Repeatability
Medium
The pattern of finding inline instantiations and mapping them to Hilt modules is structurally repeatable, but each codebase has unique dependency graphs, naming conventions, and lifecycle quirks that require case-by-case judgment. The task is not fully templated.
Ambiguity Tolerance
Medium
Success criteria are partially clear — identify instantiation points, generate modules, produce a checklist — but 'correct' Hilt scoping and module organization involve architectural opinions that aren't objectively verifiable without running and testing the app.
Data & Tool Availability
Medium
The agent needs full read access to the Kotlin source files, which can be provided, but it cannot run the app, execute tests, or inspect runtime behavior to validate its analysis. Static analysis alone may miss dynamic patterns.
Error Cost
Medium
The output is a plan and checklist, not directly executed code, so errors are caught before deployment — junior devs following a flawed checklist could introduce bugs, but the risk is contained by code review and testing. Not irreversible, but costly to unwind if widespread.
Human Judgment Required
Medium
Scoping decisions (Singleton vs. ViewModel vs. Activity scope), handling of legacy singletons, and sequencing refactoring steps to avoid breaking the app mid-migration require architectural intuition a senior Android dev brings. AI can approximate but not reliably substitute.