[FEATURE] Agent View as an alternative to Project View

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Projects have been a game changer for my workflow. They have made my coding and vibe-coding experience feel roughly 5× more productive. However, some work is better organized around an ongoing responsibility than a specific project.

Accounting, recurring operational tasks, and security reviews across multiple repositories continue over time. They need a persistent owner with a clear role, retained context, and visibility into outstanding work. Organizing these responsibilities as individual projects or sessions can feel unnatural.

This is particularly relevant for nontechnical users, who often think in terms of people, teams, and responsibilities. Concepts such as sessions, project workspaces, and agent configurations can be abstract. “My accounting assistant” or “my security reviewer” provides a more familiar way to understand and interact with the product.

Technical users face a similar challenge when responsibilities span multiple projects. If I need ongoing security reviews across five repositories, a dedicated security agent would be a more natural fit than creating another project to coordinate that work.

Local agent configurations also do not fully address the experience I am looking for: the same persistent agent, associated with my account, available from my phone, laptop, or desktop. Its identity, context, and ongoing work should remain accessible when I switch devices.

Proposed Solution

Introduce persistent, account-level agents alongside Projects, with an Agent-based View and a Project-based View.

This could appear as a toggle between Projects and Agents, or as two choices when creating a new workspace:

• Project: Organize work around a specific codebase or deliverable.
• Agent: Organize work around a role and ongoing responsibilities.

Agents would appear in a dedicated category. Each would have a name, a defined role, persistent instructions and context, access to relevant projects or resources, and a clear view of ongoing tasks, progress, and items requiring attention.

The agent would feel like a recognizable team member whose work continues across conversations. Tasks could be initiated by the user, a schedule, or a configured event; persistence would not require continuous execution.

These agents should be tied to the user’s account, with flexible execution options:

• Cloud execution: Carry out configured tasks without requiring a particular personal computer to remain online.
• Local execution: Use an authorized local machine when tasks require local files, tools, or compute.
• Cross-device continuity: Access the same agents, conversation history, and task status from a phone, laptop, or desktop.

The agent’s identity and continuity would belong to the account, while its execution environment would be a separate choice, subject to the permissions and resources available.

For example:

• An accounting agent could maintain context across monthly routines, track outstanding tasks, and let me start a conversation on my laptop and continue it from my phone.
• A security agent could review pull requests across five repositories, track findings, and provide a single view of its ongoing responsibilities.

A project organizes work around an outcome; an agent organizes work around a responsibility. Supporting both would make Claude Code more useful for ongoing work and more approachable for people who naturally think in terms of teams and roles.

Alternative Solutions

_No response_

Priority

Critical - Blocking my work

Feature Category

CLI commands and flags

Use Case Example

_No response_

Additional Context

_No response_