Agentic AI can make mistakes, just like humans, meaning it requires guardrails to understand context, validate actions and escalate issues to a human when uncertain.
Global solutions integrator pioneer Insight Enterprises is focused on the next generation of AI adoption.
Just as human employees go through training before they’re trusted with a business role, Vanamo says agentic AI requires “context engineering” to define business roles, set out the rules of engagement and set the limits of its authority.
“You also need transparency in the form of a detailed log which explains the reasoning for each decision, so you can keep a human in the loop to validate actions and ensure those guardrails are adequate,” he says.
“Those humans managing AI can be subject matter experts within your business, rather than AI developers, which keeps your people in the loop rather than cutting them out of the picture.”
Retrofitting this kind of governance can be time-consuming and expensive, which is why it is important to build in data lineage, safety filters, human‑in‑the‑loop and auditability from the ground up.
Insight Enterprises’ senior vice president and managing director, APAC, Mike Morgan.
Along with context engineering, another way to improve the performance of agentic AI in a business context is to train it as a subject matter expert, to help avoid the hallucinations which impact some generic generative AI tools.
Similar to a human workforce, performance can also benefit from using a team of multiple specialist AI agents working in collaboration, rather than expecting one AI agent to manage an entire workflow, says Mike Morgan, Insight Enterprises’ senior vice president and managing director APAC.
Ensuring each discrete AI agent is scoped more tightly to perform a specific task, rather than relying on a more generalist AI agent to handle the entire workflow, makes things easier to manage, improves accuracy and reduces risk, Morgan says.
For example, Cricket Australia Live App’s AI Insights, built on Microsoft Azure OpenAI Service, uses multiple AI agents to provide fans with real-time, text-based updates on player statistics, key narratives and notable milestones during live cricket matches.
Initially, the AI component of the Live App was built with a single AI agent which attempted to be an expert on every aspect of the game, but this generalist approach became too unwieldy, says Morgan.
“Today, the Live App’s AI Insights works more like a team, with seven specialist AI agents each contributing their expertise, managed by an overarching agent determining which insights are passed along to the fans.
“Rather than relying on developers to hone the system’s accuracy, Cricket Australia’s cricketing experts are the ones using our tools to hone its performance to ensure the platform continually delivers better outcomes.”
To discover how Insight can support you with your AI journey, click here.
