Which Tasks Suit an AI Assistant? Good and Poor Fits

Phase 1: Prepare

Step 1 Warm-up recall

From M1.L1: explain in one sentence why a reply is not a lookup.

Your answer

Private: nothing you type here is sent or stored.

Step 2 Learning goal

By the end of this lesson you can:

name tasks that usually suit an assistant (Foundation);
name tasks that need strong checking (Core);
explain the reason for each (Core).

Phase 2: Teach

Step 3 A familiar example

A new assistant at work is quick, well-read and eager. You would happily ask them to draft an email, but you would check an exact figure they quoted.

Step 4 Clear explanation

The pattern is simple. An assistant is usually a good fit where the result is easy for you to judge on sight and where a weak result costs little: drafting, rewriting, brainstorming and summarising text you can read yourself. It is a poor fit, or needs strong checking, where details must be exact: dates, quotations, citations, numbers, and recent events.

Task type	Usually a good fit?	Why
Drafting a routine email	Yes, with a read-through	The output is easy to judge on sight.
Rewriting text in plainer words	Yes, with a read-through	You supply the content and can compare.
Brainstorming options or titles	Yes	Weak ideas cost nothing to discard.
Summarising text you can read	Yes, with a spot-check	You can check the summary against the source.
Exact facts, dates, quotations, citations	Only with checking	It can produce plausible but wrong details.
Arithmetic and long calculations	Only with checking	Fluent text is not a calculator.
Events after its training	Not on its own	It may not know recent events unless connected to a source.

The right-hand column carries the reasoning. If you can judge the output quickly, a mistake is likely to be caught. If the output looks the same whether right or wrong, a mistake can pass unnoticed. Always ask: how would I know if this were wrong?

Also ask whether the tool is connected to a trustworthy source. Some assistants can search or use tools; their answers then depend on those sources, which you can read. Others rely only on what they absorbed in training, which may be out of date.

Step 5 Worked example (4 marks)

Classify: “summarise this 12-page report that I have not read”.

Summarising is usually a good fit. (1 mark)
But you have not read the source, so you cannot spot a missed or invented point. (1 mark)
Rule: only summarise what you can check; read the key sections yourself. (1 mark)
Better task: ask for a structure of the report, then read the sections it points to. (1 mark)

Phase 3: Practise

Step 6 Learner practice (instant-marked)

1. Which task is usually a good fit?

2. Which needs strong checking?

3. A useful question to ask before using output is:

4. An assistant not connected to any source may be unreliable about:

Phase 4: Close

Step 7 Recap

Good fits: easy to judge and cheap to get wrong.
Poor fits: exact details, numbers and recent events.
Always ask how you would know if the output were wrong.

Step 8 Exit ticket (ungraded)

List three tasks you do and place each as good fit or needs checking.

Your answer

Private: nothing you type here is sent or stored.

Step 9 Go further (Extension)

Choose a task from the table and design one quick check you could do in under two minutes.

Common mistakes

Mistake	Why it fails
Asking it for exact figures and trusting them	Check against a source.
Summarising what you have not read	You cannot spot omissions.
Skipping the read-through of a routine draft	Tone or detail may be off.

Limits. The table is a general guide; the right level of checking also depends on the stakes, which the next lessons cover.