CHAPTERS
Conductor in one sentence: a Mac app for running multiple coding agents in parallel
Charlie explains Conductor as a desktop app that orchestrates multiple coding agents (e.g., Claude Code, Codex) from one interface. The core workflow is: create an isolated copy of the repo, assign tasks to agents, then review and merge their work.
Business momentum: rapid adoption from indie hackers to public companies
The founders share that Conductor is growing quickly despite a small team. Adoption spans from solo builders to engineers inside large public companies, motivating them to scale hiring and execution.
Launch announcement: Conductor Cloud keeps agents running when your laptop closes
They introduce Conductor Cloud, moving from purely local execution to cloud-backed workspaces. This allows long-running agent tasks to continue even when the user’s Mac sleeps or shuts.
The real bottleneck: humans can only supervise ~3–5 agents at a time
Charlie frames the next challenge as cognitive and interface-driven rather than purely compute. Conductor proved multi-agent coding can work, but scaling to many more agents requires new abstractions to manage complexity.
Origins: manual multi-agent workflows, repo clones, and discovering worktrees
Before the product, they attempted to run multiple agents manually and hit friction. They evolved from multiple repo clones to worktrees, then gradually built the primitives that became Conductor to remove coordination pain.
Founders’ story: meeting in college, shared intensity, and the “elite team” mindset
Charlie and Jackson recount meeting in college and reconnecting later while working in SF (Netflix and Replicate). They describe a culture inspired by competitive ultimate frisbee—high-commitment teamwork against big competitors.
YC application idea that didn’t stick: AI booking reservations via browser control
Their initial YC concept was a consumer assistant that could book reservations (even tennis courts), powered by early browser automation. They quickly learned it was a “solution in search of a problem,” lacking deep conviction.
YC batch reality: rapid prototyping, frequent pivots, and learning by building
During YC they iterated through many ideas, shipping prototypes extremely fast and regularly changing direction. Advisors noted their build speed as exceptional, even if it took time to find a sticky problem.
The key advice that narrowed everything: build dev tools you personally want
A pivotal office-hours moment reframed their search: dev tools fit their strengths, and they should build for themselves. They adopted the mantra “Make something these guys want,” focusing on their own daily pain.
False start to precursor: trying to build “the thing after the IDE,” then Chorus
They tried an ambitious post-IDE interface inspired by tools like Aider, but models weren’t yet reliable enough and required too much hand-holding. They then shipped Chorus—a multi-model chat app—partly to force user feedback and keep momentum.
Conductor’s first prototype: built in days, shipped in weeks, validated by “magic moments”
Conductor emerged while building Chorus and internal dev tooling. Charlie’s rapid prototype impressed Jackson, and early experiences—like assigning multiple tasks and seeing agents finish independently—made the workflow feel immediately powerful.
Distribution playbook: authentic demos and learning virality through reps
Charlie explains how he learned to create viral demo content, citing a GPT-4 Vision / David Attenborough narration project and his growth-focused role at Replicate. Their approach emphasizes authenticity, direct language, and lots of experimentation to learn what resonates.
What top engineers do with agents: skills files, simple setups, and “slop-free zones”
From user sessions, they notice elite engineers invest in reusable “skills” documentation that agents can read, while keeping tooling surprisingly simple. They also adopt the idea of “slop-free zones,” where humans maintain tighter architectural control in critical areas.
The future of engineering: from programmer to CEO of thousands of AI workers
They predict models will become dramatically more capable, running longer and behaving more like coworkers. This shifts the job toward high-level delegation, periodic deep dives, and new interfaces for oversight—similar to managing an organization and reviewing PRs.
Roadmap and hiring: Conductor everywhere, higher abstractions, and rethinking review
They outline a vision where Conductor manages agents across an organization and can be accessed from any device, not just a Mac. They’re excited about new abstractions beyond today’s interface and about modernizing review workflows, and they share the traits they want in new hires.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome