What today's AI doesn't see when you learn

Ask a modern AI model almost anything and you'll get a fluent answer. It has read more text than any person could read in many lifetimes. On knowledge of the world, it is extraordinary.

On what happens while you learn, it sees very little.

It doesn't know what you got right last week. It doesn't know which explanation finally made something click, or which one confused you. It can't tell whether you still remember a concept or only recognise it. Most of the time, each session starts close to zero.

Memory features help a little. They can store what you said: your name, your goals, your preferences. But what people say about their learning and how their learning actually goes are different things. We are often confident about things we haven't learned yet, and unsure about things we know well. A system that only listens to what you tell it inherits those blind spots.

Learning makes the gap obvious. A good teacher doesn't only answer questions. They check. They notice the mistake that keeps coming back, the step that gets skipped, the idea that needs a second angle. And they change what they do next because of it. That judgement comes from evidence, not from a questionnaire.

What AI is missing isn't more knowledge. It's evidence of what actually helped.

That's what Waysflow is building. The idea is simple to say and hard to do well: pay attention to what happens while someone learns. Which answers were right and which were wrong. What could still be recalled days later. Whether an idea could be used on a new problem. How much help was needed. Those observations become evidence, and evidence can inform what happens next.

Any estimate built this way has to stay provisional. It covers a specific concept, it carries uncertainty, and it's allowed to be wrong. "Not enough evidence yet" is a perfectly good answer.

We're starting with learning because it's where this loop is clearest. You can teach, check, and see whether it worked. As AI becomes part of how people study and learn, it should get better at helping each person, not only better at knowing things.

The world's knowledge is now easy to reach. Knowing what actually helps someone learn it is still hard. That's the problem we've chosen.