Accelerating Business Simulation Platform Development with AI-DLC

Introducing structured AI-DLC

Brain Station 23 evolved the project’s AI-assisted development approach through several stages.

Initially, AI was used to create project documentation. As requirements changed, the team realized that documentation could quickly become outdated.

The approach was then reversed. AI was used to audit the existing codebase and create documentation that reflected what had actually been implemented.

Finally, the team introduced a structured context/ repository that became the live source of project knowledge.

New work followed a defined process:

Story / Change Request → AI analysis → Implementation → Verification → Context update

The AI agent was required to consult the relevant project context before making code changes.

Connecting requirements to implementation

The structured context included:

• Business rules
• Module specifications
• API contracts
• User stories and acceptance criteria
• Architecture Decision Records
• Change Requests
• Development guides
• Formula references
• API audits

Documentation was also linked back to the implementation through verification references.

A “code wins” policy was introduced so that when older documentation conflicted with the implemented system, the conflict would be identified and the current code would be treated as the source of truth.

AI-assisted impact analysis

Before implementing significant changes, AI was used to trace affected modules and dependencies.

This was particularly valuable for features involving scoring, accounting, permissions, and the different CEO-only and all-member gameplay modes.

Instead of relying on manual code archaeology, developers could use AI to map the potential impact across backend services, APIs, frontend portals, and infrastructure.

AI-assisted debugging

AI was also used to investigate production issues by reading actual code paths rather than relying on assumptions.

One important example involved a production issue where one player’s Round 2 balance sheet and cash flow remained frozen at Round 1 values while the income statement was correct.

The issue crossed accounting, engine orchestration, result persistence, and frontend presentation.

AI traced the relevant data sources and identified silent exception handling, non-idempotent recomputation paths, and accounting plug lines that were masking failures.

The resulting analysis was converted into a formal Change Request, Architecture Decision Record, and five user stories for controlled implementation.

AI-assisted testing and code review

AI was used to draft tests from business rules and formula audits, particularly for high-risk areas such as accounting, scoring, RBAC, and tenancy isolation.

It was also used to pre-map affected files and flows during code review, allowing developers to focus their attention on accounting logic, permissions, regression risk, and other areas requiring human judgement.