Skip to content
Chapter 05 · 8 min

Manage with Agent View, and pick the right mode

Once a team is running you need a way to watch it, and a rule for when to even start one. Agent View handles the first; a short decision framework handles the second.

“A control room with one screen showing every worker's progress, plus the discipline to know when a job doesn't need a crew at all.”

Agent View: one screen for every session

Open the dashboard with a single command. From there you dispatch new tasks, peek at any agent's progress without interrupting it, and attach to one when it needs your input. Sessions run under a supervisor, so you can close the laptop and they keep working.

claude agents
+-------------------------------------------------+
| Agent View                                      |
+----------+----------+-------------+--------------+
| Backend  | Frontend | Testing     | Review       |
| ######   | ####     | queued      | waiting deps |
| 72%      | 48%      |             |              |
+----------+----------+-------------+--------------+
| > dispatch new task                             |
+-------------------------------------------------+
Every session on one screen: progress, queue state, and a prompt to dispatch more work.

Pick the right mode

Orchestration isn't free. Using a team on a one-line fix wastes tokens; running dependent tasks as isolated sessions wastes coordination. Match the mode to the work:

  • Single prompt, single-file fix → a regular Claude Code session. No agents.
  • 3 independent tasks, no dependencies → Agent View. Dispatch all three, collect results.
  • Repeatable workflow (review, test, docs) → subagents with a YAML config, consistent every time.
  • Multi-file feature with dependencies → Agent Teams. The lead coordinates, teammates collaborate.
  • Overnight backlog drain → headless mode, left to run.

In one line each

  • `claude agents` opens Agent View: dispatch, peek, attach; sessions survive terminal closure.
  • Independent tasks don't need Agent Teams coordination; dependent tasks shouldn't run as isolated sessions.
  • Map the work to the mode: single session, subagents, Agent View, Agent Teams, or headless.

Where to go next

Manage with Agent View, and pick the right mode · AI courses · SDEN