How AI Is Taking Over Legacy Code Maintenance (And What It Means for Developers)

Bombshell Revelation for CEOs and CFOs: AI Is Replacing Costly Legacy Code Maintenance

Let’s examine what is going on today… the hype here is real, no workers, and all profit$! Just like it early 2000 when the web was going to make everything obsolete and better. Anthropic just announced that Claude can write COBOL! What is COBOL you say?

COBOL (Common Business-Oriented Language) is primarily used for high-volume, batch-processing, and transactional applications in finance, banking, government, and retail, managing critical data on mainframe computers. It processes massive, real-time datasets like ATM transactions, payroll, and insurance records.

What are the uses and applications of COBOL?

• Banking & Finance: Handles 95% of ATM transactions and 80% of in-person credit card transactions, processing roughly $3 trillion in daily commerce.
• Mainframe Batch Processing: Efficiently processes large volumes of records (e.g., payroll, inventory management, insurance billing) in bulk
• Legacy System Maintenance: Maintains billions of lines of existing, highly reliable code for large enterprises that prefer not to rewrite stable, older systems.
• Integration with Modern Tech: Modern COBOL environments integrate with APIs, Java, Kafka, and JSON, enabling legacy data to connect with modern applications.

How much money can companies save short-term to fire all developers and integrate these systems to maintain, update, and support these mainframes that do financial transactions and payroll? IBM dropped 13% on that claim, investors have seen the light, and Large Language Models (LLMs) should take over. I mean the quote below answers many questions.

Asking an LLM the same question multiple times often yields different, non-deterministic answers, as models are designed to explore varied outputs rather than rigid repetition.

The trainwreck that is coming should prompt developers for higher consultancy rates and much higher full time offers. $3 trillion in daily commerce support is no small number given to a system that is non-deterministic. Companies have tried to migrate the unique business logic out of COBOL systems to Java continuously failing. Corporate America is littered with fail projects of better re-writes that missed the logic and features needed for these legacy systems to be sun down.

If anything I would suggest learn COBOL at this point, why?

1. Few can develop and work with COBOL
2. Stackoverflow is dead, that means no more training data.
3. COBOL is very much pre-Stackoverflow
4. Financial companies will panic when the LLMs make a mistake, who will they be able to blame except the Directors, Excecutives, CFO, and CEO?
5. You will name your own price, like a lawyer, except for software development hours.

Subscribe and read https://disesdi.substack.com/

What to keep in mind when integrating this agents into your workflow to replace your developers:

• The researcher that gave access to a mock e-mail account to an offline Clawdbot would not follow instruction
• Amazon’s outages have been because of their own Kiro agent

I personally use LLMs and ChatGPT/Claude/Gemini daily, they are tools to process data, navigate problems, and develop code faster. However, leaving these tools as agents to do their best “judgement” independently, and to blindly trust them? I don’t trust my kid to not make a mess of things when I task him with chores at home, I oversee process, understand, and verify what he does!

Normally I like to write articles that help a developer solve a problem, these are my thoughts on what is coming, and things will be difficult for the next couple of years until the dust settles. LLMs are not Artificial General Intelligence (AGI), and will never be AGI. Those are my personal thoughts on it, and if you do get laid off and then a year later asked to cleanup a project… charge like you mean it.

References

• https://www.reddit.com/r/LocalLLaMA/comments/1cvpjxu/tell_the_llm_to_repeat_the_question_an/
• https://techcrunch.com/2026/01/27/everything-you-need-to-know-about-viral-personal-ai-assistant-clawdbot-now-moltbot/
• https://www.cnbc.com/2026/02/23/ibm-is-the-latest-ai-casualty-shares-are-tanking-on-anthropic-cobol-threat.html
• https://www.tomshardware.com/tech-industry/big-tech/ibm-stock-takes-a-13-percent-whiplash-after-anthropic-announces-an-ai-tool-for-writing-cobol-code-stock-has-worst-day-since-2000-and-is-down-25-percent-mom-and-counting
• https://www.bloomberg.com/news/articles/2026-02-23/ibm-shares-plunge-as-anthropic-touts-cobol-modernization-efforts
• https://news.ycombinator.com/item?id=47128745
• https://www.pcmag.com/news/meta-security-researchers-openclaw-ai-agent-accidentally-deleted-her-emails
• https://www.tomshardware.com/tech-industry/artificial-intelligence/openclaw-wipes-inbox-of-meta-ai-alignment-director-executive-finds-out-the-hard-way-how-spectacularly-efficient-ai-tool-is-at-maintaining-her-inbox