It is the briefing around the prompt
Ask AI to write a client proposal with no background and you will get something broad.
Give it the client’s problem, the agreed scope and an example proposal you like, and it has a much clearer job.
The prompt is one part of that briefing.
Context engineering also covers which documents are included, which version is current and what information should be retrieved when it is needed.
The free AI business starter shows a full briefing you can copy.
Prompt engineering vs context engineering
Prompt engineering focuses on the instructions you write.
Context engineering takes the wider view: everything available to the model at that moment.
Rewriting “make it better” can help, but it will not replace a missing client brief.
A practical example: your content voice
Suppose you want an outline for a newsletter. Give the AI the topic, your actual notes and one short sample of your writing.
Add who the reader is and what you want them to understand by the end. That is a more useful starting point than a long instruction to “sound authentic.”
Job: outline a newsletter about preparing for client calls.
Source: my notes from setting up the process.
Voice: this paragraph I wrote.
Reader: a freelancer who uses AI occasionally.
Output: an opening, three points and a practical exercise.
Why not upload every document?
More information can introduce old decisions, repeated instructions and unrelated material.
A current one-page brief may be more useful than a folder containing three versions of your offer.
Label the current version and remove information that no longer applies.
How to improve the context
- Look at the answer that disappointed you.
- Identify what the AI did not know: the audience, the constraint, the facts or the preferred format.
- Add that missing piece rather than rewriting every instruction.
- Try the task again and compare the results.
- Save the updated briefing where you can find it next time.
Is this the same as memory?
No. Memory is one way information may be retained. Context is the information available for the current task.
A saved fact is only useful when the system can bring it into the work at the right time.
Important project details should stay easy to inspect, correct and reuse.
Next: how context helps a personal AI assistant.
Definition and technical background: Anthropic: Effective context engineering.