Linear is designed to orchestrate teams of humans and agents working together. It’s the most efficient way to track progress and scale teams of agents to tackle large, complex tasks. Codegen can take a first pass at virtually any issue, breaking down work and making meaningful progress before human review. We recommend letting Codegen handle the initial exploration and implementation of most tasks.
Once you’ve enabled linear self-assign in the settings page, a codegen agent, that has been assigned to a linear issue (or has been tagged in one), can spawn child agents
by creating sub-issues and assigning itself to those sub-issues. For each sub-issue that codegen assigns to itself a child agent will be spawned and tasked with completing the sub-issue. Once the child agents are
finished with their tasks they will notify their parent by sending it a message. The parent will then incorporate the child’s work into its own as appropriate.
If you’d like to have codegen break up a linear issue into smaller issues and assign them to child agents you should instruct it to do so in the
description of the original linear issue.
Before creating sub-issues and assigning them to child agents the parent agent will produce scaffolding in the form of a git branch and include details
of this branch in the description of the sub-issues. The child agents will then work off of this scaffolding branch. If you have specific scaffolding requirements
or context you’d like the child agents to share, please include them in the description of the parent issue.