AI Agents will lead the next wave of Generative AI adoption, with one-fourth of early adopters expected to implement AI Agents in 2025, according to Deloitte. Is the attention on automation justified by the value creation opportunities it offers, or should IT leaders look beyond automation to exploit the full potential of Agentic AI?
The foundation for AI Agents was established in the early 2000s as part of Agent Oriented Software Engineering. Agents are characterized by their ability to plan and act autonomously to achieve a goal. They can adapt to changes in their environment and learn from feedback. Agents can also collaborate socially with human users and other agents in a multi-agent system. The advent of Large Language Models with the ability to mimic human reasoning has turned industry attention towards AI Agents.
Agentic AI Value Pyramid
Despite enormous advances in the last two years, Generative AI (GenAI) is still an evolving technology. Challenges such as AI hallucination could be managed to tolerable levels through better LLM training and Prompt Engineering. Effectively managing AI ethics, data protection, and AI security requires comprehensive AI governance policies, planned and implemented at organizational levels.
However, these challenges have not stopped GenAI front-runners from moving ahead with pilot implementations. There is a pattern emerging from these early implementations, highlighting the contours of an Agentic AI Value Pyramid. The Agentic AI Value Pyramid has three levels — Augmentation, Automation, and Ideation.
Augmentation
Agentic AI can help improve the speed and confidence of performing complex tasks such as analysis and decision-making. AI Agents can automatically retrieve real-time data, perform analysis, and generate reports to aid decision-makers.
An AI Agent developed at Fujitsu Kozuchi observes meeting discussions. When there is a conflict or confusion about a fact, the agent fetches supporting information from related external systems and presents them to the meeting participants, hence augmenting human decision-making ability.
Augmentation can also lead to better customer-friendly interfaces, as we can see in Amazon’s summarization of review comments for products. Instead of reading individual comments, customers can view a summary of the review comments generated by AI, to get an idea about the pros and cons of a product.
Augmentation can improve employee productivity and customer experience, leading to better retention for companies. It can cover the widest array of business activities, forming the base of the value pyramid. A survey by LinkedIn and Microsoft found that an estimated 46% of workers feel burnout at work. The same survey found that GenAI helps 85% of users to focus on their core tasks. Even as the productivity gains of Agentic AI augmentation are difficult to quantify, it is a key enabler that we cannot ignore.
Automation
Automation is getting the most attention in Agentic AI, especially with the current focus on GenAI’s return on investment. Automation is a low-hanging fruit with measurable impact. A study by McKinsey estimates that Generative AI can accelerate the pace of automation, making it possible to automate activities that make up 60% to 70% of work time.
Organizations that have established processes and governance for automation practices such as Robotic Process Automation (RPA) have an edge in implementing Agentic AI automation. RPA can automate repetitive well-defined workflows. Unlike AI Agents, RPA requires explicit step-by-step programming. There is no room for ambiguity or exceptions. The operational processes and governance policies designed for RPA can be adapted for AI Agents with AI-specific updates.
Figure 1: Agentic AI Value Pyramid
In IT Operations and Security, AI Agents can monitor metrics, analyze logs, and take proactive actions. Although this deployment model looks basic, it has enormous potential in business functions such as Manufacturing, Logistics, and Retail when integrated with the Internet of Things.
Automation forms the middle level of the Agentic AI Value Pyramid, with the potential to automate most non-core and support functions in organizations. Apart from cost savings, automation can also improve processing speed and minimize errors.
Ideation
While automation and augmentation may improve organizational agility and productivity, innovation in core competencies is a key factor that directly translates to growth and differentiates leaders from followers. Augmentation can support innovation by providing actionable insights, but the velocity and quality of idea generation still depend on the availability of human expertise.
Utilizing AI Agents for ideation may accelerate innovation by reducing dependency on human expertise. Automation and augmentation agents are powered by off-the-shelf LLMs like GPT4o. Ideation typically requires LLMs trained or fine-tuned on domain-specific data, in fields such as biology, chemistry or finance.
Consider the case of BloombergGPT, a GenAI model developed at Bloomberg for the finance domain. Training in a wide range of financial data has optimized BloombergGPT for the finance domain. Bloomberg claims that BloombergGPT’s performance in financial tasks is outstanding when compared with off-the-shelf LLMs of the same generation. In pharmaceutics, Iambic’s Enchant model for drug discovery and development has enabled a reduction of clinical risks at discovery stage.
Domain-specific LLMs require significant investment for building and maintenance, but they can clone expertise that is costly to acquire or replicate otherwise. Use cases such as Enchant show the potential of ideation agents built with domain-specific LLMs to accelerate innovation and reduce the time to market.
AI Agents will improve organizational agility and competitiveness. They will lower costs and improve customer experience. But with the looming commoditization of AI Agents, early adopters may not be able to sustain the competitive advantage for long. Organizations that are ready to adopt all three levels of the Agentic AI Value Pyramid — Augmentation, Automation, and Ideation will lead the next generation of AI enabled enterprises.
Shanmugam Sudalaimuthu is a software architect with more than 20 years of experience building innovative solutions for Fortune 500 companies across diverse industries. He specializes in Generative AI and Cloud technologies, and holds a degree in Physics and a master’s degree in Computer Applications.
