Over the past several months we have been exploring what happens when AI is trained not just on information, but on the principles of Process Education. Rather than building a chatbot that gives answers, we’re working on learning guides that ask better questions, encourage reflection, and help students think through challenges instead of around them.
That distinction turns out to be harder to build than to describe, and it starts with getting the measure of success right.
The wrong measure
Ask most people what makes an AI tutor good and they’ll describe the quality of its answers: accurate, clear, patient, available at three in the morning. All of that is real…and none of it is the point.
A learning guide’s success is not measured by the quality of the immediate answer. It is measured by whether the learner becomes more able to govern their own learning without requiring the same level of support.
That single shift changes everything downstream. A tool optimized for answer quality will always drift toward doing more of the work, because doing more of the work produces better answers. A tool built to develop capability has to do something considerably harder: work out what this particular learner needs to perform right now, and then decline to do that part — however easily it could.
Effort isn’t the goal either
The obvious overcorrection is to make the learner struggle more. Withhold things. Answer every question with another question. Make them earn it.
But that’s wrong too, and we need to be precise about why. Learner effort alone is not the goal. If it were, then tears and sweat would be enough to merit graduation. The goal is learner performance that develops capability. Confusing the two produces a tool that wastes people’s time on principle.
Nobody develops as a thinker by reformatting a table, chasing down a citation, or reconstructing a definition they already understand. There is no capability at stake in that work, and a good guide should absorb it without ceremony — because every minute spent there is a minute not spent on the performances that actually matter.
So the design question is not how much should the AI do? It is which performances must this learner carry out now in order to develop? Everything else is fair game.
In practice that sorts into three categories.
- Some work a guide should simply handle: organization, formatting, retrieval, worked examples, tracking criteria, and summarizing back what the learner has said
- Some it should scaffold but never replace: noticing, questioning, interpreting, surfacing assumptions, forming explanations, and generalizing
- And some it should never take at all — the learner authorizes what they conclude, value, intend, and carry forward. Not because a machine could not produce a plausible version, but because a conclusion someone else authored is not a conclusion. It is a quotation.
Where the help should live
Our vision is straightforward: when students need help, it should be available right where they are learning.
Imagine working on an assignment inside an online course and having access to a guide that understands self-growth, assessment, reflection, learning skills, and performance. Rather than solving the problem for you, it would help you analyze the situation, identify what you already know, recognize where you’re stuck, and decide on the next best step. The goal is not greater dependence on AI. It is greater independence as a learner.
This reflects a belief that has guided Pacific Crest for decades:
Learning is not about collecting answers, it is about developing capability. Technology should strengthen that process rather than replace it.
Embedding this kind of coaching directly in an LMS means students don’t have to go looking for support because guidance becomes part of the learning experience itself. Faculty benefit too, extending their teaching presence beyond the classroom, knowing students have access to coaching aligned with the philosophy already at work in their courses.
We are still experimenting
Every interaction teaches us something about how AI can better support reflection, improve decision-making, and reinforce the habits of successful learners. We expect these tools to keep changing as we learn from those who use them, whether they’re faculty, students, or coaches. Some of what we learn is uncomfortable, which is how we know it’s worth learning.
The future of education will not be defined by artificial intelligence. It will be defined by how thoughtfully we use it to help people become more capable, more confident, and more self-directed.
Which is why we would offer you a question to put to any AI tool your institution is considering, any from us very much included:
After a semester with this, can the student do more without it?
If nobody can answer that, that’s the answer.