Eivind Kjosbakken

Practitioner and Towards Data Science author writing hands-on accounts of scaling Claude Code. Two pieces map its two scaling axes: orchestrate-100-agents-claude-code (2026) on breadth — 100+ agents in parallel via headless mode — and run-claude-code-agents-24-hours (2026) on duration — one agent unattended for 24h. Both land on the same constraint: autonomy is gated by the verification signal and containment, not raw model ability, so human review is the bottleneck (loop-engineering).

Related

orchestrate-100-agents-claude-code · run-claude-code-agents-24-hours · agent-orchestration · loop-engineering · agent-guardrails