MOKKA SOFTWARE / AI-READY LEGACY SOFTWARE
Bring your legacy
software into
the AI era.
Prepare the context and checks
that coding agents need.
When a working system carries years of undocumented decisions, coding agents need reliable context and a way to check changes. We help prepare existing software for AI-assisted maintenance and development by reconstructing selected behaviour, validating business rules and establishing repeatable verification.
Discuss a pilot
WHAT WE PREPARE
Make the system understandable.
Make changes verifiable.
The work prepares your development process and agent context. AI functionality inside the product is a separate scope. For a selected part of the system, we agree six concrete outputs.
Reconstructed specification
Trace selected user journeys, interfaces and data flows through the code and available documentation. Record what the system does today, including assumptions and gaps, with references back to the evidence.
YOU RECEIVEA specification of observed behaviour and a system map for the agreed scope.
Business-rule register
Extract candidate rules, calculations and exceptions, then review them with people who understand the business. Distinguish required behaviour from inherited quirks, defects and rules whose intent is still unclear.
YOU RECEIVEA rule register with sources, validation status and open questions.
Domain knowledge for agents
Organise domain vocabulary, module responsibilities, approved rules and repository guidance into context that coding agents can use. Keep source references and a process for updating that knowledge as the software changes.
YOU RECEIVEA versioned domain knowledge pack and agent instructions.
Testability and baseline tests
Identify how to run the selected code, control test data and isolate external dependencies. Add focused tests that capture agreed behaviour and make the chosen change scenarios reproducible.
YOU RECEIVEA repeatable test setup, fixtures and baseline tests.
Automated verification
Connect builds, relevant tests and selected checks into an automatic verification step. Make pass criteria and failures visible so agents and developers receive evidence before a change is accepted.
YOU RECEIVEA verification pipeline with explicit pass/fail results.
Agent development workflow
Define how an agent receives a task, retrieves context, proposes a change and responds to failed checks. Specify allowed actions, escalation points and the human review required to accept changes.
YOU RECEIVEA documented workflow and a representative pilot change.
Reconstructed behaviour shows what the system currently does. Your team validates which rules are intended and which behaviour should change. We label observations, inferences and confirmed requirements separately, retain open questions and keep the deliverables within the agreed scope. Any gains measured in the pilot apply to the evaluated tasks.
A BOUNDED PILOT
Start with one part
of your system.
Choose one component or business flow and a small set of representative changes. Potential benefits include less repeated investigation, more consistent verification and lower review effort. We test these against your current process before deciding whether to extend the approach.
Choose scope and baseline
Agree the code, business reviewers, build environment and test data. Record how comparable changes are currently prepared, implemented and reviewed.
Prepare context and checks
Produce the selected knowledge and rule artefacts, improve the test setup and connect verification to the agent workflow.
Evaluate a representative change
Run the workflow on agreed maintenance or development tasks. Review the evidence with your team and decide what to improve, extend or stop.
POTENTIAL BENEFITS / EVALUATION
Measure the result.
Compare it with today's process.
- Correctness and regressions
- Check changes against client-validated requirements and agreed regression cases. Record failures, gaps in coverage and findings that still need human interpretation.
- Review and rework effort
- Compare reviewer effort, correction cycles and human interventions with the baseline. A generated change counts as accepted only after the agreed checks and review.
- Repeatability
- Repeat selected builds, tests and workflow runs. Track flaky checks, environment failures and whether the same evidence can be reproduced.
- Total time and tool cost
- Measure the whole task, including context preparation, tool runs, verification, review and rework. Record model and tooling costs alongside developer time.