Many engineering students are using AI as a tutor, not as a generator.

Most conversations about students using AI still sound the same.

People assume AI is mainly helping students get through work faster. Less effort. Less thinking. More shortcuts.

That is the fear behind a lot of the reaction.

But DORA’s research points to a more useful reality. For many students, AI is not simply replacing work. It is helping them learn. In DORA’s study of UC Berkeley students across Electrical Engineering, Computer Science, Design, and Data Science, students repeatedly described AI as a tutor or teacher rather than just a tool for output.

The real question is not just whether students are using AI. They already are. The real question is what role AI is starting to play in the learning process. And right now, that role looks a lot like support.

What DORA found

DORA’s research came from an eight month mixed-methods study focused on how students are actually using AI in academic workflows. What they found was more nuanced than the usual cheating narrative.

Students were not only using AI to get answers.

They were using it to:

• understand difficult concepts
• explain code
• fix small errors
• summarize papers
• clarify assignments
• figure out what to study first
• identify what they still did not understand

DORA also notes that in all 11 interviews, students used terms like tutor or teacher to describe their relationship with AI.

That is not a small detail. It suggests AI is becoming less of a shortcut machine and more of an on-demand support layer.

The shift is really about access

The biggest thing AI changes is not intelligence. Before AI, getting unstuck was slower and more fragmented. A student had a few options. Wait for office hours. Ask a TA. Search through forums. Open ten tabs. Try to figure out which explanation actually applied to their situation.

DORA highlights this directly. Students compared AI with older workflows built around instructor support and online search, both of which had limits. Human help was constrained by time. Search was often slow and scattered.

AI compresses that process.

Instead of hunting for the right thread, students can describe the problem directly.

Instead of waiting until tomorrow, they can ask in the moment.

Instead of asking only for an answer, they can ask for an explanation in simpler language.

That changes the experience of learning in a big way.

Because a lot of learning breakdown does not happen when the material is impossible. It happens when confusion lasts too long and nobody is there to help resolve it.

That is where AI becomes useful. Because it reduces delay.

Students are using AI to study, not just solve

One of the strongest parts of DORA’s research is that students were not just using AI to fix problems. They were using it to guide their own learning. Some used it to summarize papers mentioned in class so they could decide which ones were worth reading in full. Others used it to figure out which lecture or concept they should review before starting a project. DORA describes this as a metacognitive use of AI.

That matters because students often struggle with more than the material itself. They also struggle with direction.

They do not always know:

• where to start
• what concept is blocking them
• what to review first
• whether they need theory or practice
• which resource deserves their time

In simple terms, it helps students ask better questions.

The real value may be speed to understanding

We often talk about AI as if its main value is speed. And yes, it does make some things faster. But the more important kind of speed here is not speed to output. It is speed to understanding.

That is one of the most useful takeaways from DORA’s article. Students said AI helped them understand dense material faster because they could interact with a chatbot directly instead of spending more time searching across outside resources.

That is a healthier benchmark. Not how fast someone can submit. How fast someone can understand.

That difference matters a lot.

Because faster output does not always mean better learning. But faster understanding can.

AI is removing some low-value friction

AI is helping students remove certain kinds of low-value friction. Students said they were using it for things like syntax help, troubleshooting, and lower-level technical issues. Some explained that because AI helped with those smaller obstacles, they had more time to focus on logic, ideas, systems, and bigger-picture thinking.

There is a difference between productive struggle and pointless friction.

Productive struggle helps people learn.

Pointless friction just burns time and attention.

If a student is wrestling with a concept, that can be valuable. If they are wasting an hour on an avoidable syntax issue or a confusing error message, the learning return is much lower.

AI can help reduce that second category. And when it does, it can create more room for higher-level thinking.

That may be one of the most important educational effects of AI, especially in technical fields.

The risk is not AI itself. It is passive use

This is the part that needs to stay honest. DORA’s article is not blindly positive, and that is exactly why it is useful.

The students in the study were not treating AI as flawless. Some were very clear about the need to verify what it gave them. One student explained that they used AI for syntax, error codes, and stylistic help, but still made sure they understood the code flow if AI had contributed to something larger. They also pointed out that AI-generated code can be completely wrong.

That is the line everything depends on.

AI works best as a tutor when the learner stays active.

That means:

• asking follow-up questions
• checking whether the response is actually correct
• testing the explanation
• restating the idea in your own words
• using the answer to solve the next problem independently

The danger is not really the tool by itself. The danger is passive use.

The moment someone stops checking, stops thinking, and stops verifying, AI stops being a tutor and starts becoming a crutch. That is why the conversation should move away from asking whether AI is good or bad in education.

A better question is this:

What kind of AI use are students learning to practice?

That is where the real difference will come from.

This goes beyond education

The patterns students are building now are likely to carry into work. DORA connects these findings to broader organizational learning too. They point out that AI can support self-service learning, onboarding, access to knowledge, and better flow when it is backed by strong documentation and useful internal information.

That maps directly to how teams already work.

Inside companies, people are using AI to:

• understand unfamiliar systems
• troubleshoot faster
• summarize dense documentation
• learn new tools
• onboard faster
• keep moving when human support is not immediately available

So this is not just a student story. It is a preview of how learning is changing across knowledge work.

What educators should take from this

The biggest takeaway from DORA’s work is simple.

AI is already part of how students learn. That means the real opportunity is not pretending it is not there. The opportunity is teaching people how to use it well.

That includes teaching students:

• how to verify AI outputs
• how to document their process
• how to separate support from substitution
• how to use AI to deepen understanding rather than avoid thinking

If schools only respond with blanket resistance, students will still use AI. They will just do it without guidance. That usually creates worse habits, not better ones. A better response would be more honest and more practical.

- Teach critical use.

- Teach judgment.

- Teach verification.

- Teach when AI helps learning and when it weakens it.

That is far more valuable than acting like the tool will disappear.

Dependency is real. Misuse is real. Weak learning habits are real. But the article adds something the public conversation often lacks: Nuance.

For many students, AI is becoming a tutor before it becomes a shortcut. It is helping them understand faster, debug confusion, figure out what to study next, and spend more time on reasoning instead of getting stuck in low-value friction. That does not mean learning disappears. It means the conditions around learning are changing. And that is probably the real story.

Not that students are learning less.