Gemini Conductor: Google's New Tool is Here To Fix AI Coding

The video opens by positioning Gemini Conductor as Google's latest context engineering workflow tool, but immediately reveals it's not better than existing alternatives. The creator establishes credibility by mentioning previous coverage of similar workflows on the channel. A brief sponsor message introduces Automata, a service that helps non-technical founders build AI-powered apps and websites using the workflows taught in their videos. The sponsor positions themselves as a technical co-pilot that can turn ideas into working solutions without the need to hire or manage a development team.

Conductor is explained as an extension for Gemini CLI that bundles commands, MCPs, and rules into a shareable package (similar to Claude's plugins). The setup process begins with installation via a command, followed by five slash commands that control the workflow. The /setup command checks for existing Conductor files and initialization state, then creates 'tracks' (their version of stories) instead of traditional user stories. The system initializes a GitHub repo and prompts for project details. The creator tests it with a production-ready, scalable project to evaluate whether Conductor would recommend appropriate architecture choices. The tool creates a product.md file containing the project concept and asks clarifying questions, though these questions are described as simplistic and not particularly useful.

After approving the product guide, Conductor creates product guidelines focused on styling and design principles. The technology stack definition phase reveals one of the workflow's major flaws - despite knowing the full project requirements, it recommends an inappropriate tech stack that needs manual correction. The system includes code style guides for specific languages, automatically adding relevant ones to the project during initialization. Default workflow settings include 80% code test coverage, which is actually a positive feature. The tool automatically writes tests alongside base components, tests completed tasks, commits changes after every task, and uses git notes for tracking issues - all good practices in theory.

Conductor creates high-level product requirements to generate the first track, but this initial track is too broad and needs breaking into smaller, more manageable pieces. The /implement command starts the actual work, referencing tracks in the tracks folder. Each track contains two files: spec.md (with objectives and technical details from the tech stack) and plan.md (with specific implementation tasks). The tracks.md file uses status indicators - empty means not started, a marker indicates in-progress, and another marker shows completion. Additional commands include /status for progress reports, /new-track for adding features, and /revert for git-aware rollback. The creator also tested Conductor on a pre-existing repository, where it analyzed existing files and asked clarifying questions without creating redundant tracks.

Despite reasonable file management structure, the command files and instructions are poorly written and don't properly manage the context loop. During setup, Conductor failed to properly dissect the project idea, requiring extensive manual guidance. When defining the technology stack, it missed crucial components for a scalable app with large user capacity, only correcting after explicit instruction. The database schema generated for the initial track was completely incomplete, missing crucial elements necessary for app setup. The creator suspects Gemini itself is a capable model, but the implemented commands are causing poor behavior. The biggest problem occurred when switching from NPM to PNPM - Conductor attempted to create a backup but instead deleted the entire conductor folder containing all planning files, then tried to reconstruct everything from memory rather than following proper context workflow practices.

During actual implementation of the first track, Conductor initialized the project and core services but failed catastrophically at basic tasks. When configuring Supabase environment variables, it automatically marked the task complete while inserting dummy API keys, never asking the user to set up a Supabase project or provide real credentials. It then attempted to push the database schema with invalid keys, failed, and only then asked for verification. This demonstrates that tasks aren't being properly updated or followed correctly. The creator concludes that Conductor shouldn't be used for end-to-end spec development in its current state, recommending Be Mad as a superior alternative. For small projects, the creator prefers making custom context files. The video ends with a call to support the channel via Super Thanks and a promise of future content.