Problem solved

Development teams and individual developers often want to leverage multiple AI coding assistants—each with unique strengths—but manually coordinating them leads to chaos, conflicting changes, and quality issues. Bernstein solves this by providing a unified orchestrator that plans tasks upfront, isolates execution in git worktrees, schedules them deterministically, and enforces quality gates, thus enabling safe, parallel, and reproducible multi-agent development.

Use cases

• Automating a multi-step refactoring across a large codebase by distributing different modules to specialized agents.
• Parallelizing feature implementation where each agent works on a separate component in its own isolated environment.
• Integrating into a CI/CD pipeline to automatically plan and execute code changes from a GitHub issue with quality checks.
• Running safe experiments to compare how different AI agents handle the same coding task without affecting the main branch.
• Bootstrapping a new project by generating boilerplate, tests, and documentation concurrently using multiple agents.

Audience

• Developers
• ML Engineers
• Researchers

Pros

• Deterministic scheduling ensures reproducibility and easier debugging.
• Git worktree isolation prevents conflicts and simplifies rollbacks.
• Single LLM plan call reduces cost and complexity compared to per-agent calls.
• Supports 40+ CLI agents out of the box, with extensibility for custom agents.
• Open source with an active community and GitHub Action for easy CI integration.

Cons

• Planning quality depends on a single LLM call, which may produce suboptimal task breakdowns.
• Only supports CLI-based agents; agents requiring GUI or API-only interaction are not compatible.
• Setup and configuration may be required for each agent, adding initial complexity.