Conductor

In agentic coding conversations, 'conductor' describes a specific mode of working: the engineer stays tightly in the loop with one AI agent, steering it step by step, adjusting prompts, catching mistakes, and iterating in real time inside their editor or terminal (CLI means command-line interface). It's the evolved version of AI pair programming. The engineer isn't writing every line anymore, but they're watching every move. This is synchronous and focused: one human, one agent, one task at a time.

The term also applies to a technical pattern inside multi-agent systems. Here, a 'conductor' is the central agent (or model) that receives a high-level goal, breaks it into subtasks, selects the right specialist agents to handle each piece, and manages the flow of information between them. It separates planning from execution: the conductor thinks about what needs to happen and in what order; the worker agents just do their part. Research has shown conductor-based architectures can outperform static multi-agent setups by adapting task routing dynamically.

In May 2026, Microsoft released an open-source tool also called Conductor, a YAML-first CLI (command-line tool) for defining multi-agent workflows deterministically. Instead of letting an LLM decide the routing at runtime, you declare the workflow structure upfront in a config file, keeping orchestration inspectable and token-free. It supports mixing models from different providers in a single workflow and integrates with MCP servers (Model Context Protocol, a standard for giving agents access to tools). Whether you're thinking about the human role, the agent pattern, or the tooling, 'conductor' points at the same underlying shift: someone or something has to hold the baton.