Writing about Agentic Engineering Patterns

This post announces Agentic Engineering Patterns, a project to collect and document coding practices for getting good results out of coding agents. Its stated goal is to answer “how do I get good results out of this stuff” in one place, against a body of the author’s own existing writing on the subject that he characterises as relatively unstructured.

Before describing the project the post fixes its vocabulary. Agentic Engineering is used for building software with coding agents, with the defining feature named as the ability to both generate and execute code, which lets the agent test and iterate independently of turn-by-turn human guidance. Vibe Coding is kept in what the post calls its original definition — coding where you pay no attention to the code at all — and the two are placed at opposite ends of a scale, with agentic engineering described as professional engineers amplifying existing expertise.

The remainder is about form rather than content. The post introduces a publishing format the author calls a guide: a collection of chapters, where a chapter is a blog post with a less prominent date, designed to be updated over time rather than frozen at first publication — his answer to the problem of publishing evergreen material on a blog. It also states a personal policy of not publishing AI-generated writing under his own name, which he says will hold for this project.

Key Points

• Agentic Engineering Patterns is a new project collecting coding practices and patterns for getting the best results out of coding agents.
• “Agentic engineering” is used to mean building software using coding agents — tools whose defining feature is that they can both generate and execute code.
• That execution capability is what allows an agent to test its own code and iterate on it independently of turn-by-turn guidance from a human supervisor.
• “Vibe coding” is used in its original definition: coding where you pay no attention to the code at all, today often associated with non-programmers using LLMs to write code.
• Agentic engineering is placed at the opposite end of that scale — professional software engineers using coding agents to amplify existing expertise.
• The project is structured as a series of chapter-shaped patterns, which the author describes as loosely inspired by the format popularized by the software design-patterns literature.
• The author’s existing writing under his ai-assisted-programming tag is described as extensive but unstructured, which is the gap the project is meant to fill.
• Two chapters were published alongside the announcement: one on writing code having become cheap, and one on red/green TDD helping agents write more reliable code with minimal extra prompting.
• The intended cadence is one to two chapters a week, with no planned stopping point.
• The author states a strong personal policy of not publishing AI-generated writing under his own name, and says it holds for this project; LLMs are used for proofreading, example code and side tasks, but not for the prose.
• A “guide” is introduced as a new content shape on the author’s site: a collection of chapters, each a blog post with a less prominent date, designed to be updated over time rather than frozen.
• Guides and chapters are presented as the author’s answer to publishing evergreen content on a blog, a problem he reports having been trying to solve for a while.
• The author reports that almost all of the implementation behind the new format was written by Claude Opus 4.6 running in Claude Code for web, accessed from his phone — his own account of how the code was produced.

Context

The post is an announcement and a terminological statement rather than an argument, and its definitions are offered as the author’s own usage for the purposes of this project rather than as consensus. It is also a partial demonstration of its own subject: the author reports having built the site machinery for the new format with a coding agent while writing the prose himself, which lines up with the split his definitions draw. The project’s framing — that the field is moving fast enough to need a structured place to put what works — is stated as motivation rather than defended.