The Landscape of Generative AI Tools

Generative AI is not a single product. It is a whole ecosystem of tools that produce text, images, sound, video and code, from free assistants to professional services. In this lesson you will get a map of that landscape and, more importantly, a method for choosing the right tool for your needs without getting lost in the hype.

Now that you know how a language model produces text, you are ready to look at the different families of tools built on this kind of technology. We deliberately avoid naming products: they change quickly, and the criteria you will learn stay valid whichever tool you use.

The main families of tools

Conversational assistants (text)

These are the best-known tools: you type in a chat window and the assistant answers. They handle writing, summarizing, translating, brainstorming, explaining, and a lot more. They are the focus of this course, because they are the most versatile and the ones where prompting skills pay off immediately.

Image generation

You describe an image in words (a "prompt") and the tool creates it. Some tools can also edit an existing image: remove an object, change the background, extend a picture. Prompting here means describing the subject, style, lighting, composition and mood. The same principles apply (be specific, iterate), but the vocabulary is visual.

Audio and speech

This family includes speech-to-text (automatic transcription of meetings or interviews), text-to-speech (synthetic voices reading a text), and music or sound generation. Voice cloning also exists, and it raises serious questions about consent and misuse, which we cover in chapter 4.

Video

Some tools generate short video clips from a description, or animate still images. This area is evolving quickly and is often more demanding in computing power, so access may be limited or paid.

Code assistants

These tools suggest code as you type, explain existing code, find bugs or generate tests. Even if you are not a developer, they can help you write a spreadsheet formula, a small script or a regular expression. We will look at this use in chapter 5.

Multimodal assistants

A multimodal model can handle several types of content at once: for example, you can send a photo of a handwritten note or a chart and ask a question about it, or ask for a description of an image, in the same conversation. Many modern assistants combine text, images, voice and file reading in one interface.

Family	Input	Output	Typical use
Chat assistant	Text (sometimes files)	Text	Writing, summaries, explanations
Image generator	Text, sometimes an image	Image	Illustrations, mock-ups, ideas
Speech tools	Audio or text	Text or audio	Transcripts, narration
Code assistant	Code and text	Code, explanations	Formulas, scripts, debugging
Multimodal assistant	Text, images, audio, files	Text (and sometimes more)	Analyzing a chart, a photo, a document

Open models versus proprietary models

You will often hear about "open" and "proprietary" models. The difference concerns how the model is distributed.

• Proprietary models are accessed through a company's service (a website, an app, an interface for developers). You send your text to their servers and get a response. The inner workings are not published. They are convenient and often very capable, but you depend on the provider's terms, prices and policies.
• Open models (more precisely, "open-weight" in many cases) publish the trained parameters, with licenses that vary. Technically skilled people can download and run them on their own machines or servers, adapt them, and keep data in-house. They require more setup and hardware, and "open" does not always mean the same thing: licenses differ in what they allow, so read them.

For a beginner, a hosted assistant is usually the simplest way to start. The open-versus-proprietary question becomes important when you care about control over your data, customization, or cost at scale.

Free or paid?

Most tools offer a free tier and one or more paid plans. The differences typically involve:

• Usage limits: number of messages, images or files per period.
• Model capabilities: access to more capable models, longer context, extra features such as file analysis or web browsing.
• Priority and speed: paying users may get faster responses at busy times.
• Data handling and business features: some professional or organizational plans include stronger privacy guarantees, administration controls or an agreement that your content is not used to improve the service. Do not assume this: read the terms.

"Free" rarely means "no cost." You may be paying with your data or with limits. This does not make free tools bad; it means you should know what you are agreeing to.

How to choose a tool: six criteria

Instead of asking "what is the best AI?", ask "what is the best tool for this task?" Use this checklist.

1. The task. Do you need text, an image, a transcription, code? Does the tool handle your language well?
2. Quality and reliability. Try it on a real task you can judge. Does it follow instructions? Does it admit uncertainty?
3. Data and privacy. What happens to what you type? Is there a setting to limit retention or training use? Is the tool allowed by your school or employer?
4. Features. Can it read files, browse the web, remember preferences, export results?
5. Cost and limits. Is the free tier enough for your volume? What would you pay for, and is it worth it?
6. Ease and context. Does it fit where you already work (browser, phone, office suite)? Is the interface clear for you?

Settings and data: why two tools behave differently

Two assistants can answer the same prompt very differently, even if they use similar technology. Several factors explain this:

• The underlying model, with different sizes, training data and tuning.
• Hidden instructions from the provider (a "system prompt") that set tone, safety rules and style before you type anything.
• Settings such as temperature, maximum length and available tools (web search, file reading).
• The data the tool can access: only its training, or also the web, or your files.

This is why a prompt that works well in one tool may need small adjustments in another. It is also why you should treat your prompts as reusable recipes that you adapt, not magic spells.

Your turn: choose your tool

Exercise. Take the three tasks you listed in your prompt notebook in lesson 1. For each, write down: (a) the type of tool needed, (b) the sensitivity of the data involved (public, personal, confidential), and (c) one criterion from the list above that matters most. Then choose the tool you will use for your project and note why.

See a solution

Example for three tasks:

• Rewrite rough emails: chat assistant; data may include names, so anonymize; key criterion: privacy settings and tone control.
• Summarize meeting notes: chat assistant that can read pasted text or files; data may be confidential, so check what the employer allows; key criterion: data handling.
• Plan a weekly study schedule: chat assistant; no sensitive data; key criterion: quality of structured output (tables).

Choice: "A general chat assistant with a free tier, with history and training use turned off where the setting exists. I will switch to a work-approved tool for anything involving colleagues or clients." Notice that the decision rests on the criteria, not on a brand.