Is agentic AI posing an existential threat to COBOL and mainframes?
That said, faster code conversion does not eliminate governance obligations. Rebuilding a legacy application from extracted business logic introduces accountability questions around requirements fidelity, test coverage, auditability, and operational resilience. Leaders should treat agent-led modernization as a controlled transformation program, with explicit human sign-off on business rules, regression results, security controls, and fallback plans before retiring stable production workloads.
The operating-model implication is equally important. Many organizations have tolerated legacy estates because institutional knowledge still existed somewhere in the business or IT function. As those skills retire, AI may help capture and translate that knowledge, but it can also create a false sense of completeness. CIOs should ask whether the new target platform improves maintainability, integration, and talent access enough to justify change beyond infrastructure savings alone.
Practical next questions:
- Which legacy applications are constrained more by business risk than by technical complexity?
- What evidence will we require to prove functional equivalence and compliance after AI-assisted migration?
- Do we have a target-state architecture and product ownership model, or are we just moving old complexity onto a new stack?
Earlier this year, pundits and publications ran with a new term: SaaSpocalypse. This prognostication was based on several advancements in technology through vibe coding, legacy code refactoring, autonomous agents and Model Context Protocol server capabilities.
Without question, major change is in store for the SaaS market. This increasingly includes moving toward headless operations run by agents rather than a traditional graphical user interface.
Even Manoj Swaminathan, a president and chief product officer at SAP, said recently that the vendor is supporting this trend by exposing its APIs. But if one thing is clear, CIOs are not going to replace their major enterprise apps with something homegrown. I believe this is for two reasons:
-
Too much money and time have been invested in making these systems operational.
-
These apps are rarely a source of competitive advantage.
With this said, there are two technologies that CIOs have wanted to replace for years — and could be replaced by AI: the programming language COBOL (or common business-oriented language, designed in 1959) and that old trusty standby, the mainframe (which also dates back to the 1950s).
These technologies represent an earlier, on-premises-only era of computing. Today, companies must keep the technology modernized and agentified — and the skills needed to keep 75-year-old technology operating continue to exit the enterprise.
This isn’t a new problem. COBOL code was so ancient in 2000 that few people knew how to fix it to support going beyond 1999, as you may remember. (Personally, I booked travel early to Hawaii for me and my young family because we worried that travel systems were bound to fail as the new year approached.)
COBOL continues to age, but it also remains entrenched (and reliable) for many industries, including airlines, banking, insurance, government, healthcare and retail. See a green screen, and you’ve likely found a COBOL application powered by a mainframe.
Until earlier this month, I had not seen a trend that could replace COBOL and its workhorse, the mainframe — and I’ve worked in and written about enterprise technology for 30 years.
True, low-code vendors have slowly eaten away at these ancient lines of code for several years. But AI is having a unique effect where agents are turning the code for COBOL-based applications into a detailed specification for the application and its functionality.
With this in hand, another agent can rebuild the COBOL application using a low-code, cloud-based development stack, and another can then conduct a regression test on the resulting code.
“One of the most exciting areas is what we are doing with COBOL programs,” Woodson Martin, CEO of OutSystems, told me last month. “What we are showing enterprises is the ability to develop requirements with an agent and then build with agents is here — and then [they can] add agentic capabilities to what are COBOL programs.” OutSystems is a low-code platform focused on some of the most complicated industry use cases.
The company launched Legacy Modernization Services powered by AWS Transform in June to help companies migrate COBOL applications using a multi-step approach where agents scan code, build dependency graphs, extract business logic and ensure that a human remains in the loop.
Given what OutSystems is offering, I had to ask: Is this the death of this legacy programming language?
“Yes,” Martin said. “What we are doing will lead to the death of COBOL and the death or relegation of the mainframe.”
He pointed to the work his company is doing with the insurance industry, which hugely uses COBOL-based applications on mainframes. To manually migrate a COBOL application could take six months to a year. OutSystems is using agentic AI to cut migration time to just eight weeks. Now, Martin is hoping to get that time frame down to seven days.
“We want it to be the state of the art, to be dead simple,” he said. “This is the path for many to agentic AI.”
Outsystems’ Legacy Modernization Services enables migration of COBOL applications. Photo courtesy Outsystems
What does this all mean?
Agentic AI is clearly transforming the software industry as a whole. It will change the value proposition of SaaS applications, moving them from systems of record to systems of action.
To be clear, systems of record will remain. Over time, the agentic AI trend could affect the rents that software organizations extract from customers. It will also enable tangible business transformation as organizations pursue updates to their core business capabilities and the way they win.
The good news is that software organizations are innovating faster to support customers and avoid being left behind.
Finally, the last vestiges of the early computer era might be overcome by agents able to extract and derive a specification and build in low code a replacement for green screen apps and computing.
Success here will be a major win for CIOs who need to free resources for AI investment. And the ability to apply agents to this new code will affect companies and their customers as agents service customers for apps that are remade.
Do you think agentic AI will mark the end of COBOL and mainframes? Let us know at [email protected].
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.