This incident happened a few years ago while I was consulting for a company. A friend from the reporting team called me to help with an issue. A particular reporting table had not loaded that day, leading to an error in the reports that were generated that day. The team had looked and found that there was a mainframe-based application responsible for putting out a file which made its way to this reporting table, and that file had not arrived that day.
Identifying the source of the issue was simple; the real
problem actually started there. There was no one on the team who knew how to
use a mainframe or even log into one. Since I have some experience working on Mainframes,
my friend requested me to help him resolve the issue. I requested access to the
Mainframe and investigated the issue. The issue was simple - an incoming file
had not arrived on time and so the JCL hadn't triggered. The fix was straightforward—to
request the support team to rerun that Mainframe job.
A Symptom of a Larger Problem
It was a small fix, but the incident highlighted something
bigger. On one hand, it proved again why Mainframes remain so dominant, even in
this era of Cloud and AI. Programs written 40–50 years ago still run—and they
just run. Yes, the screens are black and green, but they never failed to do
what they were designed to do.
I might be wrong, but I believe the Y2K (Year 2000) crisis dealt the
first significant blow to the image of the Mainframe. The crisis wasn't a
failure of the Mainframe itself, but a byproduct of a different era—one where
programmers slashed the '19' from the YEAR and used the 'MM-DD-YY' format for
date simply to save on the costs of memory. Yet, Mainframes got all the bad
rap. This was further compounded by the Mainframe getting overshadowed by
newer-looking, colorful, and "cooler" technologies arriving with
cooler names, too. We are now at a point where many major corporations on the
planet use the mainframe in some form, yet there is hardly anyone left who
understands the code.
- The
developers who built the code have now retired, and there is no one ready
to take over from them to continue working on Mainframes.
- Working
on Mainframe is considered an “uncool” job. Everyone wants to work on
newer tools where they can drag and drop to build an ETL pipeline or build
colorful screens with images and videos.
- Colleges
have almost stopped teaching COBOL, JCL, and other Mainframe tools. I have
not seen any coaching institutes teaching Mainframe skills either.
- Mainframe-related
knowledge on internet forums also seems to be scarce compared to newer
programming languages and tools.
- Libraries
carry tons of books related to Cloud and AI, but hardly any related to the
Mainframe.
Bottom line: companies are running on mainframe code, yet
hardly anyone knows anything about the underlying programming language, the
technology stack, or the business logic that is running behind the scenes.
CEOs and CTOs proudly talk about how AI is developing not
just code snippets, but entire applications in minutes and hours— that
otherwise took weeks and months with a big development team. Companies boast of
how productivity has improved with AI doing all the coding with just a few
prompts, saving them tons of money. AI platforms can now detect and heal code
bugs without any human assistance. Once you lay out a basic idea, AI-based
platforms can build and implement those solutions quickly with zero human
intervention.
All this sounds fantastic, and it looks like the future is
already here. But what does all this really mean? I think this only means that
we are now adding loads and loads of AI-generated code, all running with little
to zero human intervention. AI is becoming better every day, and this process
will only hasten.
History Repeating Itself
Maybe 20–30 years down the line, we will see the "Mainframe
scenario" repeat itself, where manual programming is considered uncool.
Colleges will likely have stopped teaching programming as a subject, and
libraries will be filled only with books related to prompt engineering, RAG,
and other AI tool stacks.
My Prediction: The Y2.5K crisis of 2050
Imagine you are in 2050 and an AI-generated system has
stopped working. The 'Prompt Engineers' of that era will stare at a strange set
of lines (the actual programming code), trying to decipher what they really
mean. We will have traded the 'black and green' screens of the 1960s for the
'black boxes' of 2050.
Perhaps this will be known as the 'Y2.5K' crisis: a
world relying on a foundation of code that works perfectly until it doesn’t,
only to realize we have let the skills required to look under the hood go
extinct.
~Narendra V Joshi