Smift

Smift turns IBM Bob into a governed AI development team that plans, codes, reviews, and proves software work across IDE, CLI, web, chat, and voice. Smift is primarily a painkiller for engineering teams that need faster delivery without losing control. It also functions as a vitamin for daily developer productivity once the governed workflow is already trusted. Main User Conflict Developers want to move quickly with AI, but unmanaged AI-assisted coding can skip planning, hide assumptions, miss security issues, and leave no evidence for reviewers or judges. Motivation Users look for Smift when they need repository-aware AI help that is not a black box. They want speed, but they also need control, reviewability, reproducible evidence, and a way to trigger work from the places they already use. Smift is a governed agentic development platform built for teams that want AI speed without losing engineering control. It uses IBM Bob as the visible development partner across the full software lifecycle: Ask mode for repository understanding, Plan mode for requirements and architecture, Code mode for scoped implementation, Review mode for quality and risk findings, Verify mode for tests and checks, and Evidence mode for exported proof. The project solves a core problem in AI-assisted development: many tools can generate code, but teams still need planning boundaries, review discipline, approval gates, test evidence, and a clear record of what the AI actually did. Smift turns IBM Bob’s repository context into safe developer action. It connects IDE, CLI, web, chat, mobile, and voice workflows through a governed architecture with workflow phases, policy checks, evidence exports, and reviewable outputs. For the IBM Bob Challenge, Smift demonstrates how Bob can improve everyday software delivery, not just by writing code, but by helping teams move from idea to planned, implemented, reviewed, verified, and auditable work.