By Warner Godfrey, Principal Consultant Engineering
I’ve done enough application modernisation, architecture reviews and pre-acquisition technical due diligence to know the biggest risks lay hidden in the assumptions about how the system works.
Legacy systems harbour hidden dependencies and brittle integrations. Some areas are well understood; others only surface when someone unplugs that dusty server under someone’s desk and takes down production. That gap is where bad decisions creep in, scope, sequencing and budget get locked before anyone sees the full scope of what lies beneath. Without quick, low-risk discovery, teams often defer change altogether.
We’re genuinely excited about what generative AI can do here. Used well, it surfaces structure and unknowns at a speed we haven’t had before. We’ve validated by running it on a well-understood module first (does the map line up?), then turning it loose on low-knowledge areas and cross-checking with engineers and runtime signals (logs, traces, traffic). In our tests, the structural maps largely matched, but the proposed target architectures were often divorced from the business context. These tools excel at mapping what exists, but architects must still design what should exist.
Three things we’re seeing
See the whole system fast (code, infrastructure and data)
Use AI to map your entire system: code, infrastructure and data flows together. The real value comes from connecting the dots across all three aspects to reveal how your system actually works: where vulnerable dependencies create security exposure, which services depend on which databases, and which components share state in unexpected ways. Obviously, focus on data access patterns and data structure rather than letting AI loose on production data. Always cross-reference these static maps with runtime evidence like logs, traces, traffic patterns, because systems evolve differently than their documentation suggests.
Note: most tools still analyse application code and infrastructure separately; make sure you merge the views, because modernisation fails when decisions are made in silos.
Turn maps into a defensible plan (people, process sign-offs)
Generative AI tools produce the maps; humans decide the route. These tools don’t understand business impact, technical risk, or strategic value. The hard part is turning discovery into one sequenced plan with explicit trade-offs and decision gates, then wrapping it in governance so it stands up to scrutiny.
The bottom line is that tools create data; experience and governance turn it into a plan stakeholders will sign off and teams can deliver. Successful modernisation is not just tech-driven. It needs orchestration across multiple domains.
Diligence at speed (reduce time/cost risk up front)
Traditional discovery takes so long (and costs so much) that teams skip it, trim it, or rush through it, then the unknowns surface mid-project and blow up timelines and budgets. These AI tools don’t replace discovery; they compress the painful part, so leaders can grasp the complete scope, avoid mid-project surprises and protect budget.
One question to ask your team
Before your next modernisation, could you explain, with evidence, which parts of your system you truly understand, and what still lurks beneath the surface?
