Lesson 02 · 8 minutes
Find a task worth using AI for
Judge potential tasks by repetition, ambiguity, consequence, reversibility and whether a capable person can check the result.
Why this matters
The best demonstration task is rarely the best real-world task.
A small reversible experiment creates better evidence than a broad commitment.
The practical bit
AI is often useful where work contains language, messy information, options or drafts, and where a person can recognise a good result. Repetition and volume create possible leverage, but they do not remove the need to understand errors.
Start cautiously when inputs are sensitive, errors affect people, results are difficult to check or the action is hard to reverse. In those cases AI may still assist with a bounded part, but should not own the decision or action.
What it looks like
Turning internal meeting notes into a first draft of a weekly update is bounded and reviewable. Automatically sending performance feedback to staff is consequential and relationship-sensitive.
Creating a first packing list is easy to check. Relying on an unsourced answer about medication is high consequence and requires authoritative professional guidance.
Starting with the task that consumes the most time without checking whether failures would consume even more time, trust or money.
Use your judgement
Try it now
- Write one repeated or frustrating task.
- Rate error consequence, checkability and reversibility as low, medium or high.
- Reduce the scope until one safe test is obvious.
Use this in a real situation
Choose a practical taskCompare six bounded, checkable starting points before choosing a product.Open the practical →Keep this
Start where usefulness is real, checking is possible and failure is recoverable.