Groups | Search | Server Info | Login | Register
Groups > comp.lang.cobol > #12894
| From | Arne Vajhøj <arne@vajhoej.dk> |
|---|---|
| Newsgroups | comp.lang.cobol |
| Subject | Re: AI could keep COBOL running |
| Date | 2026-03-04 09:52 -0500 |
| Organization | A noiseless patient Spider |
| Message-ID | <10o9gug$2v1fr$1@dont-email.me> (permalink) |
| References | <772294366@darkrealms.ca> <4k8bqkhogfcinb4gaef6j5c96fhrm9dgu9@4ax.com> |
On 3/2/2026 9:49 AM, Joe wrote:
> On Sat, 28 Feb 2026 10:59:25, NOSPAM.sysop@darkrealms.ca (Dumas Walker) wrote:
>> 'Modernizing a COBOL system once required armies of consultants
>> spending years
>> mapping workflows... AI changes this': Anthropic says AI could help
>> keep COBOL
>> running for a long time to come - but IBM won't be happy
> [.......]
>> This shows that even long-established enterprise software vendors may
>> face
>> disruption as AI continues to reshape legacy system modernization.
>>
>>
>> https://www.techradar.com/pro/modernizing-a-cobol-system-once-required-
>> -armies-o
>> f-consultants-spending-years-mapping-workflows-ai-changes-this-
>> -anthropic-says-a
>> i-could-help-keep-cobol-running-for-a-long-time-to-come-but-ibm-wont-be-
>> -happy
>
> Need to see it before believing..... I've build Cobol constructs of which no AI would understand. Why? Because at the time there
> were no usable alternatives with enough speed after compiling. Call it "assembly language using Cobol".
>
> So, for now maybe a promise which will give CTO's a chance to spend money on the latest hype...
This is mostly a media hype thing.
Automatic translation from language X to language Y is not
a new thing introduced with AI. It has been possible for decades.
I would expect tools (whether traditional or AI based) to be able
to correctly translate the code. There may be exceptions but
automatic conversion of 95% of the code and manual translation
of the remaining 5% is still good.
It have been possible for decades before AI, but it is not
particular interesting, because it is not the problem businesses
are looking to have solved.
Let us say that the tool can convert COBOL to Java (nothing special
about Java in this context except that a lot of companies using
COBOL prefer to convert to Java).
First problem is maintenance. Manual maintenance of automatic
converted code is harder than maintenance of human written code.
Having AI do maintenance makes maintenance fast but raises
new issues about what really got modified and therefore need to
be tested. A cheap conversion followed by expensive maintenance
is not attractive to businesses.
Second problem is that the language is not the only problem
that businesses want solved - most likely it is not even the
most important problem business want solved. They don't just
want to modernize the programming language - they want to
modernize the entire architecture. Specifics vary, but in
many cases this means some of switching from ISAM to RDBMS,
switching from vertical scalable to horizontal scalable,
switching from a online+offline model to an always online
model, switching from EBCDIC to Unicode etc.. Sure you can ask
the AI to do that as well. But unless it receives a gigantic
input about how it should be done then the result will be a
disaster.
So the "AI will just convert the entire code base" story is BS.
What is not BS is that AI can help during such conversions.
Several companies have had good use of AI in conversion projects.
Primarily doing two things.
A) Analyzing the old code and extracting what it does. 100% accuracy
is not needed. If the developer creating the new code can use
the extract in 90% of cases and only in 10% of cases need to
ask a COBOL programmer to take a look, then it is pretty good.
B) Help write the new code. Don't think "vibe coding" - think
"intellisense on steroids". Instead of auto completing a name
it auto completed entire code blocks or even methods.
That increases productivity.
But it still requires people that know the old technology and
people that know the new technology.
And I would not worry about IBM. They have that knowledge.
Arne
Back to comp.lang.cobol | Previous — Previous in thread | Next in thread | Find similar
AI could keep COBOL running NOSPAM.sysop@darkrealms.ca (Dumas Walker) - Sat, 28 Feb 2026 10:59:25
Re: AI could keep COBOL running Joe <none@nowhere.whereo> - 2026-03-02 14:49 +0000
Re: AI could keep COBOL running Arne Vajhøj <arne@vajhoej.dk> - 2026-03-04 09:52 -0500
Re: AI could keep COBOL running Rich Alderson <news@alderson.users.panix.com> - 2026-03-02 18:38 -0500
csiph-web