Descriptions:
Charlie Holtz, co-founder and CEO of Conductor (YC Summer 2024), walks through his personal daily setup for managing multiple AI coding agents in this Y Combinator workflow video. Conductor is a Mac application that orchestrates parallel Claude instances as separate git worktree-based workspaces, each producing a PR for human review before merging — and Holtz uses it to build Conductor itself.
Holtz describes his workflow in concrete detail: launching tasks by whispering into a $20 gooseneck microphone using the Spokenly transcription app (which runs the local Parakeet speech model), reviewing agent-generated pull requests with inline GitHub-style comments, and maintaining several experimental workspaces in parallel — most of which never ship. He previews two upcoming features: cloud workspaces and a new dashboard that surfaces all active agent work in one view, designed to give users the experience of being “the CEO of a small company” directing and reviewing agent output.
Beyond the product demo, Holtz shares deliberate philosophy on human-AI collaboration in software development: the importance of humans making architectural and UI decisions rather than delegating them to AI, structuring codebases so that core infrastructure remains human-authored while peripheral features are AI-generated, and enforcing workflow constraints (like mandatory PR creation) to maintain quality and traceability. He also discusses his language stack (Elixir and TypeScript), his approach to running local models on a 128GB RAM machine, and why he recently ordered the lowest-spec MacBook to stress-test Conductor on constrained hardware.
📺 Source: Y Combinator · Published June 04, 2026
🏷️ Format: Workflow Case Study
