Jargon:Context engineering
1 min read
Say this at the party
“It's prompt engineering grown up: not just wording the question, but choosing everything the AI gets to see before it answers.”
In plain English
An AI only knows what is in front of it at that moment: your message, its instructions, any files, notes or search results it has been handed.
Context engineering is deciding what goes into that pile. Too little and the AI guesses. Too much and it gets muddled, like a person handed a 300-page binder for a two-minute question.
It matters most for AI agents, which work for a long time and keep collecting stuff as they go.
Where it came from
The phrase took off in June 2025, when Shopify's CEO Tobi Lütke said he preferred "context engineering" to "prompt engineering", and Andrej Karpathy agreed. Anthropic published a guide to it in September 2025.
See it in action
Before asking an AI to plan his week, Tom gives it his calendar, his to-do list and last week's notes, but leaves out his 2,000 unread newsletters. The plan is better because of what he left out.
What people get wrong
- More context isn't always better. Stuffing everything in can make the AI worse, not smarter.
- It doesn't replace good prompts. The wording still matters; it is just one part of the picture.
Where Ugenie fits
Ugenie keeps notes outside any one chat (a "second brain"), so the right background can be handed to whichever AI you use next.
Get Ugenie →Go deeper
- PostSimon Willison, 2025Context engineeringShort note capturing the moment the term caught on, with the Lütke and Karpathy quotes.
- GuideAnthropic, 2025Effective context engineering for AI agentsHow one AI lab thinks about choosing what an agent gets to see.
- ArticleWikipediaPrompt engineeringSets out how context engineering relates to prompt engineering.
Related jargon
Drafted with AI. The AI Council (Claude, ChatGPT and Gemini) gave counsel. Phin signed it off.