Google has announced Conductor, a new extension for Gemini CLI designed to transform AI-assisted development workflows. The tool introduces a "context-driven development" approach that moves away from ephemeral chat logs in favor of persistent, formal documentation.
Conductor's core principle rests on storing project specifications and plans as Markdown files directly within the codebase. This approach ensures that project context remains accessible and consistent for all team members, unlike AI conversations that disappear after each session.
The workflow is structured around three distinct phases. The first phase uses the /conductor:setup command to establish the project's foundational context. This step covers product definition, the chosen technology stack, and the team's workflow preferences. This information becomes the baseline for all subsequent interactions with the AI.
The second phase, triggered by /conductor:newTrack, initiates work on a new feature. The tool then generates detailed specifications and actionable task plans. This step structures the work before any implementation, favoring upfront reflection over improvised coding.
The third phase uses /conductor:implement to execute the established plan. The tool offers significant flexibility: developers can pause work, resume it later, or modify the plan along the way based on discoveries made during implementation.
Conductor addresses a specific need that is often overlooked: support for brownfield projects, meaning existing codebases. For these projects, understanding historical context and existing architecture proves crucial. The tool allows this knowledge to be documented in a structured way, facilitating the onboarding of new developers and relevant AI assistance.
The extension also allows teams to establish shared technical standards and coding guidelines. Once documented, these conventions steer AI suggestions toward practices consistent with the team's choices.
This announcement fits within the broader trend of structuring AI interactions in software development, joining similar approaches such as AGENTS.md files or skill systems. The extension is available for download on GitHub.