A tailored assistant is not a smarter machine. It is the same machine, better briefed. Standing instructions give it its job; your documents give it your world.
If you remember one thing: every tailored assistant has a tailor. When the tailor is you, that is the power. When it is someone else, that is the question.
| Method | What it does | Who can use it |
|---|
| System prompt | Standing instructions and context, written once and silently attached to every conversation, “valid for every single interaction” | Anyone, in minutes |
| Retrieval (RAG) | Searches the documents you gave it, your knowledge base, and quietly attaches the best passages to your prompt behind the scenes | Anyone, no code |
| Fine-tuning | Developers train the model further on examples from their field, so the model itself changes, not just what it reads | Specialist teams; “typically harder and more expensive” |
The first two are, in the course’s words, “something that most of us can do.”
| Put in the standing instructions | Leave for the task’s prompt |
|---|
| Who you are | The specific posting, draft, or document of the day |
| How you like output (length, tone, format) | One-off instructions for this task only |
| The assistant’s job | Anything that changes from task to task |
Rule of thumb: a system prompt applies to every conversation, so it should hold only what is stable.
| Line | Meaning |
|---|
| A tailored assistant is a specialist | Its lane is exactly as wide as what you fed it |
| Confidence does not mark the lane’s edge | Its confidence will not change at the edge, but its accuracy will; ask what it was given before you rely on it |
| Grounding reduces invention, not misreading | Retrieval can cut down invented answers, fewer not zero; for answers that matter, follow the reference back to the page |
| Question | Why it matters |
|---|
| Who wrote its instructions? | “the designer has a lot of control on how it’s going to respond” |
| Who chose its documents? | Its helpfulness is shaped by what it emphasizes, plays down, or never brings up |
| What is it tuned to want? | Nothing sinister is required; a store’s assistant is tuned to sell |
| Door | What it looks like | Current-state hedge |
|---|
| Sight | Drop in a photo, an error message, a form in another language, and ask about it | As of mid-2026, most major assistants can look at an image; what any tool does well this month is a moving target |
| Data | Upload a spreadsheet and ask for a chart in plain English | Not serious analysis, but a question that once needed a specialist is now a sentence |
| Voice | A spoken conversation at nearly the pace of a phone call | As of mid-2026, many assistants offer it; in most tools speech becomes text, the machine predicts, and the answer returns as sound |
Dated markers: the course demonstrated all this in 2024 with custom GPTs and Microsoft’s Copilot. Those products kept changing while the pattern spread; treat the names as historical markers, not a current product guide.
| Pitfall | Correction |
|---|
| Trusting a specialist outside its lane | Ask what an assistant was given before you rely on it; no politeness saves it past the edge of its documents |
| Stuffing the standing instructions with everything | Keep only what is stable; per-task detail rides in with the task |
| Forgetting the tailor | An assistant that arrives knowing its job learned that job from someone; their goals ride along in every friendly answer |
| Expecting retrieval to end wrong answers | The machine can cite your document and still misread it; verify what matters at the source |
| Line | Meaning |
|---|
| ”valid for every single interaction” | The course on what a system prompt does with whatever you put in it |
| ”something that most of us can do” | The course on system prompts and retrieval, the practical heart of the lesson |
| ”the logic behind the tools is the same” | The course on why new modalities do not need a new mental model |
| A tailored assistant is the same machine, better briefed. | The lesson’s closing punch: job description plus filing cabinet, not new intelligence |
