architecture-designlisted

# Architecture Design for Agentic Products The architectural decisions made in the first hour of a project determine whether it ships. This skill walks the user through them deliberately. ## Step 1: Determine the autonomy level (Autonomy Ladder) Never start with "build an agent." Start with "what is the minimum autonomy this task requires?" The cost of getting this wrong is asymmetric: too much autonomy = unreliable, expensive, slow, hard to debug. Too little = doesn't capture value. | Level | What it is | Cost / latency | Use when | |---|---|---|---| | **L0. Single LLM call** | One prompt → one response | Lowest | Classification, extraction, summarization, generation with known structure | | **L1. Augmented LLM** | LLM + retrieval / tools / memory, but single-shot | Low | Q&A over docs, simple structured tasks, lookup + reformat | | **L2. Workflow** | Deterministic code orchestrates LLM steps; path is predefined | Low–medium | The execution path is knowable in advance; predictability matters | | **L3. Orchestrator-Worker** | LLM plans dynamically, dispatches to bounded sub-agents | Medium–high | Parallelizable subtasks (research, multi-source synthesis); breadth-first work | | **L4. Autonomous Agent Loop** | LLM chooses next step iteratively until termination | Highest | Path cannot be enumerated; emergent behavior is the value; cost compounding is acceptable | ### Escalation rule (non-negotiable) Do not climb to level N+1 until level N delivers **≥90% pass rate on a c