Context Management 104: examples และ constraintsContext Management 104: examples and constraints

Examples are some of the strongest context you can give AI.

A rule says what you want. An example shows what you mean.

Why examples work

Compare these two instructions:

Write in a clear, practical style.

and:

Use this style:
- short paragraphs
- direct explanations
- concrete examples
- no hype
- Thai terms can keep common English words like task, context, output

Example:
[paste a short sample]

The second version is easier to follow because the model sees the pattern.

Use one good example first

Start with one strong example instead of ten weak examples.

A good example should be:

• close to the task
• short enough to inspect
• clean enough to copy patterns from
• labeled with what makes it good

Example:

Good example:
[paste example]

Why it is good:
- opens with the decision
- uses beginner language
- includes one exercise
- avoids selling the paid course

The explanation matters. It tells the model what to imitate.

Add a non-example

A non-example shows what to avoid.

Do not write like this:
[paste weak example]

Why this is wrong:
- too abstract
- uses buzzwords
- no concrete practice
- implies a certificate that the free workshop does not include

Non-examples are useful when the model keeps drifting into a pattern you dislike.

Constraints are boundaries

Constraints tell the model what must stay true.

Good constraints:

Constraints:
- Thai is the default language.
- Keep product name as "Mervia Academy".
- Do not promise instructor access.
- Use /workshops/{slug}/ route examples.
- Keep each section under 120 words.

Weak constraints:

Make it good.
Make it premium.
Make it modern.

If a constraint is subjective, make it observable.

Instead of:

Make it beginner-friendly.

use:

Beginner-friendly means:
- explain new terms the first time they appear
- avoid assuming coding knowledge
- include one copy-paste template
- end with one practice task

Acceptance criteria

Acceptance criteria are checks for the final output.

Acceptance criteria:
- Includes English and Thai sections.
- Has one practical template.
- Includes one exercise.
- Does not mention paid access.
- Ends with the next lesson link.

This helps both the model and the human reviewer.

Put examples after the task brief

A useful order is:

Task brief
Sources
Examples
Non-examples
Constraints
Output format
Acceptance criteria

Do not hide the task after long examples. The model should know the job before reading patterns.

Practice: build an example pack

Choose a task you often give AI.

Create this context block:

Task:

Good example:

Why this example is good:

Non-example:

Why this example is wrong:

Constraints:

Acceptance criteria:

Then ask AI to produce one new output and explain which parts of the example it followed.

Next: Context Management 105 - compress and refresh context.