AI commonly classifies, predicts, generates and assists. A single workflow may use all four.
Do now: Take one repeated task and label each step with C, P, G, A or H for human-only.
Automate a clear step, not a vague hope.Choosing by job keeps you from buying a shiny tool before understanding the problem. It also reveals where data, testing and human review belong.
Classify: decide what kind of thing this is
Classification assigns categories using learned patterns. It can sort messages by topic, group images by visible features or route requests by urgency. Good categories are defined before the system is judged.
Classification becomes dangerous when categories are vague, unfair or used to decide a person’s future without review. Test edge cases and allow an ‘uncertain’ route.
Useful question: What are the categories, and what happens when the system is unsure?
Predict: estimate what may come next
Prediction estimates a likely outcome from past patterns: demand next week, arrival time, equipment risk or the next word in a sentence. A prediction needs a time horizon, assumptions and a measure of error.
When conditions change, the past may be a poor guide. Weather, markets, behaviour and rare events can move outside the training pattern. A range with uncertainty is often more honest than one impressive-looking number.
Generate: create a draft or option
Generation produces new text, images, audio, video or code from learned patterns and instructions. It is excellent for variation and first drafts. It does not automatically secure truth, originality, rights or audience fit.
Treat generated work as material for selection and revision. Check sources, consent, likeness, brand voice, accessibility and the rules of the platform where it will appear.
Generation gives you options. Taste and responsibility still need an owner.
Assist: support a larger human task
Assistance combines capabilities around a human goal: summarise a meeting, propose next actions, prepare a lesson outline or compare supplier information. The person should still see the source, edit the output and make the decision.
Assistance is often the best starting point because it keeps work visible. Once a workflow is reliable, low-risk steps can become more automated while exceptions still return to a person.


