Skip to main content
The Human Bit

Lesson 01 · 8 minutes

What AI is, and what it is not

Tell apart the model, the product built around it, plain search, rule-based automation and ordinary software, so you can reach for the simplest thing that will actually work.

Recognise which kind of technology is actually involved in a task.

Why this matters

People often pay for an AI product when a rule or a feature they already have would be more reliable.

Once you can see the layers, claims about price, privacy and capability get much easier to judge.

The practical bit

A model is the part that generates or interprets content by predicting patterns. A product is what wraps a model with the things you touch, such as file uploads, search, memory and sharing. An API is a way for other software to call the same capability quietly in the background.

Alongside those sit older, plainer tools. Search looks information up. Automation follows fixed rules. Ordinary software stores, calculates and moves data predictably. Most products now mix several of these at once, so the word AI on the label tells you almost nothing about how a given feature actually works.

What it looks like

At work

If your invoices already arrive in fixed fields, accounting software with validation rules will usually beat asking a general model to reinterpret every invoice from scratch. The boring tool is the dependable one here.

In everyday life

A calendar reminder is plain automation. Asking a model to write a warm birthday message is generative help, meaning new content the model produces rather than looks up. Organising one birthday can use both, for different parts of the job.

Common mistake

Calling every handy feature AI, then comparing products by how famous their model is instead of by the whole workflow you actually need to finish.

Use your judgement

Does this task require interpretation, or would clear rules and existing software handle it?

Try it now

Name the actual intervention
  1. Pick one digital task you do regularly.
  2. Classify each step as human judgement, software, search, automation or generative AI.
  3. Circle the step that genuinely needs interpretation or generation.
Check: If every step is predictable, test ordinary software or automation before adding AI.

Use this in a real situation

Improve a familiar updateUse a practical task to separate pattern work from human judgement.Open the practical →

Keep this

Pick the tool that fits the job, not the most fashionable label on the box.