Context engineering is choosing what the AI gets to read before it answers. That is the whole idea. The rest is detail.
Where the term came from
The term spread in June 2025. Shopify's chief executive, Tobi Lütke, said he preferred it to prompt engineering. Andrej Karpathy, a well-known AI researcher, backed it a few days later.
In September 2025, Anthropic published the guide most people point to. It was written for people who build AI agents. The idea behind it is useful to anyone who types into a chat box.
Context engineering vs prompt engineering
Prompt engineering is about how you word the request. Context engineering is about everything else on the page.
That includes your instructions, your facts, your examples, the files you attached, the chat so far and anything a tool looked up. The model reads all of it together and answers from all of it.
Think of briefing a freelancer. The exact wording of your request matters a little. The brief they received matters a lot. Give a good writer a bad brief and you get a bad draft.
Why more context is not better
It is tempting to paste everything. Models can read a lot of text now, so why not?
Because attention is limited. Anthropic's guide describes it as a budget: every extra line the model reads takes attention from the others. The goal is the smallest set of information that still gets the result.
Old chat history has the same cost. A long thread full of dead ends gives the model more wrong things to pick up. This is also one reason ChatGPT seems to forget in long chats.
What it looks like for a business of one
You already do context engineering. You do it by hand, each time you retype your prices or explain your customers to a new chat. Doing it well means doing it once.
- Write your business facts down once, in plain sentences.
- Give each task only the facts it needs.
- Put a rule next to the request it applies to.
- When a fact changes, change the source. Never patch the copies.
There is a full guide to writing that page on this blog.
Three mistakes to avoid
- Dumping every document. The model reads all of it and focuses on none of it.
- Old facts. Last year's price list, still attached to a project, will end up in a quote.
- Rules that disagree. One file says be brief and another says be thorough. The model picks one, and you will not know which.
Do you need to learn it?
If you build AI agents, yes. Read Anthropic's guide, then read it again.
If you run a business, you need the habit more than the theory. Before you ask, decide what the model should know. Then check that it knew it.
Nive is being built around that habit. It holds how you run your business and gives each job the right part of it, starting with cold emails. Join the waitlist to get the demo first.