What a Video Game Taught Us About AI and Modernizing Core Systems

IBM Bob, an AI trained on 50 years of IBM platform code, was used to convert a legacy COBOL credit card system into a modern web app in a single day – work that traditionally takes months. While AI rapidly accelerates application modernization, the key lesson is that speed must be paired with deep technical expertise to verify outputs and deliver reliable results.

At a recent IBM Bobathon, I was handed IBM Bob — an AI trained on 50 years of IBM platform code — and told to build something.

I’m not a developer. So, naturally, I built a video game.

It’s called “Business Problem Solver Suite” or “Bob Blaster” for short. You play as IBM Bob, a hard-hat-wearing robot who “turns spreadsheets into firepower,” blasting through waves of business problems that have manifested as monsters. Five weapons, including the Crisis Launcher and the Problem Spreader. Insert a “Bob Coin” to begin. It is exactly as ridiculous as it sounds, and I had a genuinely good time making it.

But it also taught me something worth sharing with anyone thinking seriously about AI and their core business systems.

First, the fun part: AI makes building things wildly accessible

I can’t stress this enough: I am not a developer, and in a single day I built a working 3D game. Over about two days of tinkering, I kept asking for more: add multiplayer, add magic, add quests. The project raced through more than 20 versions, thousands of lines of code, a slick README, a changelog listing dozens of “systems.” It was genuinely exhilarating to watch an idea become real that fast.

And here’s an honest observation about generative AI as a whole — not any one tool, but the technology itself: if you ask it to keep building, it will happily keep producing plausible-looking output. Some of what I generated was fully wired and worked beautifully. Some of it was scaffolding — a title and a placeholder. That’s not a flaw in a product. It’s simply the nature of prompt-driven creation. The machine gives you plausible very fast. I later discovered I had been “vibe coding,” a term my developer friends greeted with a fair amount of skepticism. Confirming what’s complete is a separate step.

So I did that step. I went back through the code with Bob and it created an honest specification of what actually worked versus what didn’t, governed by one rule: when the documentation and the code disagree, the code wins. That verification pass — not the game — turned out to be the most valuable thing I made all week.  The implementation of Spec-Driven Development allowed me to focus on the features and let AI do the development to meet the specifications and focus on passing the automated tests that were also created when moving to this technique.

Now, the serious part: this is exactly where IBM Bob shines

My game was me freestyling — pushing a powerful tool to improvise in a domain it wasn’t built for. But that same day, my colleague Martin Donegan used IBM Bob for what it’s genuinely designed to do, and the result was remarkable.

He took a real mainframe application — a COBOL and assembly credit-card system — and, in a single day, converted it into a modern web app. Work that historically takes months or years. And it worked, because this is IBM Bob’s home turf: it’s trained on decades of real mainframe and Power code, so it understands COBOL, PL/1, DB2, RPG and the rest natively. It isn’t guessing from general internet knowledge; it’s grounded in the actual languages our clients’ systems are written in. That specialization is precisely what makes it so powerful for core systems modernization.

That principle works earlier in the cycle, too. Another Bobathoner, my colleague Paul Reed, built Power SA Studio, a structured assessment platform designed to help our Solution Architects evaluate customers’ IBM Power estates systematically — the kind of discovery work that traditionally consumes weeks and varies wildly depending on who’s running it.

Paul embedded IBM Bob into a guided workflow: five stages that transform raw discovery data into board-ready recommendations, with the AI providing intelligent analysis at each gate while keeping the architect in control of the decisions. It’s not freestyle; it’s scaffolding for expertise. The result: what used to be a time-consuming, inconsistent process became a standardized engagement that moves faster without sacrificing depth.

What’s equally powerful is that this same approach works not just for converting systems away from the mainframe or Power, but for modernizing and optimizing the platforms themselves — dramatically accelerating what used to be months of analysis into weeks of actionable insight.

Take our recent engagement with Markerstudy, a major UK insurance group running mission-critical platforms on IBM Power. Their DISC and Orion systems process 30 million requests daily, but after 25 years of growth and acquisitions, the codebase had become difficult to navigate. In just four weeks, we deployed IBM Bob to analyze 6 million lines of code and delivered a complete modernization roadmap — identifying over £10 million in annual savings, isolating legacy code for safe decommissioning, and laying the foundation for an API-first architecture. Work that historically takes many months became clear, board-ready strategy in a month.

This is IBM Bob in its element — paired with a team that knows mainframe and Power inside out, that understands the guardrails and the gotchas. The speed comes from the AI; the certainty comes from experience.

Speed and certainty, together

The lesson from my very silly game is a serious one: AI gets you most of the way, astonishingly fast. The remaining stretch — confirming that what looks done really is done — is where deep expertise earns its keep. That’s the work we’ve always done, and AI doesn’t make it optional. It makes it more valuable, because there’s now so much more that a skilled team can accomplish.

Put a tool like IBM Bob in the hands of people who know how to use it, verify it, basically get the most out of the tool as early as possible, and application modernization stops being a multi-year ordeal and starts being something you can actually move on. That combination — IBM’s purpose-built AI and Ensono’s modernization expertise — is a genuinely exciting place to be.

So yes, I made a game where you shoot business problems disguised as monsters. But I walked away convinced of something bigger: the future of core systems modernization is AI and expertise, working together.

Frequently Asked Questions

What is IBM Bob AI?

IBM Bob is an AI tool trained on 50 years of IBM platform code, including COBOL, PL/1, DB2, RPG, and assembly language, purpose-built to help modernize mainframe and IBM Power systems.

How fast can AI modernize a legacy mainframe application?

With IBM Bob, a real COBOL and assembly credit card system was converted into a modern web app in a single day. In a client engagement, 6 million lines of code were analyzed and a full modernization roadmap delivered in just four weeks.

Can AI replace human expertise in core systems modernization?

No. AI accelerates the process significantly, but deep technical expertise is essential to verify outputs, govern decisions, and ensure what appears complete is actually production-ready.

What is Spec-Driven Development in the context of AI coding?

Spec-Driven Development is an approach where AI generates code against a defined specification, with automated tests used to confirm what is truly complete — helping teams avoid acting on plausible-looking but unfinished output.

What business outcomes can AI-assisted modernization deliver?

In one example, AI-assisted modernization of an IBM Power estate identified over £10 million in annual savings, flagged legacy code for safe decommissioning, and produced a board-ready modernization roadmap in under a month.