Using AI and rapid prototyping to de-risk modernization 

AI can speed up modernization, but only when it's used in a bounded way, with a human owning the outcome. Here's how we combine AI and rapid prototyping to make modernization faster without making it riskier. 

There's a temptation in modernization to treat AI as a shortcut: point a model at a legacy codebase and hope it comes back modernized. In our experience, that's the wrong way to think about it. AI is a powerful accelerator, but it works best on clearly defined tasks, with an engineer accountable for the result. 

AI-assisted code conversion, with a human in the loop 

The clearest example we have is a QA-automation migration. A team maintained a set of automation scripts that ran quality-assurance checks against every pull request. The scripts were written in Bash, and the team needed them in Python. 

The first pass was manual conversion. The team then began using the firm's internal generative-AI tool to do the conversion automatically, and we used the migration as a testing ground for that tool, training it to handle Bash-to-Python conversion reliably. 

The point isn't that AI wrote the code. AI took on the tedious, repetitive part of the conversion, and an engineer owned the result, checking that each converted script still did exactly what the original did. That's the pattern we apply: AI speeds up the mechanical work, and the engineer owns correctness. 

Working inside the client's AI policy 

On a fintech remediation programme, the client had a clear AI policy: AI could assist with drafting and reviewing work, under human oversight, with a human remaining solely responsible for accuracy and suitability. The client was open and forward-thinking about AI, and we used it in our implementations under the same oversight. 

For senior leaders, this is the part that matters most. AI-assisted delivery only works when there's a clear line of human accountability, and we don't blur it. 

Controlling what the model can see 

Using AI inside an enterprise also means controlling data. On a support engagement, we went on to build AI tooling for the team, including a proxy that sits in front of every AI call to make sure data that shouldn't leave the organization never reaches the model. 

If you're going to use AI on real, sensitive code and data, the security boundary around the model is part of the engineering, not an afterthought. 

Prototyping to de-risk decisions 

The other half of our approach is prototyping. On a financial-infrastructure architecture, we worked in fast, iterative loops: discuss on a whiteboard, take the idea away, prototype it, come back with questions and architectural documents, and repeat. As our architect put it, "that has worked better than trying to use other processes because it's more fluid while we're correcting the architecture." We built full prototypes for several business cases, so the client could evaluate real, running software rather than a slide deck. 

The same pattern works at a smaller scale: 

• Transaction reporting. An engineer built a quick proof of concept, a "pin" feature for temporary tabs. It landed well in the demo and went into the product. 

  
  

• Rates service. We were given room to experiment with features in a development environment, demo them, and take part in decision-making. The engagement was renewed. 

  
  

The principle is the same at every scale: prototype first, so that expensive commitments are made on evidence rather than assumption. 

The outcome 

Used this way, AI and prototyping change the risk profile of modernization. AI-assisted conversion makes the mechanical parts of a migration faster without removing human accountability. A clear AI policy keeps that work defensible. A security boundary keeps it safe. And rapid prototyping means big architectural and design decisions are tested with running software before they're locked in. 

The better question isn't "can AI modernize this?" It's "where can AI take on the tedious work, with a human owning the result, inside a policy we can stand behind?" Answer that, and AI becomes an accelerator for modernization rather than a gamble. 

Facing a similar challenge? Get in touch. We'd be glad to talk it through.