Tuesday, September 8
10:15 -10:45 Australia/Melbourne Time

A March 2026 IDC survey found that 77% of organisations already have Al agents in production. The same research found that in late 2025, only 50% of organisations had delivered measurable value of any kind from their Al initiatives. The gap between "running" and "working" is not a model problem - it is an operations problem. Al agents are non-deterministic, susceptible to prompt injection, and capable of multi-step reasoning that compounds errors across tool calls in ways conventional software never did. Business teams are deploying them without centralised oversight, drawing on LangGraph, CrewAl, OpenClaw, and every other framework that shipped this quarter, creating shadow workloads that no one in security or platform engineering has visibility into. The technology that makes agentic Al exciting is the same force that makes it hard to govern.

The answer is AgentOps: the operational discipline of running agents safely, reliably, and cost-effectively in production. IDC defines five pillars - identity management and access control, guardrails, observability, evaluations, and cost tracking - and the organisations that apply them are the ones that actually get to production. Red Hat OpenShift Al makes this concrete. It is framework-agnostic by design, so agents built in any harness deploy without rewrites. What the platform adds is everything production requires: kernel-isolated sandboxes, least-privilege access control enforced at the workload level, deep execution tracing across every tool call and reasoning step, evaluation pipelines that catch model and prompt regressions before users do, cost attribution down to the agent and workflow level, and red team security testing and adversarial assessment built into the deployment pipeline - not bolted on after.

This talk shows what that stack looks like in practice: deploying a multi-framework agentic workload on Red Hat OpenShift Al, instrumenting it end-to-end, and running an adversarial probe that demonstrates exactly why none of those controls are optional. Agent harnesses and agent frameworks deployed on Red Hat OpenShift Al- governed, observed, evaluated, guardrailed, red-team tested, and cost-tracked. You need all of it. Without it, you are not getting to production safely.

Mike Hepburn, Al Senior Principal Technical Marketing Manager, Red Hat

Tuesday, September 8
11:50 - 12:35 Australia/Melbourne Time

• Top-down Al leadership: Identifying use cases that deliver tangible business value
• Evaluating tools to invest in areas that will deliver results and minimise organisational risk
• What does responsible Al adoption really mean, especially as it relates to privacy, intellectual property and ethics
• Effective workforce communication strategies to address employee apprehension

Bryon Baker, Principal Specialist Solution Architect - Al, Red Hat
Phil Laufenberg, Pro Vice-Chancellor (Al) and Chief Al Officer, La Trobe University
Rhys Weekley, CTAIO, Australia Post
Craig Lawton, Principal Al & Data Technologist, AWS
Moderator: Maile Carnegie, Senior Leader Focused on Innovation & Growth