Rethinking Performance Management on IBM Z in the Age of AI

By Josh Stiles, Product Lead, 21st Century Software

The biggest misunderstanding in performance management is thinking AI can compensate for the wrong plumbing underneath it.

Root cause analysis, trend detection, machine learning - none of it matters if your data is not being handled properly and in real time. AI is supposed to help your team get proactive. But without the right tooling, you never get the chance to catch problems before they compound.

I see this repeatedly in conversations with IBM Z teams. The intent is there. The data is there - IBM Z generates hundreds of SMF record types covering everything from CPU and memory to I/O and network activity. What is missing is the layer that turns that data into something your team can actually act on before something breaks.

The performance management gap most teams are not talking about

72% of organisations running IBM Z reported an increase in general-purpose MIPS capacity last year.¹ Workloads are growing. And nearly three in four mainframe shops have already identified generative AI as a strategic initiative.¹

That is a lot of new pressure landing on capacity planning teams - most of whom are still working with the same performance management workflows they had five years ago. The teams we work with that are getting ahead of it share one thing in common: they fixed the data layer before they added anything on top.

Two types of AI on IBM Z - and why the difference matters

AI is arriving on IBM Z in two distinct ways, and most teams are treating them as separate problems. They are not.

The first is AI as a new workload. IBM z17 can perform over 450 billion AI inference operations per day at sub-millisecond latency. ² That is a fundamentally different workload profile from traditional OLTP and batch. If your capacity planning was built around the old profiles, you already know this changes your MIPS picture in ways that are not always straightforward to model.

The second is AI as a method - applied to the SMF data your environment has been generating for years. This is where the plumbing argument matters most. If the data foundation is not right, AI cannot surface what is actually happening across your z/OS environment in time to do anything about it. The teams getting this right are not adding AI on top of the same workflow. They are fixing the data layer first.

97% of mainframe professionals now view the platform as a long-term foundation or a platform for new workloads - the highest positive reading in the 20-year history of the BMC Mainframe Survey.¹ That confidence is only sustainable if your performance management can keep pace with what you are running.

What we are doing about it

These are exactly the conversations we want to have openly - with practitioners, with partners, with the people building the next generation of tooling.

That is why 21CS is launching the Mainframe Intelligence Series. The first session - IBM Z at an Inflection Point: Performance, Talent and the AI Question - is on July 21. No pitch, no demo. Just an honest conversation about where the platform is heading and what that means for the teams running it.

Register here

Josh Stiles is Product Lead at 21st Century Software, where he leads OPTIMAn - performance and capacity management built natively for IBM Z.

Sources

1. BMC Software, 20th Annual Mainframe Survey: State of the Mainframe in 2025, 2025.
2. IBM Newsroom, IBM z17: The First Mainframe Fully Engineered for the AI Age, 2025.