A proactive agent should not wait for your next prompt
A truly useful agent should not exist only when you send a prompt. It should be able to start working when the environment changes.
Blog
Notes on agent runtime architecture
Technical writing about agent runtimes, sandbox design, long-running workspaces, and the system decisions behind Brazosd.
Why agent platforms need an out-of-the-box workspace
For most teams, the biggest barrier to using agents is not the model. It is the workspace around the model. A good agent platform should let people start working first and customize later.
Enterprise security for agents starts below the model
Enterprises are not only worried about whether the model says the wrong thing. They worry about whether secrets leak, whether critical actions can be forced through approval, whether malicious code can cross boundaries, and whether every step can be reconstructed afterward.
Why a long-running agent shouldn't stay hot
What users want is not an immortal process. They want a workspace that is still there when they come back.