Multitasking Score — Delegation & Parallelism

The Multitasking Score is a composite metric (0–100) measuring how effectively you delegate work and run AI tasks in parallel. It captures whether you're using Claude Code as a single worker or as a coordinated team.

Definition

The score combines two signals — subagent delegation and session concurrency — normalised by the number of sessions.

Formula:

Multitasking = min(100, (subagent_spawns x 2 + peak_concurrency x 3) / session_count x 10)

Components

Subagent spawns: When Claude creates helper agents to handle subtasks within a session. This happens when Claude determines that a complex task can be broken into independent pieces and delegates them to sub-agents. Each SubagentStart event increments this count.

Peak concurrency: The maximum number of sessions running simultaneously at any point during the day. This measures your parallelism at the session level — how many independent Claude Code instances you had working at once.

Session count: The total number of sessions for the day. This serves as a normalising factor — more sessions provide more opportunities for concurrency and delegation, so the raw numbers are scaled accordingly.

Data source

The Multitasking Score is deterministic — every component comes from actual recorded events:

• subagent_spawns: Counted from SubagentStart hook events, stored in the subagent_spawns field on plugin_session_summaries
• Peak concurrency: Computed from timestamp overlaps between sessions (see Concurrency for details on the sweep-line algorithm)
• Session count: Total sessions recorded for the day

Interpretation

Score	Classification	What it means
0–20	Sequential single-task	Working one task at a time, no delegation. This is the default mode for most users starting out.
20–50	Moderate delegation	Some parallelism or subagent usage. You're beginning to leverage Claude's ability to work on multiple fronts.
50–80	Heavy parallelism	Significant concurrent work and delegation. You're treating Claude as a team, not a single assistant.
80–100	Maximum parallelism	Extensive delegation and concurrency. You're running multiple AI workers and using sub-agents for complex task decomposition.

Why it matters

The Multitasking Score reveals whether you're using AI assistance at its full potential. A single Claude Code session working on one task is valuable, but it's only one worker. Running parallel sessions and using subagent delegation turns that single worker into a coordinated team.

Subagent delegation is particularly powerful because it happens automatically within a session. When Claude encounters a task that can be decomposed — like "update all test files to use the new API" — it can spawn sub-agents to handle each file independently. This is parallel execution within a single session, and it dramatically speeds up complex operations.

Session concurrency captures your own parallelism. Running a refactoring session alongside a test-writing session alongside a documentation session means three independent workstreams progressing simultaneously.

Together, these signals measure your total parallel capacity — both the parallelism you initiate (multiple sessions) and the parallelism Claude initiates (sub-agents).

How to increase your score

1. Use multiple terminal windows: Start separate Claude Code sessions for independent tasks
2. Break large tasks into parallel streams: Instead of one session doing everything sequentially, split work across sessions
3. Prompt for delegation: When giving complex tasks, encourage Claude to use sub-agents for independent subtasks
4. Identify independent work: Tasks that don't share files or state are good candidates for parallel sessions

The StarCraft analogy

The Multitasking Score maps to army splitting and macro management — the ability to coordinate multiple actions simultaneously across the map. A player who can only control one army group will always be outmanoeuvred by one who splits forces, manages multiple bases, and coordinates attacks on different fronts. In AI-assisted development, running parallel sessions and delegating to sub-agents is the equivalent of multi-front coordination: more work completed, more problems solved, more ground covered per hour.