Microsoft / GitHub
GitHub Copilot
Code completion, refactoring suggestions, and support during translation from older language versions into modern stacks. Part of active engineering work, not an automated replacement for judgement.
AI-Assisted Modernisation
AI-Assisted Modernisation:where we use AIwhere we use AIwhere we don'twhy it matters
Most companies talk about AI-assisted delivery without explaining what the AI is doing in practice. On this page we show where AI saves time and cost, which tools we use, and when and why we don't use AI.
How we use AI
A clear breakdown of what AI does on a modernisation project, what it doesn't, and how that shapes the price you pay.
100%Targeted
Targeted tooling
Each AI tool earns its place on a defined task in the modernisation process — never as a general-purpose shortcut.
Faster, broader, cheaper
Faster documentation, lower analysis cost, broader test coverage — measured against the same work done by hand.
Architecture
Senior-ledBusiness logic
Senior-ledUnreviewed AI
Never shipsHuman judgement leads
Architecture, business logic, and delivery risk stay with senior engineers — every time.
20–40% savings
Typical savings of 20–40% on tasks where AI fits — passed through to project pricing, not absorbed as margin.
Up to
40% saved
Secure by default
AI tools run under contract — never on consumer chat interfaces, never feeding model training.
- API-only, isolated sessions
- Excluded from model training
- NDA & agreed protocol
Where AI helps
Four tasks where AI-assisted delivery changes the economics of modernisation — with the time savings and review controls that come with each.
Documenting undocumented codebases
Many legacy systems carry little or no useful documentation — no current specifications, sparse comments, no record of why design choices were made. We use AI code analysis to turn large volumes of legacy code into structured documentation: functions, dependencies, workflows, and data movement. Work that often takes weeks compresses into days, so scoping happens earlier and with less wasted effort.
0x
Faster documentation
50k-line codebase: 3–5 weeks → 3–5 days
0%
Lower assessment cost
Reflected directly in project pricing
Extracting business rules from old code
In older systems, real business rules often live inside the code rather than in a written process document — pricing logic, approval steps, exception handling, edge-case rules. We use LLM-based analysis to surface those rules as readable specifications. Your team reviews them, our engineers review them, and only then do they become the basis for migration planning.
0x
Faster scope agreement
2–4 weeks of discovery → 1–2 weeks
0%
Engineer-reviewed
Every extracted rule is verified before sign-off
Generating broader test suites
One of the main risks in legacy modernisation is reaching go-live with shallow test coverage. AI helps us generate an initial suite of test cases from the existing codebase — including edge cases and failure scenarios — broader than a manual first pass. Engineers review every generated test before it is used, so coverage grows without losing control.
0x
Faster first suite
1–3 weeks manual → 1–3 days assisted
0%
Wider edge-case reach
Failure paths a manual pass tends to miss
Accelerating translation from legacy languages
Some legacy systems still rely on languages that are expensive to hire for and difficult to maintain — VB6, Classic ASP, older Delphi, FoxPro. We use tools such as Amazon Q Code Transformation and GitHub Copilot to accelerate translation into modern languages and frameworks, so engineers focus on review, architecture, and risk control instead of repetitive rewriting.
0x
Faster translation
4–8 weeks → 1–2 weeks for typical modules
0%
Engineer time redirected
Spent on review and risk control instead
Your project doesn't start from an empty repository.
A conventional rebuild often spends the first three to six months on infrastructure, deployment, access control, and setup before the client sees business value. We start from a proven application foundation, so more of the budget goes into the workflows, rules, and integrations that make the system specific to your business.
FoundationIncluded
Internationalisation & localisation
Responsive desktop & mobile layouts
Test data and demo data
Pre-configured CI/CD pipeline
Authentication & role-based access
Audit logging
Dev, staging, and production environments
Modern security baseline
Security and data handling
If AI tools are going to touch your codebase, the rules need to be clear before any work begins.
API-only, isolated sessions
Code is submitted via API in isolated sessions. We do not use shared consumer-facing AI interfaces for project work.
Excluded from model training
Data-processing agreements are in place. Your code is contractually excluded from model training data.
Minimum data exposure
Where possible, analysis uses anonymised samples or structural analysis rather than unnecessary exposure of full business data.
NDA & agreed protocol
NDA and confidentiality terms apply across the engagement. The data handling protocol is agreed before AI tools are applied.
We tell you in advance which tools will be used, what data they will process, and what controls apply. You can request changes to that approach before work starts.
The tools we actually use.
A short stack — each tool earns its place on a specific task.
01
AWS
Amazon Q Code Transformation
Structured code migration where an initial transformation plan is useful before engineers review and refine the result. Supports migration; does not determine the architecture.
GPT-5.5 and Claude via API
LLM-based code analysis
Documentation generation, business rule extraction, and data schema analysis. Code is submitted through controlled API workflows rather than consumer chat interfaces.
Task-dependent
AI-assisted test generation
Generates an initial test suite from existing code. Every generated test is reviewed by an engineer before use in the project.
Where AI does not help
AI is useful for repetitive, pattern-based, large-scale work. It is weak where the work depends on commercial judgement, architectural trade-offs, and accountability for delivery.
Architectural decisions in complex systems
If billing, reporting, inventory, and approval logic share state in undocumented ways, AI can support analysis and translation — it cannot decide how that estate should be reshaped safely.
Business decisions inside the modernisation programme
When a migration changes how the business operates, those decisions belong in structured discussion between your team and ours — not inside an AI prompt.
QA, sign-off, planning, risk, and client communication
Every AI-generated output is reviewed by a senior engineer. Project direction remains human-led throughout. Nothing ships unexamined.
A note of caution. A 2025 S&P Global finding reported that 42% of companies had abandoned most AI initiatives, up from 17% — citing data quality, system readiness, and integration failures. Our response is simple: every AI task has a review step, and nothing ships unexamined.
Frequently asked questions
The questions buyers ask before signing off on AI-assisted work.
Do you build everything from scratch using AI?
No. Each project starts from a pre-built application foundation that already includes authentication, responsive design, CI/CD, and internationalisation. AI is used in defined phases to understand the current system faster and improve test coverage; the unique business logic is engineered and reviewed by the senior team.
Is my code safe when AI tools are used?
Yes. Code is submitted via API in isolated sessions under data-processing agreements that exclude it from model training. We sign an NDA before work begins and agree the data handling protocol before any AI tools are applied.
Which AI tools do you use?
GitHub Copilot, Amazon Q Code Transformation, and LLM-based code analysis through GPT-5.5 and Claude APIs. We also use AI-assisted test generation where the task suits it, and every generated output is reviewed by an engineer before use.
How much cheaper is AI-assisted modernisation?
On the tasks where AI fits well, savings are typically 20–40%. In practice, that shows up as lower assessment cost, quicker scoping, faster translation work, and broader test coverage for the same budget.
What if I do not want AI used on my project?
That is fine. AI assistance is optional. The same quality can be delivered with traditional engineering methods, though the work may take longer and cost more. We can discuss the trade-off during the discovery call.
How do I get started?
Book a free 30-minute discovery call. We will discuss your current systems, explain the practical options, and recommend the next step — whether that is a Health Check, a scoping exercise, or direct modernisation.
See how AI-assisted modernisation would work for your systems.
A free 30-minute discovery call. We walk through your situation, explain what AI can and cannot help with, and give you a direct assessment of the options in front of you.