIs for real-time flight and accommodation data, natural language processing to interpret user requests, and machine learning algorithms to predict user preferences. By integrating these tools, agents can offer timely, personalised recommendations and solutions.

• Planning: Equipping travel agents with the ability to devise and execute multi-step plans to achieve desired outcomes is critical. In our example, this might involve breaking down a complex travel request into smaller tasks, assigning those tasks to specialised agents, and coordinating their actions to ensure a cohesive plan. Additionally, proactive planning capabilities, like anticipating travel disruptions (e.g., weather changes, flight delays), can help maintain a smooth travel experience.

• Multi-agent Collaboration: Fostering effective communication and collaboration among agents is key to creating optimal travel experiences. This could involve agents negotiating prices, sharing information, and resolving conflicts (such as prioritizing customer satisfaction over cost-saving decisions). By ensuring a seamless flow of information and decision-making across agents, a more personalized, efficient, and stress-free travel experience can be achieved.

By carefully considering these aspects, we could develop Agentic AI systems capable of revolutionising the travel industry, making travel planning more personalised, efficient, and stress-free.

Conclusion: Embracing the Agentic Revolution

Agentic AI is not just a technological advancement; it is a transformative shift in the way we interact with technology, one that will profoundly reshape both our personal lives and business landscapes. As we reflect on its historical trajectory, we recognise how far we have come, from rule-based systems to the sophisticated generative agents of today.

The journey of Agentic AI is a testament to the remarkable potential of AI to evolve from a passive tool to an autonomous agent, capable of revolutionising industries and everyday experiences.

To fully embrace the Agentic Revolution, businesses and individuals must understand both the opportunities and limitations of AI. By recognising the potential of generative agents, businesses can optimise operations, enhance personalisation, and deliver unprecedented value to their customers. But, as with any revolutionary shift, this future requires us to think ahead, adopt "AI-in-the-loop" workflows, and design AI agents that truly embody our brand values and customer needs.

“AI agent workflows will drive massive AI progress this year — perhaps even more than the next generation of foundation models” (Andrew Ng)

The future is fast approaching, and it is time for companies to act. The power of Agentic AI lies not only in its capacity to innovate but also in its potential to redefine how we connect, collaborate, and create. By proactively shaping this future - whether it’s in the travel industry, healthcare, finance, or beyond - we can build a world that is more efficient, personalised, and fulfilling.

Now is the time to prepare for the Agentic AI era, where smart, ethical autonomous agents will enhance our decisions, optimise our operations, and ultimately improve the quality of our lives and businesses. Are you ready to embrace the Agentic Revolution?

Sources:

• Dan Kraemer keynote and agentic AI toolkit at the Innovation Roundtable 2024 available on: https://iacollaborative.com/events/innovationroundtable/
• "Navigating the jagged technology frontier", Harvard Business Review https://www.hbs.edu/ris/Publication%20Files/24-013_d9b45b68-9e74-42d6-a1c6-c72fb70c7282.pdf
• Andrew Ng on AI agent workflows: https://www.deeplearning.ai/the-batch/issue-241/