Having spent the better part of the afternoon stress-testing the new team management dashboard released in the latest Claude Code workspace update, I have to say my initial impressions are... mixed. The shift towards a more granular, role-based permission system is a significant step forward for any team serious about deploying Claude Code in a production B2B or software development environment. However, the implementation feels like a version 1.0—powerful in theory but lacking some critical polish and depth in its current form.
Let's break down the core components:
**The Good (Feature Matrix Highlights):**
* **Role Granularity:** The predefined roles (Admin, Developer, Analyst, Viewer) are a sensible starting point. The ability to define custom roles is a non-negotiable for enterprise compliance, and seeing it included is a major plus.
* **Project-Level Isolation:** Assigning team members to specific projects with distinct permissions is foundational for client work or internal product teams. This works as advertised.
* **Audit Logging Foundation:** The activity feed at the workspace level is present, capturing basic user actions. This is a crucial first step for traceability.
**The Concerning (Where It Falls Short):**
* **API Key Management:** This is my primary pain point. The dashboard allows for API key creation and basic role assignment, but the management lifecycle is brittle.
* There's no obvious way to filter or search keys.
* Key rotation procedures are manual and lack any form of staged rollout or deprecation warnings.
* Crucially, there is **no integration with the usage analytics**. I cannot see which API key is being used for what volume of requests, which is essential for cost attribution and security monitoring in a financial or supply chain software context.
* **Workflow Automation Gaps:** The roles govern access to Claude Code features, but they do not yet integrate with workflow-level permissions. For example, I cannot set a rule that "Analysts can only run pre-approved workflow templates" or "Viewers cannot export data." This limits its utility for complex, multi-stage approval processes common in ERP-adjacent automation.
* **B2B Integration Blind Spot:** There is no functionality for service accounts or machine-user management tailored for system-to-system integrations. When connecting Claude Code to an inventory management API, I need a dedicated identity with specific, narrow permissions—not a human user seat.
The dashboard feels built for a small, collaborative team of developers. For those of us looking at Claude Code as a potential engine for internal tooling or client-facing supply chain automation, the current feature set lacks the administrative control and deep observability required. The role-based access control (RBAC) is a solid skeleton, but it needs muscle in the form of detailed audit trails, financial attribution (tying usage to keys/projects), and service account support.
My open question to the community and the team: has anyone devised effective workarounds for the API key attribution issue? Are we expected to create a separate project for each integration point to isolate costs and usage, or is there a pattern I'm missing? I'm currently building a usage metering layer externally, which feels like an unnecessary duplication of effort.
Data over opinions