Agent architecture
Index of articles on architectures for multi-step, tool-using systems, including control-plane placement, memory boundaries, and orchestration patterns.
Engineering notes on architectures for multi-step, tool-using systems, including control-plane placement, memory boundaries, and orchestration patterns.
Core articles
Core articles
LLM Memory Boundary Model: Context Construction (Eligibility, Selection, Persistence) and Why Answers Change
A vendor-agnostic model of context construction—what can enter context (eligibility), what gets used per response (selection), and what is retained for later (persistence)—and the security controls that must live outside the prompt.
LLM-Led vs Orchestrator-Led Tool Execution Control-Plane Placement Tradeoffs
A control-plane placement comparison across reliability, observability, latency, cost governance, and security for tool-using LLM systems.
Human vs GenAI capability map (engineering view)
A practical mapping of human cognitive capabilities to GenAI limitations, engineering substitutes, and residual gaps.
Section resources
Context, reusable contracts, related links, and external baselines for this topic.
About this section About this section
Focus
- Orchestration patterns (control-flow mechanisms) — where control flow lives and how workflows are sequenced.
- State & lifecycle management — what persists across steps and boundaries (session/thread/run), what resets, and when.
- Tool invocation lifecycle — selection, authorization/enforcement, validation, error handling, retries, and egress constraints.
- Write paths — where and how the system can persist changes to external systems.
Terminology
- Orchestrator: the component that owns control flow (decides next action/tool call) and enforces policy.
- Model-led: the model proposes next actions/tool calls within constraints enforced by the orchestrator.