AI-driven Mainframe Modernization and Why it Has Failed to Deliver

A Short History of AI in Software Development

AI in software development is not a completely new thing. In fact, already in the 80’s developers utilized “AI-tools”, like syntax highlighting to write code faster. (I know there’s a debate about what actual AI is, but bear with me) Since then, the development of AI-aides for programming has been accelerating, creating to tools like low-code platforms and todays AI-powered coding assistants like Copilot, Cursor and Devin. The focus being, to enable humans to write code as fast as possible.

And on an individual level, these tools have absolutely delivered. Developers can prototype in minutes, fix bugs in seconds, and create new features without writing a single line of code. But when we take a step back and measure how much these advances have improved the efficiency of entire software teams in enterprises, the numbers aren’t that impressive anymore.

For instance, GitHub Copilot has been reported to increase developer efficiency by 20-50%, but the efficiency gain for complete software teams is estimated to be around 5-20%. That’s remarkable progress for individuals, but less impressive when viewed at an organizational level.

It seems that the farther you zoom out, the less of an effect AI has. Why is that?

What Makes a Software Team Truly Efficient?

To dig into why there is an efficiency gap between the AI acceleration of individual developers and complete software organizations, let’s look at what makes a software organization truly efficient. I would split the essentials into three pillars:

1. A shared goal – A clear, aligned purpose that drives every decision.
2. The right context – Knowledge about users, business, system interdependencies, architectural decisions etc.
3. Psychological safety – The freedom to ask, challenge, and collaborate without fear.

Most enterprises that’ I’ve seen don’t suffer from lack of goals or safety. Mostly, they have had a clear, company-wide vision and excellent psychological safety, enabling trials with new technologies like AI.

The thing that is missing however, is the right context. Decades worth of information embedded into humongous code bases and complex processes which are a lot to share across hundreds of teams.

Since they don’t work in isolation, the lack of context is not only limited to the software team. On the contrary, the most efficient organization share basically all the context with everybody. That’s why startups are usually more efficient per capita than large enterprises: everybody shares basically all the context all the time. As companies grow, maintaining that becomes increasingly hard.

The Effect of Missing Context in Mainframe Organizations

Context becomes especially complicated in mainframe organizations, where silver-haired mainframe experts have been retiring in masses, taking the precious context with them and eroding the foundation for efficient software development. What has come to replace them, is significant outsourcing of central operations to 3^rd parties, which has made the situation worse, and the walls of silos have grown thicker. These 3^rd parties are also incentivized to keep as much of the context with them as possible, creating vendor-lockin.

A 2024 Forrester study reported that mainframe developers spend only 16% of their time programming. The other 84% is spent doing everything else, which is dominated by code comprehension: searching for functionality in massive legacy code bases. The remaining time is spent on sitting in meetings, re-discovering business logic and so on: finding or sharing context.

An anecdote about the importance of code comprehension is that the average mainframe developer spends 3-7 days just searching where they need to input their code in the code base. And that’s for every new feature, resulting in thousands of hours wasted yearly. Not to mention the security and compliance risk that comes along with not understanding how a business-critical systems work.

Nomain: on the Quest for the Perfect Context

Nomain is built to bring the right context to everyone working with legacy codebases, unlocking the real potential of AI in enterprises. By mapping logic, dependencies, and business rules across entire systems, Nomain helps teams understand how their mainframes work, without spending weeks reading code or searching for documentation. Through Nomains integration layer, we can add virtually any additional information like ticketing systems and telemetry data into one single platform and connect it into the scaffolding provided by the mainframe code base, enabling full context to developers and business analysts alike.

In other words, Nomain doesn’t just make coding faster, it makes understanding faster.

When each team member has the right context, they can focus on creating value, reduce MIPS consumption, and modernize with the big picture in mind, turning the burden of mainframe code into an asset.

A Final Thought

AI has almost doubled the speed at which developers write code. But only by accelerating understanding, organizations can unlock the real potential of AI.

The next era of mainframe modernization won’t be defined by who codes fastest, but by who understands their systems best.

And that’s exactly the problem Nomain is built to solve.