Most people quit somewhere between their first impressive result and their first useless one. Here is why tutorials fail, what three honest levels of skill look like, and how to tell when you are ready to move up.
Why tutorials fail
Almost everyone's first experience with AI is encouraging. You ask something, you get a fluent answer, and it feels like a skill has been acquired. The second experience is usually the opposite: the answer is confidently wrong, or generic, or subtly unusable, and there is nobody to ask why.
That gap is where most self-teaching stops. It is not a motivation problem. Video tutorials are very good at showing you a result and very bad at explaining the judgement behind it — when to trust an output, how to tell a plausible answer from a correct one, what to do when the tool is simply the wrong choice.
Those are the parts that take a person to teach, which is why our courses are taught by people rather than recordings.
Three honest levels
We organise learning into three stages, and the boundaries between them are about judgement, not vocabulary.
- Novice. You can get reliably useful results for everyday tasks — writing, summarising, translating, planning — and you can recognise when an answer is wrong. The goal here is confidence without superstition.
- Middle. You can build something that other people use: a repeatable workflow, a prompt that works for a whole team, an automation connecting two tools. The goal is reliability under someone else's hands.
- Expert. You can design a solution for a business problem, judge which model or approach fits, and be honest about limits and risks. The goal is responsibility for an outcome, not just a tool.
What a week of study actually looks like
Three to five hours, split between a session with a teacher and your own work on a task you actually care about. The second half is the part people skip and the part that makes it stick.
We insist that students bring real work. Practising on invented exercises produces the tutorial problem all over again: you learn the steps without learning when the steps apply. Bring the report you have to write, the data you have to clean, the enquiries you have to answer.
Progress is also less linear than course outlines suggest. Expect a week where everything clicks and a week where you feel you have gone backwards. The second kind is usually when the judgement is forming.
Group or one-on-one
Small groups work well when your goal is general capability and you benefit from hearing other people's mistakes — which is a genuinely efficient way to learn. They are also cheaper, and the pace is set for you, which helps if you would otherwise drift.
One-on-one is the better choice when you have a specific, slightly unusual goal, when your schedule is unpredictable, or when you are learning for a role where mistakes carry weight. It is also the right answer when you are further along than you think and a group would hold you back.
We teach in Vietnamese, English and Russian, and it is worth choosing the language you think in rather than the one you are most proud of. Learning judgement in a second language costs you attention you need elsewhere.
Signs you are ready for the next level
You can explain to a colleague why an output was wrong, not just that it was. You find yourself reaching for a different approach before being told to. You have stopped being impressed by fluency.
And the clearest sign of all: someone else has started using something you built, and it keeps working when you are not in the room.