For as long as I can remember, the mainframe industry has been talking about the skills gap. Experienced professionals are retiring. Finding people with deep z/OS knowledge isn’t always easy. And training someone to become an experienced systems programmer takes time. A lot of time.
There is also another problem. Much of the knowledge required to operate a complex mainframe environment isn’t necessarily written in a manual. It exists in people’s heads.
This week, IBM announced something that could begin to change that. On 6 October, IBM announced a private preview of the next evolution of watsonx Assistant for Z. The announcement is interesting because IBM isn’t simply talking about an AI that answers questions.
The next release is being designed to understand an operational problem, gather and correlate relevant information, build a multi-step plan and recommend how that problem should be addressed.
With user approval, the capabilities being previewed are also intended to carry out the recommended actions, evaluate the results and adapt the approach if conditions change.
It is important to remember that these are private-preview capabilities and represent IBM’s future direction. They aren’t generally available production capabilities today. But the direction is interesting. And it raises a question that I think is considerably bigger than the product itself: what does this mean for the people who operate the mainframe?
Mainframe Expertise Is Different
Learning mainframe technology has never simply been about learning commands. You can teach someone how to issue a DISPLAY command. You can teach JCL. You can teach them how to navigate ISPF. You can teach RACF commands. You can teach SMP/E. But that doesn’t make someone an experienced systems programmer.
Experience is knowing when to use something. And perhaps more importantly, knowing when not to.
An experienced systems programmer sees a message and may remember seeing something similar five years ago. They know which other messages matter. They understand which system components are related. They know whether something is normal for that particular environment. And they know that a problem appearing in one place may actually have been caused somewhere completely different.
That knowledge is difficult to teach. It is accumulated over years, sometimes decades. And that is one reason why replacing experienced mainframe professionals has always been difficult.
AI Changes the Knowledge Problem
This is where AI becomes interesting. Imagine a less experienced systems programmer receiving an unfamiliar system message. What would they normally do? Search the IBM documentation. Look through previous incidents. Search internal documentation. Check the logs. Ask a colleague. Perhaps eventually find the person who remembers what happened the last time.
Now imagine an AI assistant capable of helping bring some of that information together. The system information. Relevant documentation. Operational information. Previous knowledge made available to it. And potentially information from other parts of the environment.
Instead of simply returning a manual, it could help explain what is happening and suggest what should be investigated next. That doesn’t magically give someone twenty years of experience. It doesn’t turn a junior systems programmer into a senior one. But could it help them benefit from some of the knowledge that previously required years to acquire? I think it could. And that could be extremely important for the industry.
Perhaps We Have Been Looking at the Skills Gap the Wrong Way
The usual answer to the mainframe skills problem is: we need more mainframe people. And we do. But perhaps the problem isn’t simply the number of people. Perhaps it is also the amount of knowledge each person can effectively access.
Think about how much time an experienced mainframe professional has spent finding and learning information that already existed somewhere: in manuals, Redbooks, APARs, technical papers, problem records, runbooks, source code, and, very often, in another person’s head.
The information may exist. Finding the right information, understanding it and applying it to the problem in front of you is the difficult part. AI could help with that.
The next generation of mainframe professionals may therefore work differently from the previous one. They may not need to search for information in quite the same way. But they will still need to understand the systems they are working with, ask the right questions and, most importantly, evaluate the answers they receive.
That is a different way of working. Not necessarily an easier one.
Knowing the Answer Isn’t the Same as Understanding It
There is a danger here. AI can make knowledge easier to access. It can also create the illusion of expertise. Those aren’t the same thing. If an AI tells someone: “Issue this command”, there is an enormous difference between a person who understands what that command does and someone who simply copies it.
Anyone who has worked with production systems knows this. The command itself is often the easy part. Understanding the consequences is the difficult part. What else will be affected? What happens if it fails? How do I reverse it? Am I working on the correct system? Is this the right time to make the change? What happened immediately before the problem? What could happen afterwards?
AI may become increasingly good at helping answer those questions. But the person responsible for the system still needs enough knowledge to recognise when an answer doesn’t make sense.
This is why I don’t believe AI eliminates the need for experienced mainframe professionals. It changes how their expertise can be used.
The Expert May Become More Valuable, Not Less
There is an interesting paradox here.
If AI becomes good at handling some of the routine investigation involved in operating a mainframe, experienced engineers could spend less time doing those things. Their expertise can then be applied elsewhere:
- Architecture.
- Complex problem determination.
- Performance.
- Resilience.
- Security.
- Design decisions.
- Risk.
And perhaps something else that will become increasingly important: making sure the knowledge being provided to AI is actually correct. Someone needs to recognise a bad recommendation. Someone needs to understand the organisation’s particular environment. Someone needs to know that what works on one system may be completely inappropriate on another.
Because no two large mainframe environments are exactly the same. That knowledge still matters.
We May Need to Change How We Train People
This could also have an interesting consequence for mainframe education.
Historically, we have often trained people by teaching technology. Here is how JES2 works. Here is how RACF works. Here is how Db2 works. Here is how storage works. That foundation remains essential.
But perhaps we need to place even more emphasis on teaching people how the pieces fit together. Why did this happen? What else could cause it? Which information would confirm the diagnosis? What would you investigate next? What could go wrong if you make this change? In other words, less emphasis on remembering answers and more emphasis on understanding systems.
AI can be very good at retrieving information. Humans need to remain good at understanding what that information means.
Documentation Suddenly Becomes Much More Valuable
There is another consequence organisations should consider: documentation.
For years, documentation has frequently been treated as something we should do when we have time. And somehow we never have enough time. Procedures become outdated. Architecture diagrams no longer represent reality. Problem resolutions remain buried inside ticketing systems. Experienced engineers know things nobody has written down. That has always been a problem.
AI makes the quality of that information even more important.
An AI assistant cannot make good use of organisational knowledge that doesn’t exist, is inaccessible or is simply wrong. Good documentation therefore doesn’t just help people. It can improve the information available to the systems assisting those people.
Perhaps AI will finally give us a good enough reason to keep our documentation up to date. That alone would be quite an achievement.
But What Happens to the Junior Systems Programmer?
This is perhaps the question I find most interesting.
Traditionally, junior professionals learn partly by doing the simpler work. They investigate tickets. Read messages. Search documentation. Run reports. Watch experienced colleagues. Make mistakes in controlled environments. And gradually build experience.
But what happens if AI begins performing a significant amount of the work traditionally used to develop junior engineers?
This isn’t exclusively a mainframe problem. The software industry is already asking similar questions. If we automate too much of the work through which people learn, where do tomorrow’s experts come from?
We need to be careful not to solve today’s skills shortage by creating tomorrow’s.
Junior professionals still need opportunities to investigate problems themselves. They need to understand why something happened. They need to make decisions. And occasionally, in safe environments, they need to get things wrong. Because experience isn’t simply accumulated knowledge. It is accumulated judgement.
AI can potentially help accelerate that process. It shouldn’t replace it.
The Systems Programmer Isn’t Disappearing
Every significant technology change produces predictions about jobs disappearing. AI is no different.
I don’t believe the mainframe systems programmer is about to disappear. But I do think the role will change.
Systems programmers may eventually spend less time searching manuals, looking through large quantities of information and performing repetitive diagnosis. They may spend more time validating recommendations, solving complex problems, improving automation, understanding dependencies and making architectural decisions.
In other words, AI may remove some of the mechanics of the job. Not the responsibility. And perhaps that is exactly what the mainframe industry needs.
We have spent years worrying about how we replace decades of experience as people leave the industry.
Maybe the answer was never simply to replace that experience person for person. Perhaps part of the answer is finding better ways to make it accessible to the next generation.
AI won’t give someone twenty years of mainframe experience. But used properly, it may help someone with two years of experience learn from the knowledge accumulated during those twenty years. And that could change the mainframe skills conversation completely.
Be the first to comment