Measure twice, code as soon as
Benjamin Franklin stated: “Failing to plan is planning to fail”. But, within the age of AI, we regularly dive straight into implementation with out establishing a transparent understanding of what we’re constructing. Conductor, a brand new extension now out there in preview for Gemini CLI, adjustments this workflow through the use of context-driven growth. Reasonably than relying on impermanent chat logs, Conductor helps you create formal specs and plans that dwell alongside your code in persistent Markdown information. This lets you plan earlier than you construct, evaluate plans earlier than code is written, and maintain the human developer firmly within the driver’s seat.
The philosophy behind Conductor is straightforward: management your code.
As a substitute of diving straight into implementation, Conductor helps you formalize your intent. It unlocks context-driven growth by shifting the context of your challenge out of the chat window and instantly into your codebase. By treating context as a managed artifact alongside your code, you remodel your repository right into a single supply of fact that drives each agent interplay with deep, persistent challenge consciousness.
Conductor lets you:
- Plan earlier than you construct: Create specs and plans that information the agent for brand new and present codebases.
- Keep context: Guarantee AI follows model guides, tech stack decisions, and product targets.
- Iterate safely: Assessment plans earlier than code is written, conserving you firmly within the loop.
- Work as a workforce: Give your AI brokers the identical challenge and finest practices context.
- Construct on present tasks: Use your present code to tell design choices.
Help for “brownfield” tasks
As builders ourselves, most of our work includes established codebases (“brownfield”). That is usually the place AI instruments wrestle, missing the nuanced understanding of a challenge’s historical past and structure.
Conductor approaches this problem with a practical, context-driven philosophy. When launched to an present challenge, Conductor initiates an interactive session that can assist you create a foundational set of paperwork about your challenge’s structure, tips, and targets. As you construct new options and tackle new duties, Conductor updates this shared context, guaranteeing its information grows alongside your challenge.
That is simply our first step. You’ll see important enhancements on this space within the coming months as we proceed to spend money on making Conductor an indispensable companion to your present tasks.
Conductor for groups
Conductor lets you set project-level context to your product, your tech stack, and your workflow preferences. You’ll be able to set these preferences as soon as, they usually develop into a shared basis for each function your workforce builds. For instance, groups can outline a longtime testing technique that may routinely be utilized by Gemini.
By centralizing your technical constraints and coding requirements, you make sure that each AI-generated contribution adheres to your particular tips, no matter which developer runs the command. This shared configuration accelerates onboarding for brand new workforce members and ensures that options constructed by totally different individuals really feel like they have been written by a single, cohesive engineering workforce.
How Conductor works
Conductor is a structured workflow for agentic growth which is good for duties extra advanced than easy code edits. In contrast to an ordinary chat session which is confined to a single session, Conductor makes use of a set of Markdown information to plan and observe progress over time. These Markdown information persist in your repository, enabling you to pause and resume work and transfer between machines. Right here’s a high-level overview of the way it works.
1. Set up context
Once you run /conductor:setup, Conductor helps you outline the core parts of your challenge. This context can then be used for constructing new parts or options by you or anybody in your workforce.
- Product: Outline your customers, product targets, and high-level options for the general challenge.
- Tech stack: Configure your preferences for language, database, and frameworks.
- Workflow: Set your preferences for a way your workforce capabilities, corresponding to following test-driven growth.
2. Specify and plan
Once you’re able to tackle a brand new function or bug repair, run /conductor:newTrack. This initializes a observe—our time period for a high-level unit of labor. As a substitute of instantly writing code, Conductor helps you generate two essential artifacts to your observe:
- Specs: The detailed necessities for the precise job. What are we constructing and why?
- Plan: An actionable to-do record containing Phases, Duties, and Sub-tasks.
Conductor will stroll you thru creating every of those artifacts and can recommend solutions primarily based on the context it has that can assist you shortly construct high-quality specs and plans.
3. Implement
When you approve the plan, run /conductor:implement. Your coding agent then works by the plan.md file, checking off duties because it completes them. As a result of the state is saved in a file, you possibly can cease, seize espresso, and resume later with out dropping your house. We’ve additionally included checkpoints for reverting to a earlier model, in addition to the flexibility to edit the plan mid-flight.
Get began
We consider that context-driven growth results in increased high quality outcomes for advanced tasks. By treating your documentation because the supply of fact, you empower Gemini to behave as a real extension of your engineering workforce.
Set up the extension and begin utilizing Conductor at the moment right here or through the use of the next command:
gemini extensions set up https://github.com/gemini-cli-extensions/conductor
