General AI

What AI does

Reference only. Project progress and credentials come from assessed work.

AI can feel mysterious because it can write, analyse and answer in natural language. The useful explanation is much simpler: AI receives an input, uses patterns it learned, and produces an output.

The simplest useful mental model

Every AI task has three parts:

01
Input
Your task and context
02
Process
Patterns applied to the information
03
Output
A draft answer or action
PartPlain-language meaningExample
InputWhat you ask and what information you provide"Summarise this meeting transcript"
ProcessThe AI identifies relevant patterns and relationshipsIt finds decisions, owners and dates
OutputWhat the AI returnsA structured meeting summary

AI does not experience the meeting. It works from the transcript and your instruction. If the transcript is incomplete, the result can be incomplete too.

What AI is useful for

AI is often useful when work involves:

  • Summarising a large amount of information.
  • Finding patterns across documents or records.
  • Producing a first draft in a defined format.
  • Repeating a clear process across many items.
  • Comparing options against a rubric.

The best tasks have a visible finish line. "Create a table with these five columns" has a finish line. "Help with marketing" does not.

What AI cannot guarantee

AI can produce a confident sentence that is incomplete, outdated or wrong. It does not automatically know:

  • Whether your source is current.
  • Whether important information is missing.
  • Whether a company rule changed yesterday.
  • Whether an action is appropriate in your situation.

Assistant and agent

An AI assistant helps you create or understand something. You remain closely involved.

An AI agent can plan steps and use allowed tools to complete a defined task. Because an agent can do more, its goal, permissions and review points must be clearer.

You do not need to memorise these labels. Ask one practical question: Is the AI only proposing something, or can it also take action?

Recap

  • AI produces outputs from your instruction, context and learned patterns.
  • Clear tasks with a visible finish line work best.
  • A fluent output can still be wrong.
  • More ability to act requires more human control.
What AI does | Mariete Academy