Notes on agents, software, and work.
Systems / September 16, 2026 / 10 min read
Human in the loop LangGraph example: what recoverable execution requires
A recoverable LangGraph human-in-the-loop design needs risk-scoped interrupts, durable checkpoints, audit logs, timeouts, and idempotent side effects.
Systems / September 15, 2026 / 9 min read
LangGraph autonomous agents: design the operating envelope
LangGraph autonomous agents need explicit boundaries for tools, state, stopping, review, validation, and recovery.
Systems / September 14, 2026 / 9 min read
Model Context Protocol vs RAG: choose the right system boundary
MCP defines how models discover tools and data sources; RAG handles knowledge retrieval. Use either one or combine them around explicit system boundaries.
Systems / September 13, 2026 / 9 min read
OpenClaw multi-agent orchestration readiness guide
OpenClaw can run isolated agents, and six checks expose whether workload, state, routing, permissions, completion, and failure handling are ready.
Systems / September 12, 2026 / 10 min read
VS Code multi-agent orchestration: a control-first decision guide
A practical VS Code multi-agent orchestration plan covers task boundaries, shared-state ownership, approval gates, and safe retries.
Systems / September 11, 2026 / 10 min read
Human in the loop oversight: test the whole control path
Human-in-the-loop oversight is an action-level control path with risk gates, saved state, decision logs, and recovery controls.
Systems / September 10, 2026 / 10 min read
Why is context engineering important in multi agent systems
Context engineering matters in multi-agent systems because it preserves task alignment, limits authority, traces handoffs, and enables recovery.
Systems / September 9, 2026 / 10 min read
AI guardrails open source: choose the control boundary first
Open-source AI guardrails differ by enforcement point, policy ownership, error cost, observability, authorization, approval, and recovery.
Systems / September 8, 2026 / 11 min read
Open source AI agents for coding: choose by the stack you will own
Six open-source coding agents compared by interface, license, hosting, and model path, plus practical checks for free access and authority.