Part (topic)

Part is a component of a larger AI-driven process, and understanding it matters because AI development depends on breaking work into clear, manageable pieces. In the build phase, where creators spend roughly 80% of their time, you make the skills (reusable instructions for specific tasks) that your workflow needs. The borrow phase takes about 10% of your time, where you adapt skills from people inside your company rather than downloading random ones from strangers. This split matters because skills are where your unique process lives, so borrowing blindly can bias the AI toward someone else's topic instead of your own.

Consider how AI handles a long document. You ask it to list the section names of each part so you can question those parts and dig deeper when necessary. That is more than summarizing — you are extracting specific pieces you can act on quickly. Similarly, when a question needs broad coverage, the AI normally researches pieces sequentially, one after another. Sometimes you want it to spawn separate efforts instead. The lesson is consistent: identify each part, decide what it does, and rate the steps of a task rather than the task as a whole. That move drives every automation you will build.

Why does this matter for AI development? Because without structured parts, the AI behaves like a smart intern who is guessing. The context you supply — your subject matter expertise, your brain, your IP (intellectual property) — is what makes outputs unique. Context cards typically hold three parts: facts, preferences, and memories. Each orients the model so its work fits your situation.

To use Claude step by step, start by naming the task, its output, and constraints (limits on the result) so Claude has an actionable target. Then structure your prompt with XML tags (labels that mark sections), separating instructions, context, input, and documents. This layout helps Claude decide which section holds facts, which holds instructions, and which holds constraints. Add three to five relevant, diverse examples wrapped in example tags, covering a normal case and an edge case rather than repeated copies. Examples turn abstract instructions into concrete patterns Claude can imitate.

For multi-step work, use decomposition (breaking tasks into smaller steps). Inside one prompt, tell Claude to work step by step: first extract key figures, then analyze trends, then write the recommendation. Prompt chaining (separate messages using prior output as input) works too: research the market, draft the memo, then tighten it. Before Claude writes, tell it to ask you questions; it may flip and start interviewing you. When output becomes crowded, let an evaluation decide whether extra context earned its space.

These certified developer practices carry across parts: the certified architect track builds with the Claude API, while the certified associate track covers prompting and task execution.

When a Claude result misses the mark, resist the urge to pile on more instructions. Instead, find why the draft failed before asking for more work. Name the failure in plain language: off topic, invented, misplaced, slow, stale, or degraded. For example, if a two-line release note becomes a generic essay, name it a format failure (wrong shape for the job). A clear symptom stops you from sending a vague, expensive follow-up.

This habit comes from Domain 7 (the troubleshooting exam section), which is small in exam weight but appears everywhere. The certified associate path stresses inspecting the evidence before you act. Sound facts can still arrive in the wrong length, tone, or audience, so check prompt shape first and repair the smallest sound layer.

Best practice: turn each mistake into a permanent fix. When something breaks, don't just patch it and move on — work with Claude Code to ask what could be improved so it never recurs, writing the fix into a skill (a saved reusable instruction). People building agents treat every issue as a chance to strengthen the system. One creator built mistakes into their skill to make it bulletproof, saying "write this into the skill such that it never happens again."

Also remember the boundary from Part 8: Claude speeds up recoverable work like drafting, summaries, and brainstorming, but humans stay accountable for legal, medical, financial, and hiring decisions. Troubleshoot the layer, not just the symptom.