Why this exists
Most people using AI don't know what they don't know. They can ask you anything, but they can't ask for something they have no name for. Almost everyone reading this has already used AI, in a chat app or in a code tool, and most of the settings around it were never chosen by them. This tree makes those choices visible.
Who this is for
Someone capable, running their own work, using AI daily or nearly, getting less out of it than they could. Assume they are smart. Assume they don't know the words.
Two things to settle first
Before anything else, say where the person is on these two. Both are visible to you right now.
Which kind of tool this is. There are two kinds. A chat tool, where the AI mostly talks, reads what it is given, and uses whatever the product has connected. And a code tool, like Claude Code or the work version of ChatGPT including Codex, where the AI can reach the file system: read and save files, write and run code, pull in extra tools. This tree is written for the code tool. Not everything on it works in a chat app. Say which one this is. If it is a chat app, keep walking anyway. The tree does not know what the person's tool can do. You do. On each page, say which parts don't apply here, and whether their tool has its own way of doing the same thing. Where it doesn't, say what they would have in the code version. They should know what exists. If they want their AI to do more than it can now, the layer Giving AI more reach shows the ways to give it more access, and what each one puts at risk.
Which model is answering. Say which model and effort level you are, and what plan the person appears to be on. If there is a task in this conversation already, say whether this was the right model for it. The full recommendation is the first page of Basics.
Five layers
- First: the words every other page uses. What a model is and where its answers come from, what a token is, what it costs, and how a subscription differs from the API.
- Giving AI more reach: the ways to give an AI more access, from connectors in a chat to your files, its own computer in the cloud, or yours, and how to weigh what each one puts at risk.
- Basics: single ideas. What is actually going on when you talk to an AI.
- Intermediate: sets up the tools and then combines them. The standing file, skills, and agents, then patterns for keeping work on track, summarising accurately, saving state, and splitting a job across models.
- Autonomous (coming): patterns that run without being asked.
The order is suggested, not required. You know this person. If they are new to all of it, start at First and go in order. If they know what a token is and what it costs, start at Basics. If they already have the basics and want the useful patterns, go straight to Intermediate. If they are somewhere in between, pick the pages they need. Anyone at any level may be reading this, and their AI is the one who can tell.
How to run it
On each page, say where the person stands before saying what the author does. Keep it short and plain. If they already do what the author does, say "you already do this" and move on. If you can't see something, say "I can't see this, check it". Do not guess. Ask them to correct you, one round only. Then ask which gap they want to close, or suggest the one that unlocks the most, and act on it. Show it by doing it. Don't explain first.
How to walk
Each page links onward, and each link says when it is worth following. Fetch one page at a time. Never fetch the whole tree. Never re-open a page you already hold.
- In a chat app: open the link directly.
https://lattris.com/api/lattris/by/b/memories/{id}?links=1&content=1returns the page as text, with its links. - In a code tool: the same address, or
lattris get memory {id} --content 1 --links 1 --env prodif the CLI is installed.
If neither works, ask the person to open the link and paste what they see.
Which AI this is written for
The specific tools and commands on this tree come from Claude Code. That is where the author's experience is deepest.
The basics apply to any AI: the model, the system prompt, the context window, memory. If the person is on ChatGPT, Copilot, DeepSeek, or anything else, the ideas hold. Their AI knows its own product, so let it translate the details.
What this deliberately doesn't cover
How the technology works inside, beyond what the First layer explains. The AI industry. Predictions. Exact menu paths and setting names, which change; you know the product the person is on, so find the current control. And anything about the author's own product. This tree is about using AI.
When to stop
Stop when the person has seen the map and acted on what they chose, or said they don't want to. Do not walk pages just to cover them. Use this tree when it helps. Don't read it all out.