We present a Helical Pipeline Architecture (HPA) for orchestrating autonomous AI agent teams in compound software startups. Unlike flat sequential phase execution or single-ring assembly lines, HPA compounds four concurrent stage-groups across multiple ascending altitude layers, forming a helix through which software products spiral with increasing maturity. Key Contributions Formal model of helical pipelines with altitude semantics, promotion functions, and governance predicates Four altitude layers: Genesis (MVP), Integration (cross-product coherence), Production (deployment readiness), Evolution (data-driven iteration) Compounding concurrency: Up to 16 parallel workstreams at steady state across 4 altitudes x 4 stages Self-referential A3 layer: Products never terminate; they spiral through continuous evolution DAG scheduler specification targeting Claude Code subagent orchestration Three-tier compaction integration for agent, controller, and master context management Application Domain Applied to Pastoral Tech, a compound SaaS platform comprising 20+ interconnected products built atop a Unified Congregant Record, where theological guardrails, multi-tenancy, and cross-product data integrity impose non-trivial constraints on the pipeline. Document Integrity PDF SHA-256: 97addcb411d5f8403c23d7d62b23b493dfec937da66238d27212c9cebf3e0a05 TeX SHA-256: 41de0d14a5cf903bdaa7addefc45a2401a5ae0e19efa2009223cb820e7df006d

Architecture designed through iterative development of autonomous AI agent orchestration systems for compound SaaS platforms. Formalized using set-theoretic notation with invariant specification.

PDF SHA-256: 97addcb411d5f8403c23d7d62b23b493dfec937da66238d27212c9cebf3e0a05 | TeX SHA-256: 41de0d14a5cf903bdaa7addefc45a2401a5ae0e19efa2009223cb820e7df006d