In a rapidly evolving AI landscape, transitioning from traditional generative AI models to agentic AI systems represents a pivotal moment for enterprises worldwide.

In theCUBE Research’s latest analysis, discussed in the “Next Frontiers of AI” podcast, SiliconANGLE and theCUBE’s George Gilbert joined theCUBE Research’s Scott Hebner, the podcast’s host. They provided critical insights for understanding and navigating the journey toward agentic AI.

“For 60 years, we’ve built software applications by hand-coding rules procedurally,” Gilbert said. “Agentic AI fundamentally shifts this paradigm, empowering software to learn the rules of business processes autonomously.”

Understanding agentic AI

Agentic AI represents a transformative leap, moving beyond the task-oriented nature of generative AI models toward goal-driven systems capable of autonomous decision-making and reasoning, according to Hebner. Unlike assistants rooted in generative AI, agents operate with clearly defined objectives, understanding the tasks and the broader context, reasoning paths and consequences of actions.

“Assistants anchored in generative AI are task-oriented: Give them a prompt, and they perform the task,” Hebner said. “But agents empower users to understand and achieve specific goals, modeling precise chains of events and outcomes.”

Climbing the ladder: The four rungs toward agentic AI systems

The transition to agentic AI systems isn’t a sudden leap, but rather a structured climb, captured metaphorically as the “ladder to agentic AI” by Hebner and Gilbert. They are:

1. Domain knowledge

The journey begins by grounding AI systems in specific domain knowledge, according to Gilbert. Enterprises must fine-tune large language models and integrate structured domain-specific data, enabling AI to understand the nuanced language and relationships unique to finance, healthcare or retail industries.

“You need domain intelligence beyond [large language models] alone,”  Gilbert noted. “Fine-tuning models with domain-specific data or integrating runtime intelligence through knowledge graphs is essential.”

2. Decision intelligence and explainability

The second rung integrates decision intelligence into AI models, enhancing their reasoning capabilities. Techniques such as chain-of-thought, semantic reasoning and causal AI equip systems to analyze cause-and-effect relationships and explain their reasoning transparently.

Explainability is the “currency of innovation,” according to Hebner. As AI decisions increasingly influence critical business outcomes, the ability to explain decisions is vital to trust and regulatory compliance.

“Explainability will become increasingly critical,” Hebner said. “If AI agents can’t clearly articulate the ‘why’ behind their decisions, businesses will struggle with trust and compliance issues.”

3. Building AI agents

With domain knowledge and decision intelligence established, enterprises must select appropriate platforms or integrated development environments to construct or acquire customized AI agents. There will be a growing marketplace where companies purchase specialized agents, later customizing them for their unique business needs, according to Gilbert.

“Building agents requires specialized tools,” Gilbert said. “Initially generic, these tools must evolve to incorporate decision intelligence, making customization straightforward and effective.”

4. Agentic AI systems and networks

The final rung connects individual agents into cohesive, goal-driven agentic networks. Such systems, exemplified by Amazon Web Services Inc.’s parent company, Amazon.com Inc., and its advanced supply chain operations, illustrate agents collaboratively pursuing complex, organization-wide goals guided by a unifying ‘north star,’ according to Gilbert.

“Amazon deploys an army of agents managing everything from logistics to staffing, all aligning toward the overarching goal of maximizing free cash flow,” he said. “This alignment of multiple agents — each with specialized tasks yet unified by strategic goals — is the essence of agentic AI.”

Continuous learning loop

The learning loop is a critical component of tying the ladder’s rungs together, according to Gilbert and Hebner. This loop ensures ongoing improvement by leveraging multi-agent reinforcement learning and feedback mechanisms, continuously refining agents and their underlying models.

“This learning loop will become a key competitive differentiator,” Hebner explained. “Enterprises that rapidly refine their AI assumptions and strategies will excel, adapting swiftly to emerging business realities.”

A strategic, incremental journey

Adopting agentic AI will be an incremental process, according to Gilbert and Hebner. Enterprises need not overhaul existing infrastructures overnight but instead progressively integrate agentic capabilities.

“The shift from generative to agentic AI will require strategic changes, but not rip-and-replace,” Hebner said. “It’s a journey, progressively enriching your AI stack to unlock measurable business outcomes.”

As enterprises move toward agentic AI systems, understanding this structured ladder and the continuous learning loop it supports will be essential. Those who embrace this evolution strategically will lead the next era of AI-driven transformation.

Photo: SiliconANGLE/DALL-E