Every week brings another “you are already behind” thread about AI tools. Most of them describe the same five capabilities with a new interface on top.
It helps to stop thinking in product names and start thinking in jobs to be done. Once you can name the job, you can tell whether a tool is genuinely useful or just confident.
The six jobs AI assistants do well
1. Explaining a concept at your level
Ask for an explanation of a topic you are studying, then ask follow-up questions until the gaps appear. This is the most reliable student use case, because you can check the explanation against your textbook or lecture notes.
2. Turning messy notes into structure
A wall of lecture notes becomes an outline, a set of flashcards, or a list of questions you still cannot answer. The value is in the reorganisation, not in the text generation.
3. Rubber-ducking code
Paste an error, describe what you expected, and ask for likely causes. Coding assistants are strongest at narrowing a search space. They are weakest when asked to invent a whole system you do not understand.
4. Summarising something long
Research abstracts, long documentation pages and recorded lectures can be compressed into a first pass. Always keep the source open. A summary that cannot be traced back is not evidence.
5. Drafting and editing prose
For emails, report structure and grammar, AI is a competent editor. For your own argument, it is a poor author. The ideas still have to come from somewhere.
6. Practising a conversation
Mock interview questions, viva preparation and explaining a project out loud all improve with a patient, tireless counterpart that will ask “why” five times.
Categories worth knowing
| Category | Typical use | What to watch for |
|---|---|---|
| General assistants | Explanations, drafting, brainstorming | Confident errors on specifics; no access to your syllabus |
| Coding assistants | Autocomplete, error analysis, small refactors | Suggestions that compile but are wrong for your architecture |
| Research assistants | Finding and summarising sources | Fabricated citations, missing context, paywalled sources |
| Study tools | Flashcards, quizzes, note restructuring | Shallow questions that reward recognition over understanding |
| Media tools | Image, audio and video generation | Licensing, disclosure and academic-integrity rules |
Naming a tool is less useful than naming the category, because products change pricing, features and model versions constantly. Check the official pricing page on the day you need it. Do not trust a blog post from last year, including this one.
A five-question test for any new AI tool
- What exactly does it do that my current workflow does not?
- Can I verify its output independently, in reasonable time?
- Where does my data go, and does it train on my content?
- What happens when I stop paying? Can I export my work?
- Does my institution allow it for this specific task?
If you cannot answer question two, the tool is entertainment. That is fine — just do not put it in your study workflow.
Where AI tools quietly fail
- Citations. A plausible-looking reference can be invented. Open every source before you use it.
- Numbers. Statistics and dates are frequently wrong. Trace them to a primary source or drop them.
- Recent events. Model knowledge has a cutoff. If the answer depends on something from this month, search instead.
- Your specific context. A generic answer cannot know your assignment brief, your team’s codebase or your professor’s expectations.
- Learning. Copying a correct solution feels like progress and is not. If you cannot reproduce it without the tool, you have not learned it.
How to use these tools without losing the skill
Use AI after your first attempt, not instead of it. Write the ugly first draft, attempt the problem, then ask for critique. The gap between your attempt and the critique is the actual lesson.
Keep a short log of what you asked and what you changed. When a viva examiner asks how you built something, that log is the difference between explaining your work and reciting a tool’s output.
The short version
- Learn the job categories, not the product names.
- Verify anything factual, especially citations and numbers.
- Check your institution’s rules before submitting anything.
- Use AI to critique your attempt, not to replace it.
- Re-check pricing and availability on the official site, never in a listicle.