From Prompt to a Reusable AI Teammate

Welcome and what you will build

Okanagan AI Growth Opportunities (OKAI GO) is a set of hands-on lessons for anyone who wants to use generative AI in practical, thoughtful ways.

This foundation lesson helps you extend beyond one-off prompting. You will create a reusable AI teammate, give it a standing brief and useful knowledge, then discover how that context can carry into fresh conversations.

Why context matters

When you supply useful context, you can get much better results than when you do not. Context is just all of the information you provide the AI tool about your situation, your needs, what you want the tool to do, any other relevant information (although, often irrelevant information will be included too), and your desired output. Context comes from the prompt that you type, cut

Context comes from the prompts that you type, but it also comes from a bunch of other sources:

• System Prompt — A prompt written by they tool provider to give comprehensive guidance to the tool on how to behave. You can see all of the Claude system prompts here.
• Memory — information the tools retain across conversations. Most platforms allow you to turn this on or off. ChatGPT's Memory FAQ
• Your current prompt — what you just typed in.
• Conversation history — all of your past conversation in the window you are working in is included in the context (to an extent...there is a maximum number of words that can be included)
• Custom instructions — the standing role and behaviour saved in a reusable workspace or configuration.
• Uploaded files — the content of any uploaded documents are added to the context the AI can draw on.
• External tools — mail, files, websearch, and applications can be reached through connectors.
• Retrieved or searched material — sources brought into the conversation as needed.

Reusable context is worth building because it lets you return to do a similar task without restating all the details..

Two terms that will help us speak more precisely about AI tool capabilities:

• Token — an indivisible chunk of content an AI reads or writes. A token is often about the size of a word-sized, although a word is often more than one token. A good approximation is that a token is 3/4 of a word (e.g. "cat" is a token. so is "s". "cats" is two tokens). Usage limits and cost are commonly measured in tokens.
• Context window — the maximum number of tokens the AI can keep “in mind” at once. Instructions, files, and the conversation share this working space. As it fills, older material will be dropped. This is why if you only ever use one window for all your conversations, the AI will forget stuff from the beginning of the window.