Enterprises looking to harness the power of agentic AI need to rethink how technology interacts and delivers value, according to Bartley Richardson, senior director of engineering and AI infrastructure at NVIDIA.

richardson recently appeared on the NVIDIA AI Podcast to share insights on how organizations can effectively implement agentic AI systems.

“when I talk with people about agents and agentic AI, what I really want to say is automation,” Richardson said. “It is that next level of automation.”

Richardson highlighted the crucial role of AI reasoning models, which enhance planning capabilities by “thinking out loud.”

“Reasoning models have been trained and tuned in a very specific way to think – almost like thinking out loud,” Richardson explained. “It’s kind of like when you’re brainstorming with your colleagues or family.”

NVIDIA’s Llama Nemotron models stand out because they allow users to toggle reasoning on or off within the same model,optimizing performance for specific tasks.

Richardson also stressed that enterprise IT leaders need to recognise the reality of multi-vendor environments, where agent systems from various sources operate simultaneously.

“You’re going to have all these agents working together, and the trick is discovering how to let them all mesh together in a somewhat seamless way for your employees,”

“You’re going to have all these agents working together, and the trick is discovering how to let them all mesh together in a somewhat seamless way for your employees,” Richardson said.

To help organizations navigate this complexity, NVIDIA has developed the AI-Q Blueprint for building advanced agentic AI systems. This blueprint enables teams to construct AI agents that automate complex tasks,break down operational silos,and boost efficiency across various industries. The AI-Q blueprint leverages the open-source NVIDIA Agent Intelligence (AIQ) toolkit to assess and profile agent workflows, streamlining optimization and ensuring interoperability among agents, tools, and data sources.

“We have customers that optimize their tool-calling chains and get 15x speedups through their pipeline using AI-Q,” Richardson noted.

Richardson also advised maintaining realistic expectations, emphasizing that even imperfect agentic systems can deliver critically important business value.

“agentic systems will make mistakes,” Richardson added. “But if it gets you 60%, 70%, 80% of the way there, that’s amazing.”