2.3The environment

What this section will cover

• Two layers of environment: the one you work in, and the one each agent works in
• Your agent canvas: where you dispatch agents, follow their progress, and review their work
• Each agent's workspace, defined within that canvas: its own copy of the code, terminal, tools, running services, and credentials
• Kinds of agent workspace, from least to most isolated: your machine, a worktree, a container sandbox, a remote workspace
• Choosing a workspace for a task: what the agent needs, and what it could break
• Permissions and secrets in each workspace

Where the agent runs

• Local vs. cloud
• Local
  • Pros
    • Your agent can access your local environment, which you fully control
    • Nothing to set up
    • No billing-related usage limits
  • Cons
    • You eventually run out of local resources (CPU, memory)
    • Your agents sleep when your laptop closes
• Cloud
  • Pros
    • Scales to infinity
  • Cons
    • More work to set up your cloud environment

Cloud Agent options

• Claude cloud agents?
• Cursor cloud agents?