Charlie Holtz

Charlie Holtz is the CEO and co-founder of Conductor, an app for managing teams of AI coding agents through one interface. His work centers on how people direct those teams: assigning parallel tasks, following their progress, reviewing changes, and collaborating with other developers.

From quantitative finance to hands-on AI

Holtz studied cognitive neuroscience at Brown, where he met his future co-founder, Jackson de Campos, playing ultimate frisbee. He worked in quantitative finance at Point72 before moving into AI application development. At Replicate, he was a Hacker in Residence and led growth, building applications that gave people concrete ways to try emerging models.

His independent projects show a playful approach to learning through software. In March 2020, he started ShlinkedIn, an open-source social network satirizing corporate language, partly to teach himself Elixir. Later, CHARL-E packaged Stable Diffusion into a Mac application for local image generation without a complicated installation. Both made their subject approachable through an interface people could use themselves.

Holtz also built experiments connecting different kinds of models. His AI narrator turned webcam images into nature-documentary-style commentary delivered through a synthetic David Attenborough voice. It combined visual understanding, generated narration, and speech synthesis; he explored both separate stages and a vision model that could interpret the scene and write the narration in one request. The joke made the application immediately understandable while demonstrating what a developer could assemble from existing models.

With StickerBaker, he built an open-source application around fofr’s sticker-generation model. Its asynchronous workflow let image generation continue even if the browser disconnected, then stored the finished sticker and returned it to the interface. His Elixir prompting tools addressed a smaller frustration: a custom ~LLM sigil converted readable conversation text into API messages, reducing repetitive formatting when using language models inside an application.

Building the interface for an AI team

Holtz and de Campos started their company in summer 2024 and joined Y Combinator’s Summer 2024 batch. Their early product, Melty, was an open-source editor built from VS Code. They wanted an AI collaborator that understood activity across the editor, terminal, and GitHub, so developers would not have to shuttle code between disconnected chats.

Conductor emerged from the team’s own development workflow while they were building Chorus. Running several coding agents meant juggling separate repository copies. Discovering Git worktrees gave them a better foundation for an internal tool, which grew into a product for launching agents in isolated workspaces, inspecting their progress, and reviewing and merging their changes.

In March 2026, Holtz and de Campos announced Conductor’s $22 million Series A, led by Spark and Matrix. Their ambition extended from small groups of agents on a MacBook to an interface for much larger AI organizations. Conductor’s cloud workspaces provide isolated environments with repositories and dependencies prepared, while shared sessions let teammates follow work and prompt agents together. Moving execution off the laptop also allows agents to keep working after it closes and makes it possible to initiate work remotely through an agent with Conductor API access. These capabilities extend the product’s focus from individual productivity to collaboration among people and agents.

Where human judgment matters

Holtz’s approach to agent-assisted development gives people substantial responsibility for the quality and direction of the work:

• Selective workflow optimization: He favors trying new tools promptly, while reserving substantial customization for needs specific to a product or codebase. His reasoning is that broadly useful workflows are likely to become defaults in the major coding tools. Rendering long conversations quickly is a concrete reason for his team to invest in specialized optimization; endlessly polishing a general-purpose setup can consume the time it was meant to save.

• Slop-free zones: Conductor requires strict human review of database migration changes and careful maintenance of documentation and agent instructions. Holtz connects these practices to app rewrites the team undertook after giving consequential parts of the codebase insufficient care. Instruction files deserve particular attention because their contents enter an agent’s context each time it starts work.

• Organizational context for agents: The Conductor Internal Agent gathers Slack messages, Discord bug reports, and meeting recordings into Postgres. Giving coding agents access to that queryable knowledge helps them work with information about users and company decisions that the repository alone does not contain.

Holtz’s preferred image is a conductor leading an orchestra. People should be able to direct collaborating humans and agents, step into the details when needed, and retain a sense of craft. That position shapes his interest in shared workspaces: more capable agents should let teams attempt more ambitious work while keeping people involved in deciding what is worth making.