AI Concept Atlas
FIELD NOTES / 01
Guided learning · 6 lessons

What the model can see

Distinguish current context, retrieved evidence, and retained memory.

Lesson 3 of 6

Context Engineering

Once capacity is limited, choosing the information in it becomes a design task.

Context engineering is the practice of selecting, organizing, and maintaining the information a model receives while it works. It includes instructions, tool descriptions, retrieved material, and conversation state.

[Anthropic]

In practice

A travel assistant keeps the current itinerary and budget in view while removing obsolete search results from its next request.

[Anthropic]

A little deeper

The task repeats as an agent runs: useful evidence arrives while older material can become irrelevant. The aim is useful context within a finite capacity, rather than simply adding more text. [Anthropic]

A common mix-up

Context engineering means making one prompt longer.

It includes managing changing information across multiple model calls. [Anthropic]

Helpful to know: Context Window

Explore Context Engineering and its connections ↗

← Previous lesson