AI makes it inexpensive to produce another version of a trading tool. That does not make a rewrite the correct choice. The decision should depend on whether the intended behavior is clear, the current source is understandable, and the architecture can support the required operating conditions.
Repair when the defect is bounded
Repair is sensible when the source is available, rules are clear, the problem is reproducible, and the surrounding structure is stable. Examples include a compile issue, one incorrect calculation, a known alert problem, or a broker-specific constraint that can be isolated and tested.
Rebuild when assumptions are embedded everywhere
A rebuild may be safer when source is incomplete, AI-generated fragments conflict, state is hidden across duplicated branches, requirements were never written, or order and recovery behavior cannot be added without destabilizing the system. Preserve verified behavior and evidence instead of blindly translating old code.
Continue with AI when the feedback loop is strong
AI can remain useful for bounded functions, tests, documentation, refactoring proposals, and prototypes when a qualified owner reviews the output. Use small changes, version control, repeatable tests, and explicit acceptance criteria. Do not ask AI to decide undocumented trading behavior.
Test further when the answer is not yet known
Sometimes neither repair nor rebuild is justified. Create a minimal reproduction, compare historical and real-time examples, add logs, or test the platform assumption first. A small diagnostic can prevent an expensive decision based on guesswork.
Use an independent review
The person who prompted or generated the code may be anchored to its apparent success. A review should compare the source with written rules, platform behavior, failure cases, tests, and operational needs.
Start with the AI-Built Trading System Audit for existing AI-assisted code. Use the Strategy Scope Review when the rules and project direction need clarification before code review.
Risk note
This decision framework concerns software quality and technical risk. It does not validate a trading strategy or guarantee performance, safety, or live results.