Use AI Confidently
Six modules. Self-paced. Start with your first conversation and finish using AI with confidence — and a clear sense of where not to trust it.
Who this is for
Anyone new to AI or who wants to be more intentional about it. No prerequisites.
Each module is designed as an entry point and takes 5–10 minutes to get started — take it as far as you want after that.
What you’ll be able to do
- Move between AI models and see how the same prompt lands differently
- Use AI to get better at AI — have it ask questions, write your prompts, and review your sessions
- Tell what an AI can do from its tools — strong with documents, unreliable with spreadsheets
- Get AI to sharpen your thinking instead of generating from scratch
- Recognize AI's pull toward validation — and why asking it to 'be honest' won't fix it
- Save the prompts you reuse and run them from a personal library — one click or a / command away
Module 1: First Contact Log in, run the same prompt across several models, then meet Compass.
Log in and look around
Two short walkthroughs get you onto the platform and oriented. Watch these first:
A few words you'll see throughout:
A model is an AI system — Claude, GPT, Gemini, Mistral, and others. Each comes from a different company and handles things differently. You choose which model runs your conversation.
A prompt is what you type to the AI — a question, an instruction, or a rough draft you want help with.
An agent is a model that's been pre-configured for a specific task — given instructions, scope, and sometimes reference material. Agents live in the Agent Center.
Done when:
- You've logged into the platform
- You've watched both walkthroughs
Same prompt, several models
Start a new chat. The platform gives you access to several AI models, and you can switch between them while keeping the same conversation going.
Open the model selector at the top of the chat. Models are grouped by the company that made them — Anthropic (Claude), OpenAI (GPT), and others. Pick one and paste:
Now switch the model and send the exact same prompt again. Work through a few:
- Claude Haiku and GPT-5-mini — the smaller, faster models
- Claude Sonnet, then Claude Opus — larger Claude models: richer answers, but slower — save them for when a task needs real depth
- GPT-5 — OpenAI's counterpart to Claude
Same question, different systems. Read for what changed.
Take it further.
- Swap in a real task from your work this week and run it across the same models
- Ask one model: "What do you know about this platform?" — notice it knows nothing about where it's running. (More on why in Module 3.)
Done when:
- You've sent the same prompt to at least three different models
- You can name one difference between how two of them responded
Start with Compass
So far you've been driving a plain chat and picking models yourself. Compass is different — it's an agent, built for one job: look at your actual work, find something AI can genuinely help with, and do it with you.
Open the Agent Center — the small icon near the top left — and find Compass, or go straight to it: open Compass. Then tell it what your work looks like:
I spend a lot of my week on [name a few tasks that eat up your time]. I'm not sure where AI fits in — can you help me figure that out?
Because Compass is an agent, it comes pre-configured — it runs the way it was set up, so just have the conversation.
Done when:
- You've opened the Agent Center and found Compass
- You've had a conversation with Compass about your real work
Module 2: Let AI Help You Use AI Three moves that use AI to sharpen how you work with AI.
The fastest way to get better at AI is to let AI help you. These three moves work on top of whatever you're already doing — make them habits and every task gets easier.
Have it ask before it acts
Most people write a long prompt and hope for the best. A more reliable habit: before AI does the work, have it ask you what it needs to know. Make this your default opening move on almost any task — an email, a plan, a summary, a decision.
Pick a real task and paste:
I need to [TASK]. Before you start, ask me the questions you'd need answered to do this really well.
Answer what it asks, then let it produce a draft.
Take it further.
- Run the same task two ways — straight, then with the questions first. Compare.
- Try it on something you've been putting off; the questions often unstick it.
Done when:
- You've asked AI to question you before starting a real task
- You've answered and gotten a draft back
Ask AI to write the prompt
You don't have to be good at writing prompts — AI is. Instead of crafting the perfect instruction yourself, describe the result you want and let AI write the prompt for you.
I want to [DESCRIBE WHAT YOU'RE TRYING TO ACCOMPLISH]. Write the prompt I should use to get the best result from an AI, then explain why you wrote it that way.
Take the prompt it gives you and run it — paste it into a new chat and see what comes back.
Done when:
- You've had AI write a prompt for you
- You've run that prompt and seen the result
Ask AI to review the session
After a long back-and-forth, ask the AI to look at how the two of you worked together and tell you how to do it better next time. This turns any session into practice.
At the end of a working conversation, paste:
Look back at this whole conversation. How could I have worked with you more effectively? What could I have told you earlier, asked differently, or done to get a better result faster?
Done when:
- You've asked AI to evaluate how you worked together
- You've gotten at least one concrete idea for next time
Module 3: What AI Can Actually Do An AI's power comes from its tools — and so do its limits. Find both edges.
A model on its own only generates text. What makes it useful on the platform are the tools wired to it — web search, reading documents, generating images. Change the tools and you change what the same model can do. This module is about noticing that, and finding where AI is strong and where it isn't.
Same model, no tools
We've set up an agent that's a capable model — Claude Sonnet — with every tool removed. Open it: Sonnet, no tools. Paste these one at a time and watch what it does:
It can't check the weather — there's no tool for it. It answers the Super Bowl from memory, and may be out of date without realizing it. It may invent plausible-sounding details about your organization. And that last question makes it explain the reason itself: it was trained up to a fixed date and has no live view of the world.
Done when:
- You've sent a no-tools model a handful of real questions
- You've watched it stall, go stale, or make something up
Same model, with tools
Now open a regular chat — these come with tools — and run the same prompts again.
This time the answers can change: the weather and the Super Bowl can come back current, and it can actually look your organization up, because a regular chat can reach a web-search tool. Same model, very different results.
The takeaway: what an AI can do isn't the model alone — it's the model plus its tools, and you can always just ask what it has. (One caveat: models sometimes misreport their own tools, so the real test is trying.)
Done when:
- You've run the same prompts with tools and seen the answers change
- You can say in one sentence why the two sets of answers differed
Where it's strong: documents
Text is where AI is most reliable. Upload a real work document — an email thread, an SOP, meeting notes, a report — and paste:
I've uploaded a document. Do three things: 1. Summarize it in three paragraphs for someone who hasn't read it. 2. Tell me what questions a reader would still have. 3. Turn it into a different format — FAQ, one-pager, action list, or plain-language version.
Done when:
- You've uploaded a document and transformed it three ways
A trick: leave your notes in the document
You don't have to type every instruction into the prompt. Add a few comments to a Word document — the kind you'd leave for a colleague — then upload it using the "Upload to Provider" option. That option preserves your comments (it won't carry track-changes edits). Now AI can read your margin notes directly. Paste:
I've uploaded a document with comments. Work through each comment in order — make the change it asks for, and tell me what you did.
Done when:
- You've uploaded a commented .docx with "Upload to Provider"
- You've had AI act on comments left in the document
Where it struggles: spreadsheets
Upload a spreadsheet and ask for analysis:
I've uploaded a spreadsheet. Analyze the data and tell me the three most important things it shows.
Read the answer closely. The model will often narrate about a spreadsheet — guessing what the columns mean — instead of actually computing on the numbers, and it can state totals or trends that simply aren't there. This is the same lesson as the no-tools agent above: on this platform the model has no analytics tool wired in, so it has nothing to do real math with — it's reading the file like text rather than crunching it. (We're working on adding an analytics tool.) Until then, when the numbers matter, verify every one.
Done when:
- You've asked AI to analyze a spreadsheet
- You've spotted where it narrated instead of computed, or got a number wrong
Module 4: Structuring Rough Thinking Give AI your rough notes. Let it structure and draw out what's implicit.
Paste your notes
Most AI advice says "describe what you want and let AI write it." That produces generic output because AI is guessing. What actually works: give AI the rough thinking you already have — bullets, fragments, half-sentences — and ask it to organize. AI is much better at structuring than generating.
Find something you're working on where you have ideas but haven't organized them. Meeting notes, a proposal outline, talking points, a project plan. Messier is better. Paste:
Here are my rough notes on [TOPIC]. They're messy — bullets, fragments, unfinished thoughts. Don't invent new ideas. Work only with what I've given you. 1. Organize these into a clear structure 2. Fill in transitions between points 3. Clean up the language but keep my voice 4. Flag anything that seems incomplete Notes: [PASTE NOTES]
Done when:
- You've given AI rough notes on something real
- You've gotten back a structured version
Ask it what's unclear
Once you have a structured draft, let AI clarify the request itself. Ask it what would make the work better — it will surface gaps and assumptions you can fill in.
In the same conversation, paste:
Answer what it asks. Then:
Done when:
- You've answered AI's questions about your draft
- You've gotten back a revised version that reflects your actual thinking
Module 5: Pushing Past Validation AI validates by default. You can turn up the criticism — but you can't make it "honest."
Get validated
AI defaults to encouragement. It will refine your ideas, suggest enhancements, and frame things positively. Seeing this directly is the first step.
Pick an idea you're genuinely unsure about — a proposal, a process change, an approach to a problem. Paste:
I'm considering this idea for my work: [DESCRIBE IDEA]. What do you think? Is this a good approach? What would make it work well?
Read the response.
Done when:
- You've seen AI respond supportively to an idea you're uncertain about
Ask for critique
The same model, same idea, same conversation can produce entirely different feedback depending on what you ask for. Forcing critique is a skill worth having.
In the same conversation:
Now be my toughest critic. What's wrong with this idea? Who wouldn't it work for? What am I not seeing? What would someone who thinks this is a bad idea say — and would they be right? Don't be encouraging. Be direct.
Done when:
- You've gotten AI to generate concrete criticism of the same idea it just validated
Why you can't ask it to "be honest"
The obvious next move is to type "now just be honest with me." It won't do what you hope. Honesty is a stance a person takes — and AI doesn't have one. Whatever you ask, it's still predicting the response it thinks you want. It can't step outside that to tell you the truth, because there's no honest version of it underneath.
What it can do is shift between praise and criticism, and hold both at once when you ask. That's a dial you're setting, not a confession you're getting. So don't ask it to be honest — ask it to weigh both sides and leave the verdict to you:
Weigh the praise and the criticism together. Which concerns would most change whether this idea works, and which matter less? Lay out the trade-offs so I can decide.
The judgment stays yours. AI can lay out considerations from both directions — it can't tell you which one is true.
Done when:
- You've seen AI balance praise and criticism in one response
- You can explain why asking AI to "be honest" doesn't do what it sounds like
Module 6: Save What Works When you keep asking AI the same thing, stop retyping it. Save it, and build a small library of prompts you reuse.
By now you've written prompts worth keeping — the ones that ask AI to question you first, to critique instead of validate, to structure your rough notes. Re-typing those every time is wasted effort. The platform lets you save a prompt once and reuse it across every conversation. This module gets you a starter library and shows you how to run from it.
Save your first prompt
Open Prompts in the left menu. Click the plus button to create a new prompt. Give it a name, then paste this one in and save it:
I want to learn about [TOPIC]. Can you design me a short interactive minicourse we can go through together in this session that will cover an introduction to basic concepts that I might get in a college level course over the course of a semester? I obviously can't cover an entire course in one session, but I want to hit the main concepts so I have a good overview and know where I might want to dig deeper. I like interactive learning, so I want you to link me to websites where I can see things and maybe do things. I don't want to read long blocks of text. If I want to pursue something further then we can go deeper together. I want you to give me a high level overview of the minicourse and plan at the beginning and help me dig in topic by topic. Ask me questions that will help you understand what I am looking for better and design a good learning experience for me.
That's a prompt you'll want again and again — every time there's a subject you'd like a fast, guided way into. Saving it means it's one click away the next time, instead of something you half-remember and rewrite.
Done when:
- You've opened Prompts and created a new saved prompt
- You've saved the minicourse prompt to your library
Run it from your library
To use a saved prompt, go to your saved prompts and click it — that sends it straight into the chat. Run the minicourse prompt now: click it, fill in a [TOPIC] you're curious about, and send it.
Pro tip. When you create a prompt, you can give it a command. Once it has one, you can invoke it right from the chat box by typing / and the command name — no trip to the Prompts menu needed.
Done when:
- You've run a saved prompt by clicking it into the chat
- You've added a command to a prompt and invoked it with /
Write one worth keeping
The best prompts to save are the ones you'd otherwise repeat. In Module 5 you saw that AI defaults to validation, and that asking it to "be honest" doesn't fix it — what helps is giving it concrete behavioral instructions, every time. An anti-sycophancy prompt like that is worth writing once and saving. So write your own.
Open a chat, pick a strong model (Sonnet or Opus), and paste this:
Research good anti-sycophancy prompts and generate one for me. Use a snippet so I can easily copy the prompt.
The prompt above is not the one you save — it's the tool that builds it. What you save is the model's answer: the anti-sycophancy prompt it writes for you. Take that result, save it to your library, and give it a command so it's one / away whenever you want it.
Done when:
- You've had a strong model write you an anti-sycophancy prompt
- You've saved that result as a reusable prompt with a command
Use saved prompts whenever you notice you keep asking AI to do the same thing across sessions. That instinct — "I want this behavior every time" — is also the seam between using AI and building with it: a saved prompt is the first step toward an agent, which is just saved context and instructions someone else can run too. That's where the Build path picks up.
Ready to build?
You've learned how to use AI — and where not to trust it. The Build path takes you further: you'll design and build AI agents that other people can use.
Opens in Week 4Or explore: Create — Code with AI (Week 6) | All paths →