Hi, this is Ray.

I've written before in this newsletter about the risks of AI reliance… the specific way that outsourcing thinking to AI can slow your learning even when it feels like it's helping. Everything in that piece was true and worth keeping. But I'd be doing you a disservice if I left it there. Because AI is here, it's not going away, and the question for serious learners isn't whether to use it but HOW to use it. The framing of "AI good" versus "AI bad" misses that the tool can be used in ways that dramatically accelerate learning or in ways that dramatically undermine it. Same tool. Different uses. Categorically different outcomes.

I've been experimenting with this for the past couple of years, and what I've noticed (both in my own experience and in observing how other learners use these tools) is that the difference between productive and destructive AI use isn't subtle. There's a specific pattern that works, and a specific pattern that doesn't work, and once you see the distinction, it changes what you do when you sit down with an AI tool. The learner who uses AI as a Socratic tutor accelerates their learning. The learner who uses AI as an answer machine impairs it. Same tool. Different framing. Different outcomes over months and years.

Today's newsletter is about that. What the research actually shows about productive AI use for learning, the specific framework that separates helpful use from harmful use, and how to actually structure your AI interactions so they build your capability rather than replacing it. This is the constructive companion piece to my earlier newsletter about AI reliance risks. Together they give you the picture the research supports rather than either the utopian or dystopian version. Let's get into it.

AI is moving fast and most people are falling behind. 

The Rundown AI keeps you ahead of the curve. 

It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.

Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses — tailored to your needs.

Sign up to start learning.

Let me start with what the science actually shows, because there's genuine research emerging that clarifies the productive versus unproductive uses.

According to a 2025 framework published in Frontiers in Education, the Cognitive Mirror paradigm aims to reframe the role of generative AI in education, shifting its purpose from an omniscient tutor to a metacognitive partner who reflects the quality of a learner's explanation. While the broader AI deployment amplifies access and efficiency, it simultaneously induces cognitive dependency that suppresses active recall and problem-solving. Consequently, learners are less likely to engage in effortful but essential processes, such as retrieval practice, knowledge reconstruction, and error analysis. Read this carefully. The researchers explicitly distinguish between AI as "omniscient tutor" (giving you answers) and AI as "metacognitive partner" (reflecting on your thinking). The framing determines the outcome. Answer-provision suppresses learning. Reflection-provision supports it.

The specific mechanism the researchers identified is important. According to the same framework, in Oracle-style pipelines, learning devolves into a passive, one-way flow of information from AI to humans. When the system positions AI as an omniscient oracle, it diminished teaching accountability, weakening the Protégé Effect. Furthermore, a plain answer-providing LLM removes key desirable difficulties. Suppressed Reflective Practice: If the AI routinely supplies the correct solution path, learners lose the drive to try → observe → adjust, which is the core of reflection-in-action. This maps directly to what I've been writing about in previous newsletters about desirable difficulty, the protégé effect, and the importance of productive struggle. When AI removes the struggle, it removes the mechanism that produces the learning.

A 2025 systematic review of AI's role in metacognitive learning found similar patterns. According to the researchers, AI-based tools such as predictive models, intelligent tutoring systems, adaptive platforms, learning dashboards, and generative or conversational AI support goal setting, monitoring, strategy adjustment, and reflective evaluation through feedback, progress visualization, and personalized recommendations. Most studies report improvements in SRL strategies, metacognitive awareness, motivation, engagement, and learning outcomes. The productive uses focus on supporting the learner's own thinking rather than replacing it. Goal setting. Monitoring. Strategy adjustment. Reflective evaluation. These are all functions where AI supports what YOU are doing rather than doing it for you.

A specific study on generative AI environments found that metacognitive support matters critically. According to the researchers, the experimental group receiving explicit metacognitive support showed enhanced self-regulated learning compared to the control group. Metacognitive support has a significant advantage in enhancing self-regulated learning, but fewer studies have explored the effects of its role in generative AI environments. Same students. Same AI access. Different outcomes based on whether the AI use was structured to support metacognition or not. The framing again determines the effect.

Let me get specific about what productive AI use actually looks like. Based on the research and my own experimentation, there's a specific pattern that works.

The distinction is between using AI to skip cognitive work versus using AI to enhance cognitive work. Skipping cognitive work means asking AI to solve your problem, write your essay, produce your answer. Enhancing cognitive work means asking AI to help you think better about your problem, react to your draft, or check your reasoning. Same tool. Wildly different learning outcomes.

Here are the specific patterns of productive use:

AI as thinking prompter. Instead of asking AI to answer your question, ask it to help you think through the question. "What are the key considerations I should think about when analyzing this?" "What questions should I be asking about this topic?" "What framework might apply here?" This uses AI to structure your thinking rather than replace it. The thinking still happens in your head.

AI as feedback provider. Do your own work first. Then ask AI to critique it. "Here's my analysis of X. What am I missing?" "I've worked out this solution …what's wrong with it?" "Here's my draft essay …what would strengthen it?" This is like having a tutor available who reviews your work. The learning still happens through your own attempts, with AI providing the feedback that helps you improve.

AI as counter-example generator. After you've reached a conclusion, ask AI for cases where your conclusion might not hold. "I've concluded that X. What are the strongest counterarguments?" "In what cases would this framework not apply?" This forces you to encounter the limits of your own thinking, which is where deep understanding lives.

AI as explanation checker. After you think you understand something, try to explain it clearly to AI. Ask AI to identify where your explanation is unclear or where you might be misunderstanding something. This uses the protégé effect (you learn by teaching) while getting feedback on the quality of your teaching.

AI as question generator. When you're studying material, ask AI to generate questions about it that test different levels of understanding. Then answer the questions yourself. This is like having an infinite supply of practice tests. The learning still happens through your own retrieval work.

AI as connection finder. When you're trying to integrate new material with existing knowledge, ask AI for possible connections you might not have seen. "This concept is X. What other things does it connect to?" You then evaluate the connections and integrate them yourself. AI is showing you possible bridges. You still cross them.

AI as metacognitive coach. According to the research summary I cited, AI prompts reflection and self-regulation through structured questions. The AI helps students think about how they learn, what went well, and what to adjust next time. The main value here is metacognition. Students often struggle to step back and examine their own learning processes. An AI coach can ask the right questions at the right time to prompt that reflection. Use AI to ask you questions about your own learning process. "I've been studying X for a month. What patterns should I be noticing? What might I be missing?" This uses AI to support your own metacognition rather than doing the thinking for you.

Now let me name the specific patterns that undermine learning, because avoiding them is half the game.

Asking AI to solve your problems for you. "Here's a problem. Give me the answer." This is the pattern that damages learning most reliably. You get the answer without doing the cognitive work that would have produced learning. Same problem completed. Very different learning outcomes.

Asking AI to write your work. "Write me an essay about X." "Draft this report for me." The words appear, but no learning happened. You didn't struggle with organization, argument construction, or precise expression… the specific cognitive work that writing produces.

Using AI to skip the struggle. When something is hard, asking AI to make it easy. This bypasses exactly the productive struggle that produces learning. The Cloud Strife principle applies here… Cloud doesn't get stronger by skipping fights, he gets stronger by fighting through them. AI is a tool that can be used either to fight through hard material with better support, or to skip fights entirely. Only one produces character growth.

Accepting AI answers without verification. AI hallucinates. It gets things wrong. It presents confident-sounding answers that are sometimes incorrect. Accepting these without checking builds knowledge that's actually misinformation. Even when AI is right, accepting answers without verifying them prevents you from developing the ability to evaluate answers on your own.

Using AI to feel productive without producing learning. Some learners use AI extensively while doing very little actual thinking. They get a lot of text produced. Their brains do very little work. This produces the appearance of productivity without the substance of learning.

Confusing familiarity with understanding. Because AI can quickly familiarize you with almost any topic, it's easy to confuse knowing about something with actually understanding it. This is a genuine risk. AI-mediated familiarity often stays at the surface level. Actual understanding still requires deeper engagement than surface familiarity can provide.

Okay, here's the specific framework I use when working with AI for learning. Based on the research and my own experience.

Rule 1: Do your own thinking first. Before consulting AI, actually try to figure out the thing yourself. Attempt the problem. Draft the essay. Work through the concept. This is essential. AI use after your own attempt produces different results than AI use as a substitute for your attempt.

Rule 2: Show your work to AI, don't ask for its work. When you consult AI, share what you've already done and ask for feedback, refinement, or counter-perspectives. This structures the interaction so AI is responding to your thinking rather than replacing it.

Rule 3: Ask questions that require you to think more, not less. Prefer prompts like "What am I missing here?" over "What's the answer?" Prefer "Help me think through this" over "Do this for me." The framing of your prompts shapes what happens next.

Rule 4: Verify important claims independently. When AI tells you something factual, especially something that matters, check it against reliable sources. Don't accept AI statements as authoritative. This builds both accurate knowledge and the skill of evaluating information.

Rule 5: Explain concepts to AI after learning them. Use AI as an audience for your own teaching of the material. This activates the protégé effect while providing feedback on the quality of your understanding.

Rule 6: Use AI to identify weaknesses in your work, not to eliminate them for you. Ask AI to point out where your reasoning is weak, where your writing is unclear, where your analysis is incomplete. Then YOU fix the weaknesses. AI is diagnostic. You're the surgeon.

Rule 7: Set AI-free periods. Have deliberate periods where you don't use AI at all. This maintains your ability to think without the tool. Complete reliance on AI produces the specific weakness of not being able to work without it.

Rule 8: Track what your brain is doing, not just what got produced. When using AI, notice how much of the cognitive work is happening in your head. If very little, you're not learning. If substantial, you're learning better than you would without AI. The output matters less than what your brain was doing while it happened.

Here's a diagnostic question I've come to rely on. After any AI interaction, ask: "Did that make me think harder or think less?"

If the answer is "think harder"… the AI helped you engage more deeply with material, forced you to defend or refine your ideas, showed you connections that expanded your thinking (that was productive use. If the answer is "think less") the AI did work you would have done, gave you answers you would have figured out, replaced struggle with efficiency… that was probably counterproductive for your learning.

Same tool. Same time invested. Different quality of use based on what your brain was actually doing.

The learners I know who are getting the most benefit from AI are consistently answering "think harder" to this question. The learners who are getting less benefit are consistently answering "think less" without realizing it. This diagnostic is available to you in every AI interaction. Use it.

Here's what I want you to take from all this. AI is neither the magic learning tool the enthusiasts claim nor the learning-destroying force the skeptics fear. It's a tool that can be used in ways that dramatically enhance learning or in ways that dramatically undermine it. The specific patterns of use determine which outcome you get.

If you've been avoiding AI entirely because you've heard about the risks, please consider that skilled use can genuinely accelerate your learning. You're leaving value on the table by not using it at all. If you've been using AI extensively without seeing much learning benefit (or maybe seeing your capacity decline), please consider that the specific patterns of use might be the problem. The tool isn't destroying your learning. The specific way you're using it might be.

The framework I've laid out (AI as Socratic partner, not answer machine) captures what the research supports. Do your own thinking first. Show AI your work and ask for feedback. Use AI to prompt questions that make you think more, not to provide answers that make you think less. Verify important claims. Explain to AI to activate the protégé effect. Track whether your brain is engaging or coasting.

This isn't complicated. But it does require deliberate practice, because the default use of AI (the way most people interact with it when they don't think about it) tends toward the answer-machine pattern. Building the Socratic-partner pattern requires ongoing attention to how you're using the tool.

In Attack on Titan, the ODM gear is a powerful tool. Soldiers who use it as intended (as an extension of their own skill and decision-making) become highly effective. Soldiers who become dependent on it, using it as a substitute for their own judgment, get killed. Same tool. Different relationships. Different outcomes. AI works similarly for learners. Use it as an extension of your own thinking. Don't let it substitute for your thinking. The distinction sounds simple but it makes all the difference across the years of learning you'll do with these tools available.

Keep learning (and keep using AI thoughtfully),

Ray