A friendly map to artificial intelligence

AI, explained in one page

Think of this as a coffee with a well-read friend: what AI is, why and when to reach for it, how to use it well — plus the words people throw around, the worries worth having, and which tool fits which job.

What · Why · When · How

What is AI?

Software that learns patterns from large amounts of data to perform tasks that usually need human judgement — writing, summarising, answering, generating images or code. Today’s most-used systems are generative AI, built on large language models.

Why use it?

To remove friction from thinking work: a first draft, a summary of something long, a second pair of eyes, or a quick prototype. It widens what one person can do — it doesn’t replace your judgement.

When to use it

When the task is open-ended, repetitive, or a starting point — brainstorming, restructuring, explaining, translating, drafting. Skip it when you need guaranteed accuracy without checking, or you’re handling sensitive data.

How to use it well

Give clear context and examples, ask for the format you want, and iterate. Treat the output as a draft from a fast, confident intern: useful, but always verify facts and sources yourself.

The lingo, decoded

LLM large language model

The engine inside tools like ChatGPT, Claude and Gemini. It’s trained to predict the most likely next piece of text, which lets it write and reason convincingly. Key point: it predicts plausible language — not necessarily true language.

Agent AI that takes actions

An AI that doesn’t just answer but does — it plans steps, uses tools, browses, writes files or builds an app, then checks its own work. “Agentic” tools complete multi-step tasks with less hand-holding.

Vibe coding build by describing

Coined by AI researcher Andrej Karpathy in February 2025 — building software by describing what you want in plain language and letting AI write the code. Named Collins Dictionary’s Word of the Year 2025. Brilliant for prototypes; production code still needs human review.

Common confusions & honest concerns

Common misunderstandings

“It understands like a person.” It predicts language patterns; it has no lived understanding.
“It’s always right.” It can hallucinate — state false things with total confidence.
“Vibe coding replaces engineers.” Great for drafts and MVPs; real products still need review.
“Newer is always better.” Fit-for-task beats novelty.

Valid concerns to hold

Accuracy: verify facts and citations before relying on them.
Privacy: don’t paste confidential or personal data.
Copyright & IP: generated work can raise rights questions.
Bias: outputs reflect the models’ training data.
Over-reliance: keep your own skills and judgement sharp.

The most recognised AI tools

ChatGPT OpenAI’s general assistant for writing, brainstorming and Q&A.

Claude Anthropic’s assistant, strong at writing, analysis and long documents.

Gemini Google’s multimodal assistant, woven into Search and Workspace.

Microsoft Copilot AI built into Windows and Microsoft 365 apps.

Perplexity An answer engine that searches the web and cites its sources.

NotebookLM Research assistant grounded in documents you upload.

Lovable Builds full web apps from plain-language prompts.

Replit Build, run and deploy apps from the browser, with an AI agent.

Midjourney Generates images from text descriptions.

Features and pricing change often — confirm details on each tool’s own site before relying on them.

Where to keep learning

Podcast — How I AI, hosted by Claire Vo. Each episode shows one practical way someone uses AI in real work.

One Useful Thing — evidence-led writing on using AI well.
Google AI for Developers — official docs and AI Studio.
Anthropic’s prompting guide — practical tips for better prompts.
NotebookLM explained — an overview of source-grounded AI.