From Newsgroup: comp.lang.cobol

'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

By Efosa Udinmwen published 19 hours ago

Engineers can now modernize complex COBOL systems using AI

AI automates COBOL code exploration, maps dependencies, and
analyzes
structural risks quickly
Engineers can prioritize modernization based on technical risk and
business
value efficiently
Automated tests verify that migrated COBOL components produce
identical
outputs to legacy systems

Modernizing legacy COBOL systems has long been a costly and
labor-intensive
process that requires extensive human effort, as traditionally, teams
of
consultants spent months or even years mapping workflows, documenting dependencies, and untangling decades of accumulated business logic.

Hundreds of billions of lines of COBOL still run in production
worldwide,
powering critical systems in banking, government, and airlines, yet
finding
developers with the knowledge to interpret these systems has become increasingly difficult.

Now, however, Anthropic is looking to supplant this, with its Claude AI platform aiming to take much of the heavy lifting away from human
workloads.

How AI aids code exploration and analysis

This scarcity of expertise has historically slowed modernization
projects and
increased costs - however, Anthropic now believes AI can automate much
of the
exploration phase that once consumed most human effort.

"Modernizing a COBOL system once required armies of consultants
spending
years mapping workflows... AI changes this," the company said in a blog
post.

Tools like Claude Code can map dependencies across thousands of lines
of COBOL,
trace data flows between modules, and document workflows that current
staff no
longer actively remember. These automated processes identify risks,
isolate
tightly coupled components, and flag duplicated or potentially fragile
code.

By analyzing these structural and functional relationships, AI can
prioritize
which components to modernize first based on technical risk, business
value,
and organizational priorities.

The best laptops for programming allow engineers to integrate AI
outputs
efficiently while maintaining oversight of the modernization plan, and
once
components are prioritized, AI can generate preliminary function tests
to
verify that migrated code produces identical outputs to the legacy
system.

Human teams then decide whether these automated tests are sufficient,
which
scenarios require manual verification, and what performance benchmarks
must be
maintained. Implementation proceeds incrementally, with each module
tested and
validated before additional changes are made.

AI tools can translate COBOL logic into modern languages, create API
wrappers
around legacy components, and build scaffolding that allows old and new
code to
operate side by side. This reduces the risk of large-scale failures
and
enables organizations to move forward with complex modernization
projects.

AI also provides detailed insights into potential technical debt,
isolated
modules, and high-risk areas, allowing teams to plan modernization strategically - as engineers can review these recommendations and
sequence the
work to align with regulatory requirements, business priorities, and operational constraints.

Automated documentation and analysis give teams comprehensive
situational
awareness, but final decisions still rely on human judgment.

While this is a major win for many engineering teams, IBM, a major
vendor of
COBOL-powered mainframes and enterprise systems, will not be pleased.

The company saw its stock fall sharply after Anthropic announced that
Claude
Code could automate much of the labor-intensive modernization process.

AI's ability to replace work traditionally done by human consultants
threatens parts of IBM's business model.

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

$$
--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: comp.lang.cobol

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...
--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: comp.lang.cobol

BWAH-HAH-HAH-HAH-HAH-HAH!!!!!
--
Rich Alderson news@alderson.users.panix.com
Audendum est, et veritas investiganda; quam etiamsi non assequamur,
omnino tamen proprius, quam nunc sumus, ad eam perveniemus.
--Galen --- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: comp.lang.cobol

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


--- Synchronet 3.21d-Linux NewsLink 1.2