3/5 GitHub Copilot Beyond Code: Your AI Writing Partner
Previous: VSCode Isn’t a Code Editor. It’s an AI Workspace.
In Post 1, you wrote a fairy tale outline. In Post 2, you put it in VSCode with live preview and Git. The outline is good — but it’s still an outline. Bullet points, not a story.
Now Copilot turns it into one.
Copilot Chat vs Inline Suggestions
GitHub Copilot works in two modes:
Inline suggestions — Copilot completes what you’re typing. Start a sentence and it predicts the rest. Useful for code, less useful for creative content because it guesses where you’re going rather than where you want to go.
Copilot Chat (Ctrl+Shift+I) — You ask, it responds. This is the mode that matters for content. You select text, describe what you want, and Copilot transforms it. You stay in control.
For content creation, Chat is your primary tool. Inline suggestions are a bonus.
The Context Advantage
When you chat with an AI in a browser, it sees one message at a time. In VSCode, Copilot sees:
- The file you’re editing — your fairy tale outline
- Other open files — related content, reference material
- Your selection — the specific section you want to transform
#filereferences — you can point Copilot to any file in your project
This is why the workspace model from Post 2 matters. Copilot doesn’t work in a vacuum — it works in the context of your project.
Hands-On: From Outline to Story
Open tales/lady-binary.md in VSCode. Select the entire “Lady Binary’s Method” section:
## Lady Binary's Method
1. Open the middle scroll
2. Too high? Ignore the upper half
3. Too low? Ignore the lower half
4. Repeat with what remains
5. Each peek halves the searchOpen Copilot Chat and type:
Expand this into a narrative fairy tale section. Keep the voice whimsical but technically accurate — Lady Binary is a clever librarian explaining her method to a young apprentice. Each step should be a scene, not a bullet point. About 300 words.
Copilot will generate something like:
## The Method
"Watch carefully," Lady Binary told the young apprentice, who had spent
all morning searching the scrolls one by one, starting from the very first.
She walked past thousands of shelves to the exact center of the great
library. "Scroll five hundred thousand," she read aloud. "We seek scroll
forty-two. Is forty-two greater or lesser than five hundred thousand?"
"Lesser," said the apprentice. "Far lesser."
"Then everything above this point" — she swept her hand across the
upper half of the library — "we shall never visit." Half the library
vanished from consideration in a single glance.
She walked to the center of what remained. "Scroll two hundred fifty
thousand. Still too high." Another half disappeared.
The apprentice's eyes widened. "You're not searching. You're eliminating."
"Precisely." Lady Binary smiled. The library that held a million scrolls
was shrinking with every step. A million became five hundred thousand.
Then two hundred fifty thousand. Then one hundred twenty-five thousand.
Twenty steps later, she held scroll forty-two.
"But how?" the apprentice stammered. "I would have searched for weeks!"
"You searched *linearly*," Lady Binary said gently. "One after another,
like counting grains of sand. I searched by *halving*. Each peek removes
half of what remains. A million scrolls, twenty peeks. That is the power
of the middle."Review it. Keep what works, rewrite what doesn’t. Commit the good version.
Hands-On: Adding a Mermaid Diagram
Now let’s visualize Lady Binary’s method. In Copilot Chat:
Create a Mermaid flowchart showing the binary search algorithm as Lady Binary would describe it. Use fairy tale language for the labels. The hero arrives, checks the middle scroll, then goes left or right.
Copilot generates:
```mermaid
graph TD
A["Lady Binary enters the library"] --> B["Open the middle scroll"]
B --> C{"Is this scroll #42?"}
C -->|"Yes!"| D["Quest complete!"]
C -->|"Too high"| E["Ignore the upper half"]
C -->|"Too low"| F["Ignore the lower half"]
E --> G["Walk to the new middle"]
F --> G
G --> B
`` `Paste this into your Markdown file. The Mermaid preview extension from Post 2 renders it as a real diagram — right in your editor.
For the full catalog of Mermaid diagram types (flowcharts, sequence diagrams, state diagrams, and more), see Mermaid Diagrams: When Text Becomes Visual.
Prompt Patterns That Work
After working through dozens of content projects, these patterns consistently produce good results:
| Pattern | Example | When to use |
|---|---|---|
| Expand | ”Expand these bullets into a narrative paragraph” | Turning outlines into prose |
| Transform | ”Rewrite this section for a younger audience” | Adapting tone or level |
| Visualize | ”Create a Mermaid flowchart showing this process” | Generating diagrams from descriptions |
| Structure | ”Organize these notes into a Markdown document with headings” | Structuring raw material |
| Critique | ”What’s missing from this outline?” | Getting feedback before expanding |
When Copilot Helps vs When It Struggles
Copilot is good at:
- Expanding structure into prose (outline → story)
- Generating diagrams from descriptions
- Reformatting content (bullets → table, list → narrative)
- Consistent style across sections (once you establish the voice)
Copilot struggles with:
- Original voice and humor (it defaults to generic)
- Domain nuance (it doesn’t know your specific fairy tale world)
- Knowing when to stop (it will happily overwrite your best lines)
- Maintaining constraints across sessions (it forgets your rules)
That last one — maintaining constraints — is the biggest problem. You tell Copilot “keep the voice whimsical” and it does. For one response. Next prompt, it defaults to generic. Next session, it’s forgotten everything.
Post 4 solves this with AGENTS.md — a file that teaches Copilot your rules permanently.
What You’ve Built
Your fairy tale project now contains:
- An outline (Post 1)
- A VSCode workspace with Git (Post 2)
- A full narrative expanded by Copilot (this post)
- A Mermaid diagram of the algorithm (this post)
All in Markdown. All version-controlled. All editable by both you and AI.
But Copilot still doesn’t know your rules. It doesn’t know about the five-part story structure, the character archetypes, or the boundary that says “never use actual code syntax in the tales.” Next post, we fix that.