CodePilot – AI-Assisted Developer Workflow

Developers often spend significant time debugging repository-level issues because the problem is rarely limited to a single line of code. They must understand the bug report, trace the relevant API and service layers, identify the root cause, implement a fix, run tests, investigate regressions, and verify that the system works correctly. CodePilot is an AI-assisted developer workflow built around IBM Bob 2.0 and demonstrated on a real FastAPI application called TaskFlow API. It structures debugging into a verification-first workflow: Bug Report → Investigate → Evidence → Tests → Verify. The workflow starts with structured bug reports from a small synthetic, team-created dataset. A developer selects an issue and CodePilot presents the affected code, relevant source files, expected versus actual behavior, and the investigation path. IBM Bob 2.0 was used during development to inspect the repository, trace bugs across API/service/repository layers, identify root causes, implement fixes, and review regression scenarios. The prototype demonstrates three debugging scenarios, including an incorrect project-progress calculation, an ignored task-status update, and swapped task-filter arguments. For example, the progress bug produced 60% instead of the expected 40% completion for a project with 5 tasks and 2 completed tasks. The root cause was an inverted condition, which was corrected and verified. CodePilot also executes the project's actual pytest suite rather than displaying simulated results. The final suite contains 41 tests, including 10 additional regression tests, with 41 passed and 0 failed in the demonstrated environment. The target users are software developers and development teams working with existing repositories. CodePilot provides a structured way to move from a reported issue to evidence-backed verification while preserving the development evidence produced with IBM Bob 2.0.