Concurrency — Parallel Session Count

Concurrency measures how many Claude Code sessions you're running simultaneously. It captures your ability to multiply your AI workforce by running independent tasks in parallel.

Definition

At any given moment, concurrency is the number of sessions that have started but not yet ended. The Control Center reports both peak concurrency (the highest overlap at any instant) and average concurrency (the time-weighted mean over the day).

How it's measured

Concurrency uses a sweep-line algorithm over session timestamps:

1. Collect all started_at and ended_at timestamps from plugin_session_summaries
2. Create a timeline of events: each session start adds +1, each session end adds -1
3. Walk through the timeline chronologically, maintaining a running count
4. The maximum value reached is peak concurrency
5. The time-weighted average across the day is average concurrency

This approach gives an exact count at every moment — there is no sampling or polling.

Peak vs average

• Peak concurrency: The highest number of sessions running at the same instant. This tells you the maximum parallel workload you achieved.
• Average concurrency: The time-weighted mean across the day. A peak of 5 with an average of 1.2 means you briefly ran 5 sessions but mostly worked with 1-2.

Peak is useful for understanding your maximum capacity. Average is useful for understanding your typical working pattern.

Data source

Concurrency is deterministic — computed entirely from actual timestamps:

• started_at: Recorded when a SessionStart hook event fires
• ended_at: Recorded when a SessionEnd or SessionStop hook event fires
• Both timestamps are stored on plugin_session_summaries

No estimation is involved. If a session was active during a time window, it is counted.

Why it matters

Running parallel sessions multiplies your AI workforce. A concurrency of 4 means four AI assistants are working on different tasks simultaneously — four sets of file reads, edits, test runs, and code generation happening at once.

Practical scenarios for high concurrency:

• Feature development + test writing: One session builds the feature while another writes tests
• Multi-repository changes: Parallel sessions working across different repositories
• Research + implementation: One session explores documentation while another implements the solution
• Code review + bug fixing: Review code in one session, fix issues in another

Even modest concurrency (2-3) can dramatically increase daily output. The key constraint is usually your ability to context-switch between sessions and provide direction, not the AI's capacity.

Interpretation

Concurrency	Pattern
1	Sequential — one task at a time
2-3	Light parallelism — common for experienced users
4-6	Heavy parallelism — managing multiple workstreams
7+	Team-scale — typically indicates multiple team members or automated pipelines

The StarCraft analogy

Concurrency maps directly to multi-front battles — managing multiple bases, armies, and expansions simultaneously. A player who can only control one army group at a time will always lose to one who coordinates attacks across the map. In AI-assisted development, running parallel sessions is the equivalent of multi-front operations: more ground covered, more tasks completed, more problems solved per hour.