Orientation: what this chapter changes
Artifact: A clear picture of what changes between Ch 22 and Ch 23, and what you'll have shipped by the end
1. The promise
You'll know what makes this chapter different from Ch 22, what the assignment asks for, and what the chapter does not try to teach. By the end of this unit you'll have picked the feature you'll ship in the assignment.
2. The mental model
Ch 22 taught you to drive a car with a local model under the hood. Ch 23 teaches you to dispatch a courier. The courier — an agent, a program that takes a goal, picks tools, calls them, and reports back — runs jobs you assign. The harness is the dispatcher's office around the courier. You'll spend this chapter learning the controls in that office.
(DIAGRAM: Ch 22 model vs Ch 23 agent — comparison illustration — left: editor with inline completions, right: terminal with the agent driving tools)
3. You do it, with me
There's nothing to install in this unit. Read along.
Open the Click to Career chapter index in your browser and find Ch 23.
(SCREENSHOT: the Click to Career chapter index for Ch 23 — full window — the orientation unit highlighted)
Here's what changes between Ch 22 and Ch 23.
Before this chapter. You open a chat box, ask a question, copy an answer back into your editor. The model gives you a paragraph; you do the work.
After this chapter. You hand the agent a folder, a goal, and constraints. The agent reads files, writes files, runs commands, and reports back. You're the dispatcher, not the typist.
(SCREENSHOT: a terminal with Claude Code prompt visible, ready for input — full window — prompt highlighted)
What this chapter teaches. The control surface of one specific harness — Claude Code. Permission modes. Slash commands. Skills. Memory. Hooks. Subagents. Plan mode. Loop mode. MCP. Cost.
What this chapter does not teach. Prompt engineering as a craft. The Vibe to Spec course covers that. You'll learn enough prompt shape here to ship the assignment, no more.
The assignment, previewed. You'll ship one real feature end to end through the harness. Not a refactor. A feature a user can interact with. Pick it now so the units feed it. Examples: a new page with a form, a new API endpoint your existing frontend calls, a new database column with the migration and UI to set it.
The honest cost statement. This chapter uses real Anthropic API calls or a Claude subscription. Expect to spend on the order of A$10 to A$30 on the assignment if you're on the API path. Less if you're on Pro. Unit 23.14 covers cost honestly.
(SCREENSHOT: a finished feature shown side-by-side — the student's terminal on the left running Claude Code, the result in VSCode on the right — full screen)
Terms you'll meet first here.
- Agent — a program that takes a goal, decides which tools to call, calls them, reads the results, and decides the next step.
- Harness — the program around the model that runs the tools, enforces permissions, and keeps state between turns.
- CLI (command-line interface) — the program you talk to by typing text and reading text. No buttons. No windows. Met in Ch 2; re-introduced here.
4. What you should be seeing
Three things in your head by the end of this unit:
- The split. Ch 22 was the model. Ch 23 is the agent and the harness around it.
- The assignment. One real feature, shipped through the harness, with a one-page reflection.
- A feature picked. You wrote down the one feature you'll ship for the assignment.
If you can't name the feature you'll ship, stop and pick one before moving on.
5. Common stumbles
- You think this chapter teaches you to "talk to AI better". That's Vibe to Spec. This chapter teaches the harness controls — permissions, hooks, memory, the rest.
- You pick a half-baked feature for the assignment. Scope down now. A new page with a form beats a new feature with auth, payments, and a dashboard.
- You skip this orientation thinking the chapter starts in 23.1. The framing here — agent versus model, harness as dispatcher — runs through every unit.