context engineering
The suggestion quality depends more on what the model can see — open files, types, tests, neighbouring code — than on the prompt you type.
§ 02 — Eras · 2021
Copilot
the comment is the spec; the model writes the code
Left: era context, features, and the 101 hello. Right: 3 deeper 201 lessons, the factorial specimen, and where Copilot lives today — each column scrolls on its own.
The problem it solves
Most programming is boilerplate already written somewhere. The model predicts the code from intent — a comment as spec.
The problem it carries
Confident wrongness: plausible code, wrong licenses, and a debate about what skill remains when the first draft is free.
Key features
- prediction as the interface
- The next lines are guessed from context; you accept, reject or steer.
- the comment as spec
- A comment in the right context is often enough to define the function to be written.
- whole-file awareness
- Suggestions consider the file, not just the line under the cursor.
101First contact — hello world. 201 — 3 lessons on mechanisms you meet after the factorial specimen: the sharp edges practitioners actually use.
201 — going deeper
The 201 assumes you have read the 101 and the specimen. Each lesson is one mechanism — what it is for, how it looks in code, and the trap or payoff that makes Copilot distinctive at scale.
review and refactor
The strongest workflows treat the model as a first draft: generate, then diff, then test, then simplify — never accept a large change unread.
provenance
Generated code carries the licence of whatever it was trained on, and the liability of whatever it produced. Review provenance like any other dependency.
The specimen — factorial
The completion: a comment is the spec, the model writes the body — prediction becomes a programming interface.
# reverse a linked list, iteratively
def reverse(head):
prev = None
while head:
head.next, prev, head = prev, head.next, head
return prevRun it today
No toolchain — an editor. Install the Copilot extension in VS Code, type the comment, and accept the ghost text with Tab.
Where it is now
Millions of developers accept its suggestions daily. Treat it as a very fast, very confident junior pair: verify everything, mind the license question, and keep reading the code it writes.