Conductor for Gemini CLI: Taming AI's Amnesia in Software Development

Every time you open a chat with an AI coding assistant, you're essentially talking to a goldfish. A brilliant, code-generating goldfish that forgets everything about your project the moment you close the window. Google's Conductor extension for Gemini CLI, released in preview on December 17, 2025, isn't content with this state of affairs. It's not here to incrementally improve AI coding, it's here to fundamentally change how we think about AI in software development.

Conductor introduces Context-Driven Development (CDD), a methodology that treats your project's context as a first-class citizen. Instead of relying on ephemeral chat sessions, CDD mandates a structured workflow where your project's specifications, architecture decisions, and coding standards live in version-controlled Markdown files. The AI reads these files on every run, ensuring it always has the full picture before touching your codebase.

This isn't a minor quality-of-life improvement. It's a seismic shift in how we approach AI-assisted development. Here's the core thesis: By forcing developers to codify their project's context, Conductor doesn't just make AI more consistent, it makes human developers more deliberate and aligned.

Let's break down how Conductor actually works:

The Command Layer is what you interact with, six slash commands inside Gemini CLI. The Artifact Layer is a conductor/ directory in your repo containing Markdown files that hold project state. The Version Control Layer is Git, which Conductor uses to create per-task commits and support rollbacks.

Setting up Conductor requires Gemini CLI, a Google API key or Vertex AI setup, and an initialized Git repository. Installation is straightforward:

gemini extensions install https://github.com/gemini-cli-extensions/conductor --auto-update

The real test of commitment comes when you run /conductor:setup. This command analyzes your existing codebase (if any) and guides you through creating six critical context files:

1. product.md: Vision, users, goals, and success criteria
2. product-guidelines.md: UI standards and error handling behavior
3. tech-stack.md: Your chosen languages, frameworks, and infrastructure
4. workflow.md: Development preferences like TDD and commit strategies
5. code_styleguides/: Language-specific coding standards
6. tracks.md: A registry of all feature 'tracks' (initially empty)

This process isn't quick. It's not meant to be. It forces you to articulate aspects of your project that may have been implicit or inconsistent. And that's the point.

With this context in place, starting a new feature becomes a structured process. The /conductor:newTrack command creates a specification for your feature, ensuring the AI understands exactly what you're trying to build before it writes any code.

The key insight here is that Conductor isn't just a thin wrapper around Gemini, it's a new development philosophy. It forces developers to think deeply about their project's architecture and goals upfront. This isn't just about making AI more effective; it's about making humans more effective by demanding clarity and intentionality.

However, this power comes with a steep learning curve. Developers accustomed to quick, ad-hoc AI coding assistance will need to adapt to a more deliberate, planning-focused workflow. The upfront time investment in creating and maintaining context files will feel burdensome, especially for those used to diving straight into code.

Is Conductor worth this shift in approach? For small, throwaway projects, absolutely not. It's overkill. But for any substantial codebase, especially those with multiple contributors, Conductor's ability to keep AI assistance aligned with project goals and standards could be transformative. It's not just about writing code faster; it's about writing the right code, consistently, across an entire team.

As AI becomes an increasingly integral part of the development process, tools like Conductor that help manage and direct that AI assistance will become essential. But they come with a challenge: Are we, as developers, ready to be as rigorous and intentional as we're asking our AI assistants to be? Conductor doesn't just promise better AI coding, it demands better human planning. And that might be its most disruptive feature of all.