Articles for people who
ship the stack
Original rewrites on React, Node.js, TypeScript and AI — practical notes from the same engineering practice behind our operator software. Article bodies are in English.
Tagged: LLM
A Production Harness for Claude Code: Six Layers From Slack to Merged MR
How a six-layer harness turns Slack tickets into merged code: event intake, gated agent loops, durable state, self-healing, observability and operator control.
42 min readFifty Years of Retrieval Research Hiding Inside Your RAG Pipeline
RAG chunking, hybrid search, HyDE, reranking and evals all have decades-old ancestors in information retrieval. A tour of the research and where it falls short.
17 min readCutting Agent Token Costs: Caching, Routing and Context Hygiene
Shrink LLM agent bills with prompt caching and semantic caching, lazy-loaded tools, model routing and cascades, subagents, and plain old context hygiene.
16 min readAI Agent Foundations: Purpose, Prompt Design and Model Selection
How to lay the groundwork for a production AI agent: define purpose and success metrics, write prompts with goals and guardrails, then pick the right model.
21 min readContext Engineering in Practice: A Claude Agent SDK Research Agent
Build a Python research agent with the Claude Agent SDK using write, select, compress and isolate patterns, then compare it against a naive baseline in measured runs.
20 min readWhat a Model-Diffing Agent Finds: Style Shifts Overshadow a Narrow Fine-Tune
A model-diffing agent audited a finance LoRA fine-tune: 82% of runs caught the length and structure shift, while the intended reasoning change stayed hidden.
15 min readAgent Harness, Loop and Graph: Three Layers Behind Reliable AI Agents
Why unreliable agents are usually missing a layer: the harness around the model, an evidence-driven loop and an explicit graph, with code patterns and a debugging checklist.
15 min readThree Ways Production RAG Fails: Fragmented Context, Drift and Misuse
Three ways production RAG fails: fragmented context from chunking, drift as the knowledge base grows, and misuse despite good retrieval, with layer-level fixes.
11 min readRebuilding a Claude Code-Style Agent With LangChain Deep Agents
How a coding agent's harness works, from the core loop to tools, planning, memory, subagents and sandboxed approvals, rebuilt step by step with deepagents.
16 min read
About these articles
Request a 24h estimate
Need the same stack in a production operator layer? Send the brief — estimate within 24 hours.