Claude Code has a new UI (pair it with Claude OS)

Lines worth screenshotting.

• Claude Code's native Agent View terminal dashboard is Anthropic's answer to the proliferation of third-party multi-agent UIs — it ships the management layer that the community had been building as workarounds.
• Sorting agents by repo rather than status is the view that makes multi-project work legible: you can see all tasks in a specific codebase or client folder grouped together without switching terminals.
• Backgrounding existing sessions with /bg and migrating them into Agent View converts fragmented terminal windows into a unified dashboard without interrupting the work in progress.
• Pinning specific agents with ctrl+t and reordering with shift+up/down means the dashboard reflects the human's current priorities rather than chronological session order.
• An Agentic Operating System as a folder structure that injects context at the right time is compatible with Agent View because the OS is not a UI — it is the data layer that each session reads, regardless of how sessions are managed.
• The client subfolder limitation — sorting by repo but not by subfolder — is the gap that matters for agencies operating multiple client contexts: until subfolder-level sorting ships, a workaround is required to separate client work visually.
• Agents getting to 90% of the output from a good prompt is the capability threshold that makes multi-agent management the bottleneck — the problem is no longer 'can the AI do this' but 'how do I manage multiple agents doing multiple things simultaneously.'
• Talking to a session directly from the summary view — without jumping into the full terminal — is the productivity behavior that Agent View enables: brief responses, approvals, and status checks without losing the dashboard context.
• The pattern of building third-party dashboards (Vibe Kanban, TMux, custom command centers) before Anthropic ships a native solution is the standard adoption curve for developer tools — the community solves the pain first, then the platform formalizes it.
• Terminal-only availability at launch means desktop app users cannot use Agent View yet — which is the constraint that determines whether an experienced terminal user will switch from their existing workflow.
• A Agentic OS folder structure that injects brand voice, client context, and scheduled jobs is the abstraction layer that makes every session in Agent View context-aware without manual setup per session.
• Managing multiple agents from a single dashboard without five terminal windows is not a convenience improvement — it is the operational requirement for any solo operator running concurrent work streams across multiple clients or projects.
• Seeing the last output and time-since-last-input for each session in the summary view enables triage: you know immediately which agents need attention and which ones are running without checking each one individually.
• Approving agent actions from the summary view — type approve without jumping to the full terminal — is the workflow that keeps the human in the loop without the context switch cost of switching terminals for each approval.
• The multi-agent management problem is the next frontier after the single-agent quality problem: once individual agents reliably execute complex tasks, the constraint moves to how many can a single human effectively supervise simultaneously.