Map
Framework

The five-part context bundle

The question it answers

When a team gives AI the same kind of task every week, such as reports, customer replies or meeting summaries, what bundle should be ready so results are not generic and nobody rewrites the brief?

Origin
Boonstra, n.d.; Anthropic, 2025; Brown et al., 2020
Diagram type
Hub diagram
Level of analysis
Individual
Topics
AI for leaders
Glass sphere with a glowing neural-network core, icon for AI for leaders

Diagram

The five-part context bundle

The context bundle sent to AI
Role
Context
Examples
Output spec
Constraints

When to use it, and when not to

Use it when

Use it when the same kind of task keeps coming back and you want a ready-made bundle to paste in each time. The five parts correspond to techniques described in (Boonstra, n.d.): assigning a role, supplying context, giving sample examples, being specific about the output, and setting constraints.

Avoid it when

The five parts are this site’s grouping of (Boonstra, n.d.) and (Anthropic, 2025); neither source prescribes this exact set of five. For a one-off task a short question is usually enough: (Anthropic, 2024) advises starting from a simple prompt and adding parts later.

How to apply it, step by step

  1. Role: who is doing this job and who will read the result ((Boonstra, n.d.)).
  2. Context: only the information this task needs. (Anthropic, 2025) advises finding the smallest set of information that still lets the model do the right thing.
  3. Examples: a few representative samples rather than a list of every edge case ((Anthropic, 2025)); examples in the prompt let a model take on a new task without retraining ((Brown et al., 2020)).
  4. Output spec: the format, length and structure you want, stated concretely ((Boonstra, n.d.)).
  5. Constraints: what must not be done. (Boonstra, n.d.) leans towards instructions that say what to do rather than lists of prohibitions, and keeps constraints for blocking harmful content or enforcing a strict format.

Pitfalls

Pitfall

Stuffing whole documents into the bundle. (Anthropic, 2025) describes “context rot”: as the number of tokens in the context grows, the model’s ability to recall information accurately falls.

Sources

References

Listed A–Z by author surname, Harvard style. A source with a DOI links to https://doi.org/…; one without links to its page or says plainly that it has no DOI.

  1. Anthropic (2024) Building effective AI agents. Anthropic Research.No DOIanthropic.comCited in 6 articles
  2. Anthropic (2025) Effective context engineering for AI agents. Anthropic Research.No DOIanthropic.comCited in 4 articles
  3. Boonstra, L. (n.d.) Prompt engineering [whitepaper].No DOIkaggle.comCited in 1 article
  4. Brown, T.B., Mann, B. and Ryder, N. (2020) Language models are few-shot learners. (arXiv preprint arXiv:2005.14165).doi:10.48550/arXiv.2005.14165Cited in 1 article