Work that informs the practice

Agentic test generation with coverage closure

A long-lived codebase had accumulated a manual regression suite that was costly to maintain and weighted towards brittle end-to-end tests. Simply asking a code-generating model to fill coverage gaps produced plausible tests, but also duplicated existing behaviour and placed tests at the wrong level.

The team built an agent plugin and composable skills that treat test creation as a closed loop. Existing tests are inventoried before anything new is proposed. Gaps are classified and routed to the appropriate level of the test pyramid. Generated tests are run, revised against real failures, checked with coverage and mutation feedback, and held for human review before merge.

This judgement has a longer history. In 2006, the same underlying problem was addressed with a parser that read call-flow diagrams and generated test plans automatically. The technique changed; the judgement about which tests are worth writing did not.

A unified customer API across every channel

The website, mobile app, retail stores, and call centre had each developed separate integrations with the same back-end systems. Customer journeys varied by channel, while each new vendor created another bespoke integration to maintain.

The work involved designing and delivering a unified REST API layer for internal systems and external vendors. It provided a consistent account and journey view across channels while operating across cloud and on-premise infrastructure. The scope included monetisation flows, customer financing, and location information for stores and payment centres.

Moving routine service enquiries into self-service

Customer care representatives were handling avoidable enquiries about installation appointments, technician arrival, and outages already known to the operator.

The team delivered responsive customer-facing modules for rescheduling and cancelling appointments, following same-day technician status, and notifying customers about outages affecting their area. The experience worked across mobile, tablet, and desktop screens.

A shared middle tier for digital channels

Android, iOS, and browser applications were integrating directly with different back-end systems for television recording, home phone activity, and feature management. Each channel repeated the same logic, and back-end changes created failures in several places.

The team delivered a middle-tier application that consolidated those systems behind a common set of web services, giving every channel a stable contract to build against.

Stabilising a platform with recurring downtime

Applications were causing repeated performance degradation and server downtime. Diagnosis was slow because logging varied across the codebase and was tightly coupled to a particular implementation.

The engagement covered code re-engineering and performance tuning, alongside an aspect-oriented logging framework that separated application code from the logging library and applied consistent entry, exit, and unhandled-exception reporting. The work also defined code review, coding, and exception-handling standards for later projects.