Most tools hand you one conversation, with one agent, in one repo. That's fine, right up until you're juggling three repositories, a background reviewer, and a feature that can't start until another one lands.
Frontier treats each piece of work as its own place — a real workspace, with its own worktree checked out from whichever repos it touches, and its own thread history, plus agents you can teach to run on a schedule or jump in the moment something changes. Attach the same repo to five different workspaces and you get five separate branches moving at once, each in its own worktree — nothing one is doing can collide with what another is building, even against the exact same codebase. When one agent hands work to another, it happens the way it would between two people: a message left in a thread, not some invisible shared memory.
This is the first in a series about the ideas behind Frontier — Domains, Strategies, and the Conductor, the pattern that lets a single workspace carry an entire roadmap forward on its own.