1
Set depth
Updated July 10, 2026
Starting AI map
AI Models Made Useful
A visual, step-by-step starting map for choosing the right AI workspace, mode, and prompt for real tasks.
Coffee-first move: bring one real task with you. “Help me answer this customer email” will teach you more than “show me AI.”
Starting AI
Follow the path once, then jump where you need.
This page is meant to be used in order the first time: set your detail level, learn the map, understand the tradeoffs, apply it to a real task, then dig deeper only when it helps.
Bring this sentence: I need AI to help me ___ using context from ___.
Open Advanced AI Onboarding
2
Start the map
Watch how the pieces connect.
Start with the visual walkthrough below. It explains why model, app, mode, prompt, and checking all matter.
Start the map
3
Follow the trail
Let each section hand you to the next.
After the slideshow, continue through limits, why AI helps, the model/app/ecosystem map, and then the picker.
Limits
Why
Map
Picker
4
Jump later
Use shortcuts after the first pass.
Starting AI
See how the pieces fit together
Click through the map first. Each step zooms into one part of the workflow, then the next buttons walk you through the sections in order.
Map move: do not memorize the whole page. Follow the route once, then use the picker for the task in front of you.
What to do
Start with the task before choosing a model.
Use your AI limits wisely
importantStronger models and thinking modes often use more of your plan limits. Use them when the task needs extra reasoning, not for every quick draft.
Limit move: draft fast, then upgrade only if the answer affects money, trust, time, safety, or a decision someone else will act on.
Routine work
Drafts, summaries, simple rewrites, basic emails, quick lists, and low-risk brainstorming.
lower usageHard work
Planning, research synthesis, code architecture, long documents, important decisions, and sensitive question prep.
higher usageSimple rule
If a bad answer would barely matter, use fast mode. If a bad answer would waste time, money, trust, or safety, use stronger mode and verify.
Human-centered why
AI should give you breathing room
Use AI to organize confusing situations, break big tasks into smaller steps, prepare questions, compare options, and get unstuck.
Use AI to check your own blind spots.
A draft can make perfect sense inside your context bubble, then sound totally different to someone seeing it cold. Ask AI to read it like an outsider, then use another model to challenge facts, tone, missing context, and assumptions.
Mental model
Model vs app vs ecosystem
Model = the brain
The underlying AI capability: GPT, Claude, Gemini, Grok, Llama, Mistral, DeepSeek.
App = where you use it
ChatGPT, Claude, Gemini app, Grok, Mistral Vibe, Codex app, Genspark, Perplexity, Microsoft 365 Copilot.
Ecosystem = what it connects to
Apple/iCloud, Microsoft 365, Google Workspace, files, mobile photos, desktop context, X, GitHub, APIs, MCP, and coding tools.
Interactive picker
Find good fits for your task
Start with what you need to do or where the useful context already lives. If an answer feels off, add clearer context, use a stronger mode, or compare with a second model.
Better answer move: add four details to any suggested prompt: who it is for, what context matters, what format you want, and what would make the answer unusable. One specific example usually beats a broad request.
Model families
What each model family is for
These cards are quick mental maps, not full product limits. Each family includes apps, modes, connected tools, agents, and developer paths; use the short label as a starting handle, then expand the card for what it actually does.
Model move: open the card that matches the job, not the brand. The question is “what role do I need right now?”
Deep dives
Three everyday AI ecosystems
Open a card to understand what people use each ecosystem for, why they pick it, and which app surface fits the work in front of them.
Claude ecosystem
Thoughtful writing, long-document work, Artifacts, desktop/tool connections, and Claude Code for developer workflows.
- Why people pick it: polished tone, careful reasoning, long context, and better critique of messy drafts.
- Everyday use: messages, documents, meeting notes, writing polish, plans, and decision framing.
- Workspace use: Artifacts for reusable content/tools; Desktop/connectors when Claude needs surrounding files or apps.
- Developer use: Claude Code for terminal, IDE, GitHub Actions, code review, and repo-aware implementation workflows.
What this ecosystem is: Claude is the chat surface for writing, reasoning, documents, and analysis. Claude Desktop adds local desktop extensions. Connectors can bring in cloud tools. Artifacts turn substantial outputs into editable/reusable content beside the conversation.
Why people pick it: it is often a comfortable choice when the work is sensitive to tone, nuance, or long context: difficult emails, policy drafts, meeting summaries, strategy notes, essays, scripts, and “read this carefully” reviews.
How it is used: paste rough notes, drafts, or files and ask Claude to preserve intent while improving clarity. For bigger outputs, ask for an Artifact so the result can be edited, reused, or shared outside the chat.
Best app surface: Claude web/mobile for ordinary chat and document work. Claude Desktop when the work should connect to local files or desktop apps. Claude Code when the work is a real software project, terminal task, GitHub issue, or PR review.
Try this: “Read this as someone with no background context. What could sound unclear, cold, defensive, too intense, or different than I intend? Then rewrite it in my voice.”
Connector habit: before connecting tools, decide whether the context is local/private, cloud/team-based, or safe to paste manually. The best surface is the one that can see the right context with the least friction.
Advanced angle: Claude Code and Claude Code GitHub Actions fit repo-aware implementation, automated code review, PR/issue workflows, and custom agent workflows that should follow project standards.
ChatGPT + Codex ecosystem
General-purpose AI, files, images, voice, data analysis, research, writing/code blocks, custom assistants, and Codex for coding agents.
- Why people pick it: broad coverage: ask, draft, analyze files, work with images, talk by voice, research, and build.
- Everyday use: planning, learning, writing, spreadsheet/data cleanup, image/screenshot help, and quick second opinions.
- Workspace use: writing/code blocks for co-writing, debugging, and structured editing; file upload and data analysis for documents, PDFs, CSVs, and spreadsheets.
- Developer use: Codex across app, CLI, IDE extension, web/cloud, and GitHub-style review workflows.
What this ecosystem is: ChatGPT is the general assistant surface for conversation, writing, learning, file work, images, voice, search, deep research, writing/code blocks, custom GPTs, and memory. Codex is the coding-agent side of the ecosystem.
Why people pick it: it is the easiest “one place to start” when you are not sure whether the task is writing, research, planning, images, data, coding, or a mix of several things.
How it is used: upload a file to summarize or extract information, ask about a screenshot, talk through a problem by voice, use data analysis for spreadsheets/CSVs, or use writing/code blocks when you want a cleaner draft or implementation snippet inside the chat.
Best app surface: ChatGPT web app for files, research, and longer sessions. Mobile app for voice, photos, screenshots, and quick capture. macOS/Windows app for working beside desktop apps. Codex app / CLI / IDE extension for repos, terminals, multi-file changes, and reviewable diffs.
Try this: “Give me a quick answer first. Then give me the careful versión, what you assumed, and what I should check before using it.”
Codex habit: use ChatGPT when the problem is still fuzzy; use Codex when the work should touch real project files, tests, branches, PRs, or terminal commands.
Advanced angle: Codex is built for supervising coding agents across CLI, IDE, app, web/cloud, and repo workflows, with reviewable diffs and permissioned command execution.
Gemini + Google ecosystem
Google apps, Search, Android, Gmail/Drive/Docs context, photos, YouTube, Workspace, Deep Research, and NotebookLM.
- Why people pick it: the useful context is already in Google: Gmail, Drive, Docs, Sheets, Slides, Calendar, Photos, YouTube, Maps, or Android.
- Everyday use: find details in email/files, summarize Google docs, plan from Calendar/Maps, ask about photos, or understand YouTube content.
- Workspace use: Gemini in Workspace for docs, sheets, slides, drive, mail, meetings; NotebookLM for source-grounded study/research packets.
- Developer use: AI Studio, Gemini API, Android Studio, Vertex AI, and Google Cloud workflows.
What this ecosystem is: Gemini is Google’s AI assistant across the Gemini app, Android, Search, Workspace apps, and developer tools. NotebookLM is the source-grounded research/notebook surface for working from a defined set of materials.
Why people pick it: it reduces copying context around when the work already lives in Google. That can mean Gmail threads, Drive files, Docs, Sheets, Slides, Calendar items, Photos, YouTube videos, Maps plans, or Android screenshots.
How it is used: ask Gemini to summarize email or Drive context, turn a photo or flyer into a calendar/event task, create or improve Workspace docs/slides/sheets, or use NotebookLM to turn sources into briefings, study guides, audio overviews, mind maps, reports, and Q&A.
Best app surface: Gemini app for everyday Google-connected help, Android/mobile for voice/photos/screenshots, Workspace surfaces when you are already inside Gmail/Docs/Sheets/Slides/Drive, Deep Research when you need a report, and NotebookLM when the source packet matters more than open-ended chat.
Try this: “Which Google surface has the useful context for this task: Gmail, Drive, Docs, Sheets, Slides, Calendar, Photos, YouTube, Maps, Search, or NotebookLM? Tell me where to start and what to ask.”
Google habit: decide whether you need personal/app context, Workspace context, web research, or source-grounded NotebookLM work. Those are different jobs.
Advanced angle: Gemini fits AI Studio, Gemini API, Vertex AI, Android Studio, multimodal workflows, and Google Cloud/Workspace automation concepts.
Workspaces and maps
Packaged workflows and strength map
Workspace move: pick the workspace that already has the useful context. Copilot for Microsoft work, NotebookLM for source packets, Codex for project files.
Genspark
AI workspace for packaged outputs like slides, docs, research-style deliverables, creative assets, marketing workflows, and content systems.
Perplexity
Search and answer assistant with sources. Useful for starting research and checking what appears current.
Microsoft Copilot
Microsoft’s AI family: free Copilot for web help, Copilot Chat Basic for secure work chat, and Microsoft 365 Copilot for deeper Office, Teams, Outlook, and company context.
Free/personal Copilot: use it for web-grounded questions, writing, brainstorming, image creation, Edge page help, and everyday learning. The visible modes can include Quick response, Think Deeper, Study and learn, Smart, and Search.
Copilot Chat Basic for work: use it for secure work chat with web grounding, file/image upload, Copilot Pages, standard model access, and the model selector. It can work with uploaded files or context open in supported Microsoft 365 apps.
Microsoft 365 Copilot: use it when the answer needs meetings, emails, chats, people, files, or app-specific editing inside Word, Excel, PowerPoint, Outlook, Teams, OneNote, OneDrive, or SharePoint.
Model picker: Auto is the everyday default. Quick response is for speed. Think deeper is for harder reasoning. If your tenant shows named GPT-family options, treat them as current Microsoft-controlled options that may change.
Prompt pattern: “Using the Microsoft 365 context I can access or attach, summarize what matters, list assumptions, draft the output, and tell me what names, dates, files, numbers, or permissions I should verify.”
Watch for: check your label: Copilot Chat (Basic), M365 Copilot (Basic), or M365 Copilot (Premium). That label affects app access, work-data grounding, agents, and priority access.
Useful second check: paste the draft into another model and ask it to read with zero company context. Ask what sounds unclear, risky, overconfident, or missing.
Common strength map
Notes and knowledge workspaces
Turn scattered context into better AI answers
Notetaking is how you capture raw context. Knowledge workspaces are where that context becomes reusable: project briefs, meeting notes, research sources, decisions, and prompts you can bring back to a model later.
Notes move: after a useful AI answer, save the context and final decision. Next time, paste the note instead of re-explaining everything.
A practical workflow for notes + AI
The point is not to collect endless notes. It is to keep enough organized context that an AI can help without guessing.
Capture
Organize
Ask
Compare
Reuse
Capture: save messy notes, links, screenshots, transcripts, files, and decisions before they disappear.
Organize: turn raw notes into headings, bullets, dates, owners, source links, and open questions.
Ask: give the model the note plus a specific job: summarize, draft, compare, plan, or critique.
Compare: use another model or source-backed tool when tone, facts, or decisions matter.
Reuse: keep the cleaned note as future context instead of restarting from scratch every time.
Intermediate habit: make one reusable project brief per important project: goal, audience, decisions, constraints, open questions, source links, and next actions.
Advanced habit: keep AI-ready context files in Markdown so they can move between Obsidian, local folders, Codex, GitHub issues, Claude, ChatGPT, Gemini, and other tools without lock-in.
Use AI to teach yourself
Turn AI into a coach, not just an answer machine
The best learning use of AI is active: explain, practice, get feedback, apply it to something real, then save what you learned in notes.
Learning move: make the model quiz you before it explains more. If you cannot use the idea, you have not learned it yet.
The learning loop
If you only ask “explain this,” it is easy to feel fluent without being able to use the idea. Ask the AI to make you practice.
Explain
Example
Practice
Feedback
Apply
Explain: ask for the idea at your level, with no jargon unless it is defined.
Example: ask for a real example from your work, hobby, class, or project.
Practice: make the AI quiz you, give drills, or ask you to solve a small scenario.
Feedback: have the AI grade your answer and show what you missed.
Apply: use the idea on a real task, then save the result in your notes.
Intermediate habit: ask for spaced review: “quiz me tomorrow on the parts I missed today,” then save the weak spots in a note.
Advanced habit: build a source-backed study packet in NotebookLM or Markdown, then use Claude/ChatGPT/Gemini for practice, critique, and application projects.
AI judgment and blind spots
Use AI without letting it become an echo chamber
AI is designed to be helpful, but helpful can become too agreeable. A strong workflow asks models to challenge your framing, not just polish it.
Challenge move: after the first good answer, ask “what would a smart person disagree with here?” That is where the useful thinking starts.
The challenge loop
When the answer matters, treat the first response as a draft. Make the model name assumptions, critique weak points, read it from another perspective, and then improve it.
Frame
Challenge
Compare
Verify
Decide
Frame: state your goal, what you believe, and what outcome you want.
Challenge: ask what you may be missing, over-weighting, or assuming.
Compare: run the same issue through another model or perspective.
Verify: check facts, tone, numbers, policies, source material, and real-world constraints.
Decide: use the critique to improve the work, not to outsource judgment.
Intermediate habit: after a model agrees with you, ask the opposite: “make the strongest case that I am wrong or missing something.”
Advanced habit: use different models for different jobs: one to draft, one to critique, one to fact-check, and one to simplify the final decision.
Comparison
Useful comparison table
| Family | Common strength | Mode to try | Useful surfaces | Watch out for |
|---|
Prompt engineering without jargon
Prompt stack: task + context + constraints + format + freshness + check
Prompt move: do not chase perfect wording first. Add missing context, ask for a useful format, then ask the model what it still needs.
Prompt stack visual
Task
Context
Constraints
Format
Freshness
Check
Use the stack as a checklist when a prompt is vague: say what you want, add useful context, set limits, ask for the format, flag whether current info matters, then ask what should be checked.
Prompt repair tool
This builder uses prompt templates, not live AI. It does not save the text you type.
Prompt + freshness builder
Follow-up prompt library
Second opinion prompt: Review this answer from another AI. Do not rewrite it yet. Tell me what is strong, weak, unsupported, outdated, confusing, or missing. Then suggest the top 3 improvements.
Prompt technique cards
Take it with you
Download the prompt cheat sheet
Includes the mode map, context map, prompt stack, freshness prompt, accuracy habits, starter prompts, and second-opinion prompts.
Carry-it move: keep the cheat sheet nearby, then customize one prompt for your actual work instead of starting from a blank box.