Context engineering

Definition

The deliberate selection and maintenance of instructions, knowledge, tools, state, examples, and artifacts within a finite model attention budget. The control surface is no longer only the prompt; it is everything the model can see at the moment of decision.

A concrete practice with rapidly evolving techniques and tooling.

Why it matters

The control surface is no longer only the prompt; it is everything the model can see at the moment of decision.

The operating shift is from “More tokens and larger prompts” to “Smallest sufficient high-signal state.”

System anatomy

Practice 1

Use progressive disclosure and just-in-time retrieval

Practice 2

Budget and prune model-visible state

Practice 3

Reset context with structured handoff artifacts

Important distinctions

More tokens and larger prompts

This concept moves the engineering system toward smallest sufficient high-signal state.

Implementation signals

• Use progressive disclosure and just-in-time retrieval
• Budget and prune model-visible state
• Reset context with structured handoff artifacts

Failure modes

• Context accumulation that buries the objective
• Untrusted retrieved content silently steering action

Sources and further study

1. Anthropic — Effective context engineering for AI agents

   A working model for treating model-visible context as a finite resource that must be selected and maintained.

   Use in this library: First-party case study. This first-party account documents one organization, product, or implementation context and should not be generalized without local evidence.

   engineering · engineering · Published 2025-09-29
2. Anthropic — Effective harnesses for long-running agents

   Patterns for incremental progress and durable handoffs across context windows and extended execution.

   Use in this library: First-party case study. This first-party account documents one organization, product, or implementation context and should not be generalized without local evidence.

   engineering · engineering · Published 2025-11-26
3. OWASP GenAI Security Project — Memory Is a Feature. It Is Also an Attack Surface

   A focused account of why persistent agent memory requires integrity, provenance, and poisoning defenses.

   Use in this library: First-party case study. This first-party account documents one organization, product, or implementation context and should not be generalized without local evidence.

   engineering · engineering · Published 2026-05-13