What belongs here
- Tool use
- Planning and task decomposition
- Reflection and repair loops
- Short-term and long-term memory patterns
- Subagents and delegation
- Browser and computer-use patterns
- Human-in-the-loop workflows
- Long-running task design
- Deep research patterns
Editorial intent
Pages in this lane should explain mechanism, tradeoffs, and failure modes. They should stay useful even as individual frameworks rise or fall.Current pages
- Agent Memory And Retrieval: how agents separate active state, durable memory, retrieval, and explicit artifacts.
- Reasoning And Control Patterns: how think-act-observe loops shape control, explainability, and tool use.
- Planning And Reflection: how plan-first and critique-and-refine patterns improve quality on longer tasks.
- Agent Runtime Building Blocks: how modern runtimes combine sandboxes, tool and protocol boundaries, control planes, memory, and adaptive endpoint surfaces.
Example starters
- Agent Memory Retrieval Starter: a small code sketch for separating active notes, verifiable RAG retrieval inputs, citations, and durable artifacts.
- Prompt Cache Agent Starter: a small code sketch for keeping stable prompt prefixes separate from dynamic memory and current-turn inputs.