#coffee tracking Startups & Tools

AI coding agents promise rapid development, but they struggle with a fundamental challenge: distinguishing completed work from merely plausible-sounding code. Factory directly addresses this gap by inserting verification gates between intention and delivery. The product wraps AI coding workflows in a structured process that treats each ticket as a journey through nine discrete phases. Rather than trusting an agent's assertion that work is done, Factory demands evidence. It translates informal requests into testable acceptance criteria, isolates changes in dedicated worktrees, and enforces gates where claims must be supported by passing tests and fresh-context review before proceeding. What distinguishes Factory is its refusal to be optimistic. When acceptance criteria aren't fully satisfied, the system blocks delivery with explicit reasons. The proof edition demonstrated in the materials shows this in action: a false delivery claim fails because open blocking questions remain unanswered and acceptance criterion statuses are invalid. That kind of honest gating is rare in tools designed to accelerate development. The workflow itself is thoughtful. The process begins with intake and diagnosis to establish baselines, moves to human-approved planning where blockers must be resolved before proceeding, then transitions to agent-driven building with failing tests first. Final proof phases map each acceptance criterion to evidence before a terminal gate either approves or rejects the change. Only state.json persists between sessions, forcing decisions to be recorded rather than ephemeral. The design includes deliberate friction points. Work happens in isolated worktrees. Human approval is required at specific gates. Tests must fail before implementation can begin. Reviews happen with fresh context. These constraints work against speed but directly serve the core problem: unverified delivery masks real issues. Factory operates as an open-source agent skill for Claude Code, which positions it as a specialized workflow rather than a standalone platform. That architecture keeps the tool focused and integrated into existing developer environments. The fundamental contribution here is structural rather than technological. Factory doesn't replace code review or testing; it systematizes when each happens and what triggers progression. For teams where AI agents write code at scale and false positives create compounding problems, that discipline addresses a genuine pain point.