---
title: Agentic coding craft
tags:
  - wiki
  - concepts
sources:
  - "[[Here's how I use LLMs to help me write code]]"
  - "[[How I Use Claude Code]]"
  - "[[Cursor agent best practices]]"
  - "[[Beyond the Prompt Claude Code]]"
  - "[[Zen of AI Coding]]"
  - "[[Writing about Agentic Engineering Patterns]]"
  - "[[Shipping at Inference-Speed]]"
  - "[[Night Shift Agentic Workflow]]"
  - "[[AI-Augmented Software Development Manifesto]]"
date: "2026-07-17T06:24:14.541Z"
created: "2026-07-17T06:24:14.541Z"
includeInRss: false
---

# Agentic coding craft

How humans + coding agents ship software without drowning in context or vibe-slop.

## Recurring patterns

- **Context is the craft** — what you put in AGENTS/skills/files beats clever prompts.
- **Plan → execute → verify** — explore/plan modes; TDD or explicit tests as the agent’s ground truth.
- **Code is cheap; judgment is scarce** — bottlenecks move to review, product sense, and shipping.
- **Inference-speed shipping** — trust streams, CLI-first tools, engineer *for* agents.
- **Night Shift** — day: specs & review personas; night: autonomous loops.
- **Agentic engineering ≠ vibe coding** — structured loops, not autocomplete theater.

## Tension

Speed vs accountability. Manifestos push “no dedicated QA / engineer owns outcomes” — useful as pressure, dangerous as dogma.