newsfilter.io
Product Demonstration, Interview

Conductor CEO Charlie Holtz Walks Us Through His AI Coding Setup

  • Predictions and Expectations:

    • Charlie expects the industry to quickly move to a world where agents run "10 times longer" and are "10 times smarter," necessitating environments unconstrained by local Mac CPU.
    • Charlie believes the world is "slowly waking up" to the conclusion that with next-generation models, old code won't matter as much as the ability to rerun prompts.
    • Charlie anticipates that software will evolve into "malleable" forms similar to video game mods, where users can "build in your own workflows."
    • Charlie predicts that as the technology advances, the concept of code will become more like "sawdust" in the development process, shifting focus entirely to describing desired outcomes.
  • Timelines and Milestones:

    • Charlie notes the highest token spend occurred "in July 2025."
    • Charlie mentions the "open in" button pattern decision was made "for a long time" before implementation.
    • Charlie indicates the current "workspace" abstraction in the sidebar is "going to change soon."
  • Technology and Product Direction:

    • Conductor plans to introduce "Cloud Workspaces" as a sneak peek of upcoming features.
    • The company aims to improve boundaries between "slot free zones" (human-written) and areas with "free reign for the AI."
    • Charlie is "always like pushing for more" Elixir usage in the codebase despite currently being 90-95% TypeScript.
    • The team intends to build the core of the app around "human-written APIs and contracts" to reduce AI contribution to critical infrastructure.
    • Conductor plans to allow users to "mod conductor" and build custom workflows, similar to video game mods.
  • Market and Industry Outlook:

    • Charlie believes the industry is moving away from terminal interfaces toward GUIs because "humans are spatial visual creatures."
    • Charlie expects the future workflow to feel like "conducting at the orchestra level" with multiple people working with AIs in "multiplayer chats."
    • Charlie suggests the world is beginning to understand that "prompts" will become the primary asset rather than the code itself.
  • Company Plans:

    • Conductor is working on a new "dashboard page" to provide a single view of all agents, with the ideal outcome of making the user feel like a "CEO of a little company."
    • The team is "messing around" with the interface of the new dashboard to determine what it should "look and feel like."
    • Conductor recently added a "garry mode" which defaults to not collapsing tool calls, specifically designed for a user named Gary.
    • The company plans to enforce a workflow where workspaces create "work trees" and "PRs" rather than allowing direct file edits.
  • Financial Guidance:

    • Charlie reports spending "$22,000 on tokens" in a single month (July 2025).
    • Charlie states they are not being measured on "lines of code" but rather on efficiency and minimal code additions.
  • Risks and Caveats:

    • Charlie warns that without careful boundaries, AI can enter a "vicious cycle" where it reads bad code and writes more bad code.
    • Charlie notes the current boundaries between human-written and AI-written code are "a little murky."
    • Charlie suggests that letting AI make UI choices can result in interfaces that "don't feel crafted."
    • Charlie admits that "most of them [experiments] don't make it in," indicating a high rate of discarded ideas.
  • Confidence and Disagreement:

    • Charlie is "pretty against" showing icons in the app bar at first but has since "really like[d]" the decision, noting it feels like "advertising a different app."
    • Charlie states it is "really important to us" that the software feels "crafted" and that humans must make the core architectural and design decisions.
    • Charlie expresses a strong belief that the future requires agents to run in environments "constrained by your Mac's CPU" is no longer sufficient.
    • Charlie is "not big on analytics" or A/B testing, relying instead on "gut feel" to make product decisions.