I had many business conversations in 2024, but I keep coming back to just one phrase. I was chatting with an analyst who specialises in AI and infrastructure and what she said during that conversation – that “it’s now or never for legacy modernisation” – has stuck with me ever since. The reason she said that is simple: AI has upped the stakes. What was once a decade-long marathon of expensive enterprise migrations has become a sprint, demanding immediate action.
Why? In AI, there will be winners and losers. The winners will not be building on legacy technology, like relational databases. After all, relational databases can block the modern, flexible and scalable use cases that will enable AI, which relies on ready access to data that is highly detailed, unstructured, interrelated, timely, and nuanced.
If an organisation spends the next five years getting its infrastructure in order to build AI applications, they will be too late to the party.
But just as AI is piling on the pressure to modernise, IDC predicts AI will be a $631 billion market by 2028, it is also now evident that AI can be part of the solution. When used correctly, AI is the crucial ingredient to a new and transformative way of approaching modernisation that is far cheaper, quicker and more effective than anything that was possible previously.
In 2025, there will be a cohort of about 30% of APAC enterprises that have the people, processes and platforms in place to go on this modernisation sprint and to capitalise on AI. It’s now or never.
Don’t defer: Legacy tech debt is holding you back
About a third of an enterprise’s technology landscape is made up of legacy systems. Just maintaining it can take up to 80% of an IT budget. After security, it’s the single biggest headache for CIOs.
Three key issues stand out:
- Rigid and hard to change: there are unnecessary constraints on developers, an estimated 42% of developer time is consumed by maintenance rather than innovation.
- High cost: Enterprises are spending billions just maintaining outdated systems’ expensive hardware, punitive licensing, cloud lock-in and intrusive audits
- Blocks modern use cases: Not flexible enough to easily meet today’s application requirements like gen AI.
To deliver valuable AI experiences it’s all about unlocking your own organisational data and plugging that into an intelligent machine. Without ready access to your own data, your app is just a shiny wrapper on ChatGPT. If all your data is stuck in silos of legacy infrastructure, accessing it in this way becomes very difficult.
If you’d like to try building an intelligent, fast and cost-effective AI platform on top of your 30 year old infrastructure, I wish you the best of luck.
Given these challenges , it’s not surprising companies have wanted to modernise these environments. But the decision to do this is often deferred because modernisation is seen as being too hard or too risky.
It’s hard because these are complex estates and there’s a lack of necessary institutional knowledge and skills. There are some usual suspects, such as Oracle, Sybase, Cobol and their ilk – but each stack is unique which means there’s not standardised modernisation expertise and tooling.
Modernisations have happened, but they’ve historically been slow, expensive and very disruptive to the business. Or they’ve been faux-modernisations, in which a legacy estate was simply lifted and shifted from on-premises to the cloud, leaving the data and applications layers essentially untouched. At best this delivers only modest benefits.
So what are the other options?
How AI is transforming modernisation
This year we found that a new AI powered modernisation process can dramatically speed up and lower costs for modernisations.
We did this by combining Large Language Models (LLMs) with a great team of experts and some of MongoDB’s existing proprietary tooling. This accelerated a number of steps in the modernisation journey, particularly in the most complex layer: the application itself.
There are numerous steps in a genuine modernisation and AI can help at each point. But here are three areas where AI has a an outsized impact:
- Early stages: Use the LLMs to analyse legacy codebase in order to understand its structure, eliminating the need for input from original developers.
- Analyse: see how users interact with the app and then feed that information to an LLM to create tests that ensure the viability of the new application.
- Partially automate: use AI to automate the creation of microservices by using existing source code and recorded end-user behaviour with LLMs.
This works. For just one example look at Bendigo and Adelaide Bank. While modernising a core banking application the bank was able to:
- Cut developer time required to migrate off legacy database by 90%
- Improve testing speed from 80 hours to 5 minutes.
- Complete the experiment at one-tenth the cost of a traditional migration.
And it’s just the beginning. The bank is now repeating the process on other core banking apps.
When done correctly, this not only massively reduces time and costs, but also leaves an organisation with a modern tech stack that is ready for AI.
Experiment with an AI powered app modernisation factory
We all should be asking our partners and vendors how they can bring simplicity to the modernisation process. And to me, the key is to start with a quick experiment on a single application that enables the organisation to learn and demonstrate the value of new processes. Then use that to create a playbook and set of tools for modernisation. This becomes an app modernisation factory that allows the organisation to prioritise and then churn through the modernisation of its applications.
These results and the potential presented by AI are why I’m convinced by the analyst’s argument: 2025 will be now or never for legacy modernisation.