Key takeaways

• Prompt engineering is wording one instruction well. Context engineering is building a system that assembles everything the model sees, automatically and dynamically.
• Context includes the system prompt, the user request, conversation history, long-term memory, retrieved information, tool definitions, and output format rules.
• Most agent failures today are context failures, not model failures.
• The same model can produce a robotic answer or a magical one; the difference is the context it was given.
• Agents make this skill critical because they run in long loops that flood the context window unless it is actively managed.

Continue to Chapter 2 to see the specific ways context goes wrong.