How Everforth Quinnox’s Governed AI-Assisted Engineering Helped Upgrade Five Critical Application Workflows for 20,000 Drivers
Client Overview
A Fortune 500 environmental-services enterprise operates a long-running Android application used by approximately 20,000 drivers, with approximately 17,000 active on a typical day. Managed devices installed in field vehicles support critical daily work, including authentication, vehicle checks, safety communication, performance visibility, route activity, operational events and end-of-day completion.
Everforth Quinnox was engaged to support a technical platform upgrade while maintaining the existing user experience, workflows and field-visible behavior. The engagement covered five connected application flows that needed to move toward a more reusable and testable architecture.
The work established a repeatable delivery model across five connected flows, 450 AI-generated test methods and 169 defined internal checks before client-owned validation.
Business Challenges
The engagement incorporated approved enterprise AI-assisted engineering into the platform upgrade while maintaining strong human oversight.
The key objectives were to:
- Improve source-backed understanding of complex legacy code and dependencies.
- Use AI assistance across bounded analysis, implementation, testing and documentation activities.
- Establish a repeatable approach to AI-assisted engineering across the delivery team.
- Build an automated test foundation for the upgraded scope.
- Maintain architecture, security, privacy and code-ownership controls.
- Preserve the existing field experience while upgrading the underlying platform.
- Ensure production changes were supported by sufficient technical and operational evidence.
Technology/Operational Challenges
Complex legacy dependencies
Changes that appeared locally could affect session state, background processing, navigation, route activity and other downstream application behavior.
Limited automated test foundation
The legacy scope began without an assessed unit-test base, making it difficult to establish fast, repeatable feedback when modifying the application.
Compatibility constraints
The upgraded application needed to coexist with established build, device and dependency requirements.
AI adoption and trust
AI assistance was new to parts of the engineering team. Without shared instructions and evidence gates, plausible AI-generated output could be accepted before the underlying source, architecture or acceptance criteria were verified.
Data and security concerns
The use of AI assistance required clear boundaries around credentials, Personally Identifiable Information, operational details, prompts, logs and telemetry.
Maintaining the existing field experience
The technical platform upgrade had to happen without disrupting the workflows and experience relied upon by field users.
Everforth Quinnox Solution and Approach
Everforth Quinnox established a structured delivery model that embedded AI assistance across the engineering lifecycle while keeping accountable engineers in control.
How We did it
Our team worked through a controlled Understand → Decide → Implement → Review → Validate cycle for every feature. Here’s how:
1.Understand
The team first analyzed the existing application, tracing classes, methods, state, configuration, services and dependencies.
AI assistance was used to summarize legacy behavior, identify potential dependencies and surface areas requiring further investigation.
Engineers validated these findings against the source code and runtime behavior.
2.Decide
AI was used to explore implementation options and highlight technical trade-offs.
Architects and engineers retained responsibility for selecting the appropriate architecture, compatibility approach and implementation boundaries.
3.Implement
AI assistance helped draft and refine Kotlin code and documentation and generated the unit-test methods for the in-scope upgraded area.
Developers remained responsible for the resulting code and reviewed, corrected or rejected AI-generated output where required.
4.Review
AI assistance helped identify potential risks, inconsistencies and possible improvements.
Pull Request reviewers independently evaluated the changes before approving them.
5.Validate
AI assistance supported test design, unit-test generation, test execution and result reporting while human reviewers retained ownership of quality decisions.
Quality Assurance and release stakeholders remained responsible for acceptance and release approval.
Governance and Security
Everforth Quinnox established repository-level instructions and approved prompt patterns to control how AI assistance was used.
Application-level controls were also strengthened around sensitive information, including filtering, masking and sanitization of relevant logs and telemetry.
The approach incorporated relevant Open Worldwide Application Security Project logging guidance without claiming formal certification.
AI could recommend or generate changes. Accountable engineers remained responsible for architecture decisions, code ownership, Pull Request approval, test adequacy, Quality Assurance acceptance and release decisions.
Building an AI-ready Engineering Practice
Rather than relying on ad hoc prompts, the team created reusable prompts incorporating:
- Business and technical context
- Role and task
- Source information
- Constraints
- Examples
- Acceptance criteria
- Expected evidence
This helped establish a consistent approach to AI-assisted development across the team.
Because AI-assisted engineering was new to parts of the team, adoption was reinforced through shared repository instructions, reusable prompts, a recorded enablement session and ongoing review feedback during development.
Business Outcomes Delivered
The engagement strengthened the delivery and validation foundation while progressing the upgraded scope from source analysis through controlled enterprise production use.
| Observed result | Evidence | Business meaning |
|---|---|---|
| Shared adoption | Six licensed users worked from a common operating model | Assistance became a team practice rather than isolated experimentation |
| Production proving ground | Five connected flows progressed from bounded pilots through full-site rollout and enterprise production | The delivery model operated under real constraints at increasing scale |
| AI-generated test foundation | AI assistance generated 450 test methods across 47 test files for the in-scope upgraded area; engineers reviewed and executed the tests | Created repeatable technical feedback while keeping validation and acceptance human-owned |
| Internal readiness | 169 of 169 defined internal checks passed before client Quality Engineering: 123 platform-feature checks and 46 selected end-to-end regression checks | Provided a complete internal evidence set before client-owned validation |
| Staged field progression | Three pilot devices expanded to approximately 140 field users before enterprise production for approximately 20,000 drivers, with approximately 17,000 active on a typical day | The upgraded flows moved from demonstration scale into an enterprise operating context |
Quantifiable Impact
The engagement demonstrated that AI-assisted engineering can support change in a complex legacy application without turning generated output into an unchecked production dependency.
The value came from combining AI assistance with source context, accountable engineering decisions, structured testing and controlled validation.
The resulting delivery model enabled the team to use AI across legacy analysis, implementation, test generation, documentation and review while retaining human accountability for architecture, security, quality, compatibility and release decisions.
The engagement did not run a controlled before-and-after study to isolate a tool-only productivity percentage; the observed outcomes reflect AI assistance working together with architecture, engineering, Quality Assurance and controlled rollout.
The advantage was not access to an AI coding assistant. It was the delivery discipline built around it.