Conductor

Conductor is an AI-native development startup and career inbound lead from People Network Ledger. It is relevant because its product thesis overlaps Kevin's multi-agent coding and Agent Machines orchestration interests.

Startup at conductor.build - recruiting Kevin for an engineering role (LinkedIn inbound Jun 2026). The public careers page describes Conductor as a San Francisco company building the future of AI-native software development: a Mac app for running a team of coding agents, positioned around the emerging "AI orchestrator" category. Source: Conductor Join Us page, 2026-06-18

Public positioning

Conductor's current public story:

Dimension Detail
Product Run parallel Claude Code, Codex, Cursor, and OpenCode agents in isolated workspaces on a Mac
Category AI orchestrator / primary UI for AI managers
Location San Francisco, in-person about five days a week
Team page roles Charlie Holtz, CEO; Jackson de Campos, CTO; Sam Lehman, Ops
Public traction claim 10x growth since January; $24M raised; used by engineers at companies including Linear, Vercel, Ramp, Notion, and Stripe
Current product docs Each task gets its own workspace, branch, files, terminal, diff, and review path
Current release signal 0.70.0 on June 26, 2026 added multiple run scripts and repo script configuration

Source: Conductor Join Us page, 2026-06-18; https://www.conductor.build/; https://www.conductor.build/docs; https://www.conductor.build/changelog, 2026-06-27

Product stack and setup surface

Conductor's stack, as exposed through public docs, is a Mac desktop orchestrator around local git worktrees. The core unit is a workspace: a branch plus a worktree plus agent chat, terminal, run environment, diff, checks, and PR/review state. It treats repo setup as product surface: .conductor/settings.toml and .conductor/settings.local.toml define setup/run/archive scripts, environment variables, files to copy, prompt customizations, and git behavior. Source: Conductor docs and official agent skill, 2026-06-27

The harness surface currently covers Claude Code, Codex, Cursor, and OpenCode. Claude Code, Codex, and OpenCode are bundled or managed by Conductor; Cursor runs through the Cursor API and needs CURSOR_API_KEY. Conductor does not resell model usage: provider accounts, subscriptions, and API keys determine billing and available models. Source: https://www.conductor.build/llms-full.txt, 2026-06-27

Important implementation implications:

  • worktree compatibility is a first-class product constraint
  • repo setup is mostly TOML, scripts, env, copied gitignored files, and port allocation
  • CONDUCTOR_PORT gives each workspace a 10-port range
  • shared settings only apply after landing on the default remote branch
  • local-only settings belong in .conductor/settings.local.toml
  • MCP setup is per harness: Claude Code, Codex, and Cursor Composer each have their own config path

Why it matters to Kevin

Conductor is not only a recruiting lead. It is a neighboring product thesis. Conductor focuses on the human UI for managing parallel coding agents on a local Mac; Agent Machines focuses on persistent remote workers, substrate/runtime routing, and browser/CLI operation. The overlap is useful for comparison:

  • both assume agents become teams, not single chat threads
  • both care about orchestration as the operator surface
  • both have to make agent state, parallelism, and handoff visible
  • both need strong taste because the interface becomes the engineer's cockpit

Kevin has a 20-minute intro call scheduled with People Network Ledger on Fri Jun 26, 2026, 2:00-2:20 PM PT (Google Meet). Career signal, not Agent Machines fundraising. Source: Gmail sync, 2026-06-06; User, 2026-06-12

Programming interview

See Conductor Programming Interview Prep for the focused prep page. The programming round is about a 40-minute open-ended CLI script problem where normal tools, IDEs, and AI agents are explicitly allowed. The strongest signal is technical ownership under agent acceleration: Kevin should restate the invariant, inspect SQLite before designing around imagined tables, use Codex for scoped scaffolding/review, and keep the interviewer in the loop while shipping a working vertical slice. Source: User-pasted interview instructions, 2026-06-27


Timeline