Three tiers. Real skills. No prerequisites. Each lesson takes 5–10 minutes, with practical exercises you can try immediately.
5 lessons · ≈ 30 min
0 / 5 complete
6 lessons · ≈ 45 min
0 / 6 complete
5 lessons · ≈ 45 min
0 / 5 complete
What AI actually is, how chatbots work, and your first real conversation with one.
Forget the sci-fi version. AI in 2025 is software that predicts what comes next — in text, images, or code. Think autocomplete, but dramatically more powerful. Large Language Models (LLMs) like GPT-4 and Claude are trained on billions of text examples so they can generate helpful, human-like responses. They don't "think" — they pattern-match at an extraordinary scale.
Open ChatGPT, Claude, or Gemini and type: "Explain what you are in 3 sentences, as if I'm a university student who's never used AI before." Read the response. Then try: "Now explain it in one sentence." Congratulations — you've just done your first prompt iteration. You gave feedback to an AI, and it adjusted. This back-and-forth is the foundation of everything that follows.
These are the three AI chatbots you should know. ChatGPT (by OpenAI) is the most popular — great for general tasks, writing, and brainstorming. Claude (by Anthropic) excels at nuanced, thoughtful responses and handling long documents. Gemini (by Google) is tightly integrated with Google's ecosystem and strong at research. Try the same question in all three — you'll notice they each have a different "personality."
AI is great at: drafting emails, summarising articles, brainstorming ideas, explaining concepts, translating languages, and writing code. AI struggles with: real-time information (it has a knowledge cutoff), maths requiring precision, citing real sources accurately, and understanding your specific personal context. The golden rule: always verify important facts. Use AI as a first draft, not a final answer.
Before you go further, three essential principles. 1) Don't share sensitive data — anything you type into a chatbot could be used for training. Never paste passwords, private documents, or confidential work data. 2) AI can be biased — models learn from human text, which includes human prejudices. Stay critical. 3) Give credit — if AI helps you write something, be transparent about it, especially in academic or professional contexts.
Let's see what stuck. Pick the best answer.
What is the best way to think about what an LLM does?
Prompt engineering, research workflows, and getting 10x more value from AI.
The quality of your output depends almost entirely on the quality of your input. Here's the framework: Role ("You are an experienced copywriter"), Task ("Write a LinkedIn post about…"), Context ("My audience is UK marketing managers"), Format ("Keep it under 150 words, use bullet points"). Try adding each of these layers and watch how the output improves with each one.
When you ask AI to solve something complex, add: "Think step by step." This single phrase dramatically improves accuracy on reasoning tasks. Why? It forces the model to break a problem into stages instead of jumping to a conclusion. Try it with a maths problem, a business decision, or a tricky essay question. Compare the result with and without — the difference is striking.
Google's NotebookLM lets you upload PDFs, articles, and notes — then chat with them. It's like having a research assistant that has read everything you've ever highlighted. Upload a lecture slide deck and ask: "Summarise the key arguments" or "Create flashcards from this material." Unlike general chatbots, NotebookLM only uses your sources, which means no hallucinated facts.
AI is not the best writer — but it's the best writing partner. Use it to: brainstorm (give me 10 angles for an essay on X), outline (structure a 2,000-word report on Y), draft (write the introduction in a formal academic tone), and refine (make this paragraph more concise). The trick is to stay in control — edit everything, add your voice, and never submit AI output as-is.
By now you've used a few tools. Time to get strategic. Choose one primary chatbot (the one whose tone and speed you prefer), one research tool (NotebookLM or Perplexity), and one writing tool (the chatbot, or a dedicated tool like Grammarly with AI). Stick with these for two weeks. Mastery comes from depth, not from jumping between 20 tools.
You will sometimes get bad outputs (hallucinations, ignored instructions). Don't just regenerate—debug it. Isolate the error by telling the AI what went wrong ("You ignored the word count"). Challenge it ("Are you sure about that statistic?"). If it's still failing, simplify the prompt or try breaking the task into smaller steps.
Test your prompt engineering knowledge.
In the Role-Task-Context-Format framework, what does "Context" refer to?
AI coding assistants, automated workflows, and building with AI agents.
Claude Code lets you have a conversation with AI that can read, write, and edit code across your entire project. Think of it as a senior developer pair-programming with you in your terminal. Cursor takes a similar approach but lives inside a code editor with inline suggestions. You don't need to be a developer to benefit — these tools can help you automate spreadsheets, build simple apps, or understand code you've inherited.
Antigravity is a new breed of AI tool — an intelligent workspace where AI doesn't just answer questions, it actively helps you build. Upload documents, paste research, describe what you want to create, and Antigravity helps you structure, write, and iterate — all in one place. It's designed for people who want AI to be a true collaborator, not just a chat window. Think of it as the difference between texting a friend and actually working alongside them.
An AI agent is an AI that doesn't just respond — it takes actions. It can browse the web, run code, manage files, and complete multi-step tasks autonomously. Tools like Claude Code's agentic mode and Antigravity already let you experience this. Instead of giving AI one question at a time, you give it a goal: "Research competitor pricing, create a comparison table, and draft an email summary." The AI plans the steps and executes them.
You've learned the tools — now put them together. Here's a real workflow: Morning research — use Perplexity or NotebookLM to catch up on a topic. Deep work — use Claude or ChatGPT to brainstorm, outline, and draft. Code/build — use Claude Code or Antigravity for technical tasks. Review — use AI to proofread, check logic, and format. The key is building habits, not just knowing tools. Start with one AI touchpoint per day and expand from there.
"Shadow AI" refers to employees using unsanctioned AI tools at work. The biggest risk? Data leakage. If you paste proprietary code, client data, or financial reports into a public consumer model (like the free tier of ChatGPT), that data can be used to train future models. Always check corporate policies and verify if a tool trains on user data before using it for work. Paid enterprise tiers usually guarantee zero data retention.
Final knowledge check.
What makes an AI "agent" different from a regular chatbot?
Head to our prompt templates and start using AI right now.