Almost every AI tool does one of six jobs: generate text, generate images, work with audio, work with video, work with structured data, or connect other tools together.
Do now: Take one task from this week and write which of the six capabilities it actually needs.
Choose the job first. The tool is the last decision, not the first.Tool names change every few months. Capabilities change slowly. If you learn the map, a new tool is just a new entry in a category you already understand — and you stop paying for things that do a job you did not have.
Six capabilities, not six hundred products
Text tools generate, rewrite, summarise, classify and extract. This is the largest category and the one most people meet first. Image tools generate pictures from descriptions, edit existing pictures, or remove and replace parts of them. Audio tools transcribe speech into text, generate speech from text, translate, and clean up recordings.
Video tools generate clips, cut footage, add captions, and increasingly assemble rough edits. Data tools read spreadsheets and documents, explain what is in them, find patterns, and produce charts. Automation tools do not generate anything themselves — they connect the others together and make things happen when a trigger fires.
A product may combine several. A single assistant might handle text, read a spreadsheet and describe an image. That does not create a seventh category; it means one product covers three. Describing tools this way lets you compare things that market themselves as completely different.
If you cannot say which of the six a tool does, you are looking at marketing, not at a capability.
Start from the task, not the product
Write the task as a sentence with a verb and an output. 'Turn Monday's meeting recording into a list of decisions with owners' names a capability chain: audio to text, then text to structure. 'Make my shop look professional' names nothing, which is why it leads to buying five tools and finishing none.
Many real tasks need two or three capabilities in sequence. That sequence is your workflow, and it is worth writing down before you shop, because the handover between capabilities is where quality is usually lost. A perfect transcript feeding a careless summary produces a careless result.
It is also where you discover the task does not need AI at all. A recurring report that has the same shape every month may need a template and fifteen minutes, not a subscription.
What the map does not tell you
Capability is not suitability. Two tools in the same category can differ enormously in accuracy for your language, your accent, your file sizes, your country and your budget. The map narrows the field to three candidates; it does not pick the winner.
Nor does capability imply permission. A tool that can generate a voice does not tell you whether you may clone that particular voice. A tool that can read a spreadsheet does not tell you whether that spreadsheet is yours to upload. Those are separate questions, and they are the ones that get people into trouble.
The map is a filter, not a verdict. It saves you from the wrong category, not from the wrong choice inside it.