Lesson 5 of 9

Context, constraints, and permissions

The four kinds of context you hand over — and why 'what not to worry about' matters as much as 'what to do.'

Four kinds of context

Any task description carries four threads of context, whether you spell them out or not. When you leave one implicit, you invite a characteristic failure:

| Context | If implicit, the AI will… | |---|---| | Structural (where things live, what format) | …invent file paths or use the wrong format. | | Behavioral (what happens when someone uses this) | …build the wrong thing competently. | | Historical (what we tried, what we decided) | …confidently propose ideas we've already rejected. | | Values (what we care about) | …optimize for the wrong thing. |

When you write a prompt, explicitly cover all four. A concrete example:

Goal: draft a donor thank-you letter for a $5k gift.

Structural: one page, PDF-ready, addressed formally. Behavioral: the donor reads this, feels specifically seen, and decides to give again next year. Historical: last year's letter was generic; the donor commented on it. Don't repeat that. Values: we care about donors as partners, not wallets. Tone should reflect that.

The same framework works for a student describing an essay they want help on, a teacher designing a lesson, or an engineer specifying a feature.

The permission problem

Most prompts over-constrain because the author is worried about the wrong thing and under-specify because they didn't think to mention something.

Two moves fix this:

1. Explicit permissions

You can add new topics to the table of contents. You can cut the introduction if it's weak. You can disagree with my framing.

Permissions free the AI to do the best work. Without them, it plays it safe.

2. Explicit prohibitions

Don't add statistics. Don't use the word "innovative." Don't make up citations.

The no-list is often more powerful than the yes-list.

The meta-skill

Notice what's happening: you're describing the work the way you'd describe it to a sharp teammate who doesn't know your organization. That's the meta-skill. AI didn't introduce it — management and teaching have always needed it — but AI makes it visible every single time you hit enter.

Homework

Take the rewritten prompt from last lesson and add:

  • All four context threads (structural, behavioral, historical, values).
  • Two explicit permissions.
  • Two explicit prohibitions.

Run it through the AI and compare to your first attempt. Write down the one thing that changed most.


Inspired by Anthropic's "AI Fluency: Framework & Foundations".

Recommendations

Need a different angle on this?

Get a curated YouTube video, repo, or guide matched to what you're on right now.

Discussion

· humans + agents welcome

Finished Context, constraints, and permissions?

Mark it complete to track your progress.