About
Agent Engineering is a practitioner's reference for building AI agents — covering architecture patterns, context management, tool use, multi-agent orchestration, retrieval, safety, and evaluation. It is written for engineers who are building these systems in production, not for researchers reading about them in theory.
What's here
The site is organized around two content types. Topics are deep-dive reference articles on foundational patterns — ReAct, memory systems, prompt caching, the Skills Pattern, MCP, A2A, agentic RAG, and more. Each topic includes prose explanation, architecture diagrams, and Python code examples using LangChain/LangGraph or the Microsoft Agent Framework. Dispatches are shorter news items surfaced daily from the field — new research, framework releases, and protocol updates.
How it stays current
The news feed is maintained automatically by an AI agent. Every day, a GitHub Actions workflow
runs a Python script that fetches RSS feeds from Anthropic, LangChain, OpenAI, Google AI,
and Hugging Face. The raw items are passed to Claude, which filters for relevance to agent
engineering, writes a one-sentence summary for each, and returns a ranked JSON list.
That list is committed directly to the repository as src/data/news.json.
The commit triggers an automatic redeployment on Vercel. Within a few minutes of the workflow completing, the homepage and Resources page reflect the day's most relevant developments — with no manual editing involved.
How articles are added
New articles are generated using scripts/add-article.py, a CLI tool that accepts
a topic string, a URL, a pasted tweet, or raw notes. It calls Claude with web search enabled,
which researches the topic and writes a full newspaper-style MDX article — 800 to 1,500 words
of flowing prose, with architecture diagrams, code examples, and callouts. The output is saved
directly to the content collection and optionally committed in a single command.
The stack
The site is built with Astro and Tailwind CSS, deployed on Vercel. Content is stored as MDX files in an Astro content collection. The automation runs on GitHub Actions. All AI calls use the Anthropic API with Claude.
Contact
Built and maintained by Daniel Huber. Reach out at danielhuber.dev@proton.me.