Notes on agents, software, and work.
Systems / October 8, 2026 / 9 min read
RAG vs context engineering: find the system boundary before you fix it
RAG retrieves external knowledge; context engineering governs the full inference-time information environment, so the right fix follows the failure source.
Systems / October 7, 2026 / 11 min read
Python for agentic AI: inspect the controls before choosing the framework
Python for agentic AI starts with visible state, bounded loops, scoped tools, traces, approval points, and recovery controls.
Systems / October 7, 2026 / 11 min read
AWS agent harness: choose by responsibility, not feature count
An AWS agent harness choice comes down to who owns the loop, runtime, controls, and failure handling across AgentCore and custom paths.
Systems / October 7, 2026 / 10 min read
Google ADK vs LangGraph: choose by workflow truth and recovery responsibility
Google ADK favors defined tool capabilities and faster initial deployment; LangGraph favors explicit state, transitions, and auditable maintenance.
Systems / October 7, 2026 / 12 min read
Agentic workflows examples: seven reusable control shapes
Seven agentic workflow examples expose planning, routing, state, human gates, execution limits, and checkpoint-based resumption.
Systems / October 7, 2026 / 10 min read
Databricks managed MCP servers: choose the right control path
Databricks managed MCP servers split into managed, custom, and external paths; compare control, credentials, client limits, costs, and recovery.
Systems / October 7, 2026 / 10 min read
Agentic AI vs chatbot: choose by authority, not the label
Chatbots handle short interactions; agentic AI can pursue longer delegated tasks by orchestrating tool calls and interacting with environments.
Systems / October 7, 2026 / 10 min read
Meta context engineering via agentic skill evolution: a controlled-adoption guide
MCE co-evolves agent skills and context artifacts, and adoption also turns on evaluation, authority, tracing, monitoring, cost, and recovery.
Systems / September 30, 2026 / 11 min read
Agentic AI versus generative AI: output generation or delegated execution?
Generative AI creates outputs; agentic AI delegates multistep action and requires decisions about authority, evaluation, oversight, stopping, and recovery.