Faster discovery, higher productivity, less friction to production
It seems that every client conversation now includes generative AI in some form: strategy, risk, compliance, data, models and tools. Everyone wants to compare notes, an inevitable consequence in a tech landscape that is still evolving. More and more, these chats converge on one thing: using generative AI to modernise legacy. The interest has been overwhelming, so we’re bringing it to the front.
Here’s what we’re doing:
- Running an event with AWS on using generative AI for legacy modernisation to share potential, patterns and lessons (more info further down).
- Starting a regular series here on LinkedIn called The Tech Trajectory* to put what we are seeing on the ground in one place, so people can compare notes with less noise. Today’s issue focuses on legacy modernisation, future issues will cover other areas where AI, data and product are reshaping our industry.
So why the energy around modernising legacy? Probably because it’s painful. It’s common to have years of tech debt: multiple systems doing the same job, undocumented code, and critical dependencies often running both ways. The system may depend on critical services, and critical services may depend on it. Decisions get deferred because the unknowns are heavy and decoupling takes time while keeping operations running.
We’re seeing that generative AI supports key parts of the modernisation process. But specifically we’ve had great success in understanding what you have and getting legacy code bases to a “better” spot more quickly. It can analyse large codebases, map dependencies, generate or update missing documentation, extract key information buried across artefacts and support quick prototypes on top of legacy. The result is faster assessment, enabling more focus on conversations and options around the best way to tackle the modernisation process.
If the analysts are close, expect the next two to three years to retire more legacy than the past decade, with generative AI driving cheaper, clearer discovery and steadier, lower risk modernisation. Of all the AI developments that have landed, our sense is that this is a space that will drive real value for clients.
*If you know us well, yes, it shares a name with our podcast.
Productivity gains are real: more motivation for modernisation
The evidence is clear on AI-powered engineering. Not everywhere and not all the time, but used well it materially lifts individual productivity. Nothing is for free though, there’s still a need to focus on quality and disciplined software engineering whilst ensuring the guardrails are in place to support this new way of working.
Many greenfield environments are exploring and adopting AI to deliver team gains across the delivery lifecycle, baking it in as part of how they build and ship products. In legacy systems, the same gains are capped by weak foundations like outdated databases, end-of-life operating systems, and brittle virtual machines. Modernisation creates the conditions AI works best in: explicit versioned APIs, aligned data schema and real observability. And the benefits compound as you modernise. If you want the productivity lift, prioritise modernisation where the risk–return makes sense.
Shipping generative AI in enterprises with established systems
Recent MIT research found 95% of enterprise generative AI pilots aren’t yet delivering measurable returns. That doesn’t mean enterprises are failing at generative AI. We’re working with plenty of large organisations that have invested and are already seeing returns.
Looking deeper, it’s the usual culprits: application of technology where it’s not strong/not a fit, integration, legacy systems, data access and quality, governance and trying to fit with existing workflows vs assuming some level of change management. They’re not unique to enterprises but certainly amplified there. It’s also reflective of the reality that you’ve got to work with what you’ve got. Waiting to modernise or redo that process before you start with generative AI is just not feasible. Instead, we recommend a low-risk, incremental and iterative path to value while you identify where to modernise or streamline.
Even highly-regulated organisations are driving success with generative AI. In these environments, nothing ships until all key stakeholders (product, risk, legal and ops) are comfortable with releasing to clients. It may sound strange, but in fact, in these early stages, the AI model itself may well be the smallest part of an AI-powered feature in your product. The real work is the engineering, controls and sign-offs that make it safe to run in production.
Starting with a low-risk path where the key complexities are organisational, educational and process related is where the success is, based on what we’re seeing. Some are starting with public data, no customer data and no internal records to avoid most integration and data governance hurdles. This way organisations can start with low risk, but high/good value use cases and quickly begin the journey of understanding what it really means to deliver AI into a production environment, laying the foundation for future enhancements and more complex (and agentic) solutions.
Shipping something small and meaningful to exercise the full release path and build momentum is a great place to start. In parallel, identify what you need to modernise for clearer interfaces, consistent data contracts and real observability. Then, when you do need internal data or a feature that touches multiple parts of the business, integration is faster, safer and more predictable.
You get early, bankable wins with generative AI while you steadily reduce legacy or cross-organisational drag. And you bring the organisation on the journey with you.
