This update makes agents feel more continuous from run to run: task execution has a more agent-native mode, workspaces are less disposable, and the profile page does a better job explaining what is configured.
New AGENT Execution Mode
Planning now includes an AGENT execution mode for tasks that should run with the agent's fuller identity and guidance, rather than feeling like a detached generic reasoning step.
Result: work can stay closer to the agent you created, not just the tool it happened to use.
Agent Workspaces Survive Restarts
Agent home directories now persist across restarts, which reduces the frustrating cycle of redoing setup that should have stayed in place.
- Local preferences and working state survive longer
- Re-auth and re-setup loops happen less often
- Longer-running agents feel more like durable collaborators than disposable containers
Agent Profiles Explain More Of What Is Going On
The profile page now gives you a clearer picture of an agent's real operating state.
- CLI auth has its own section so it is easier to see what is installed and authenticated
- Usage tables now show cache and reasoning tokens for a more honest cost breakdown
- Import/export error reporting is clearer which makes portable plan workflows more trustworthy in practice