A new term is spreading through the AI community: context engineering. In a June 30 blog post, Philipp Schmid argues that the discipline of building effective AI agents is less about crafting clever prompts and more about designing dynamic systems that provide the right information, tools, and format at the right time. Tobi Lutke calls it ‘the art of providing all the context for the task to be plausibly solvable by the LLM.’

Schmid breaks down context into multiple components — system prompts, user prompts, conversation history, long-term memory, retrieved data, available tools, and structured output — and argues that most agent failures today are not model failures but context failures. He contrasts a ‘cheap demo’ agent that responds robotically with a ‘magical agent’ that uses rich context to craft a perfect reply.

On Hacker News, the post drew over 900 points and 500 comments. Simon Willison, who has been writing about context engineering independently, says the rebrand is needed because prompt engineering has become associated with ‘typing prompts full of stupid hacks.’ Drew Breunig notes that many of these techniques date back to pre-ChatGPT NLP research. But some commenters voice skepticism, calling it a manufactured buzzword driven by top-down tweets from influential figures.