Agentic Engineering with Claude
Ongoing · 6 parts
in engineering
A working dev's field guide to coding agents — what "agentic" actually means, Claude Code in the terminal, the feedback loops and guardrails that make it trustworthy, and shipping real features from prompt to deploy.
- 1 What "Agentic" Actually Means Agent, chatbot, autocomplete — three different things the industry keeps calling the same word. What makes a tool agentic, and why a loop plus tools plus judgment changes how the work feels.
- 2 Claude Code in the Terminal Where the agent lives is the shell — the same place I always worked. Setup, project memory with CLAUDE.md, slash commands, and hooks, using the rebuild of this very site as the case study.
- 3 Designing the Feedback Loop An agent is only as good as the ground truth it can check itself against. Tests, build scripts, type checks, and screenshots as the agent's eyes — and why a fast loop beats a smart prompt.
- 4 Subagents and Parallel Work One agent in one context window hits limits. Delegating research to subagents, isolating context, and working in parallel with git worktrees — plus the trap of over-delegating.
- 5 Guardrails and Trust An agent that can run any command is powerful and dangerous in the same breath. Permission modes, destructive-action gates, and reviewing diffs — how I let an agent ship to a live Synology box without losing sleep.
- 6 Building Real Things The whole loop, end to end: one feature from a prompt to a deployed change — the wins, the dead ends, and the honest accounting of what agentic engineering is and isn't good at yet.