Most people are content to be passive users, but the danger is that we are being conditioned to accept the first output that looks decent. If you aren't building with the intent to understand the system, you aren't leveraging the technology, you are just being optimized by it.
Thank you, that’s the direction I want to take it. “Think critically about AI” is too vague to be useful. I'll definitely share more articles on this! Thank you, Certainly!
I really like your approach and I'm happy that I already follow your suggestions. I'm building and I can see the limits just as you said. Congratulations
The 'become the product' framing is the right one, and sharper than most AI-literacy pieces manage. I'd push it one step further though. Running an AI assistant across a few businesses daily, the thing that separates using AI from being used by it isn't awareness, it's friction. Deliberate friction. Approval queues, verification steps, a human sign-off before anything goes out. The people who get quietly turned into the product are the ones who stripped out all the friction because the demo felt like magic. Critical literacy without a workflow that enforces it is just a nice intention. You end up trusting confident output because checking it is slower than accepting it, and that's precisely the behaviour these tools are built to reward. Attitude is necessary. It isn't sufficient. The question I'd add to your series: what does critical AI literacy look like as a habit you can't skip, rather than a mindset you're meant to hold? Mindsets erode the moment the tool is convenient enough.
I really appreciate this take, especially the "challenge" aspect of critical AI literacy. Lately, I'm finding a ton of value in pulling the thread with Claude in situations where I previously might have been quick to take its word as the truth. I'm simultaneously bumping up against some of AI's limitations and finding that these limitations can strengthen my own skills as long as I learn to recognize them and call them out.
The guidance is right on spot but it remains hard for a lambda user to enforce.
As a starter, I’d recommend users to expend the tiny “Read” that shows you what your AI is doing and how it thinks (available in Claude Cowork). Then testing every new automation multiple times until you’re happy with the results.
I’m a product builder at heart and thinking about adding a control layer (fifth word in your list) inside my Claude OS.
If interested to brainstorm what it should do and look like, DM me!
Great to hear you're all in! Control and enforcement are really useful extensions of the idea, thank you! Feel free to share your writing with the rest of us in the community chat.🤗
The "become the product" line is the right frame, and there's a builder's version that sits a layer below judgment: where the system actually runs. I got tired enough of the steering that I run my own AI stack on my own hardware ... on-prem k8s, the private data never leaves the building, the core on my own metal instead of someone's usage dashboard. Your human-only zones are the discipline; owning the substrate is that same instinct pushed down to the infrastructure. Use it hard, and own the floor it stands on.
Exactly, that's the builder move. Where we go deeper into also “who controls the runtime, the data path, the logs, the permissions, and the failure modes?” Thank you for reading and taking the time to comment, Jeff! 🤗
As to methods there may be a million and then some, but principles are few. The man who grasps principles can successfully select his own methods. The man who tries methods, ignoring principles, is sure to have trouble—Harrington Emerson
One thing I'd add to the "notice what the tool is doing to you" framing: the sneakiest version of this isn't in the outputs, it's in what questions you stop asking because the AI made the first answer feel sufficient. I've caught myself skipping the "is this even the right problem" step because the model gave me a confident, well-formatted wrong answer fast. Have you noticed critical AI literacy eroding faster for people who are otherwise strong critical thinkers offline, or does it track pretty evenly?
Great article! Critical AI literacy is four verbs:
- use, understand, evaluate, challenge
Most people stop at use.
Here is the part I want to push further.
The skill-erosion test she proposes is not a metaphor. It is a muscle test. Can you still draft the email cold. Can you still defend the answer under pressure. Can you reconstruct the reasoning without the chat window open.
If the answer is no, the tool is not augmenting you. It is eating you.
Run the test on yourself this week. One task. No model. See what is still yours.
Your point about becoming the product is incredibly sharp. Mainstream tech advice sells AI as a way to eliminate friction so we can move faster. But when we rush to automate all the friction in our workflows, we're in danger of accidentally automating our own thinking right out of the process. Convenience is a very expensive trade when the currency we are paying with is our own judgement.
I love how you explained this. I know a lot of people who definitely use it passively. I use it a little, but I'm geared to using an app someone made for the task rather than spit balling with the free open access that is far harder to point at what you want without the right 'box' to work in for proper context and framework. I don't have time to build it myself, so I research whose tools might help me with what I want to do regardless of AI or no AI in the app.
That makes total sense! Sometimes the smartest use of AI is choosing the right tool for the job. Thank you very much for reading and taking the time to comment, Athea 🤗
This lands, Karo, and I want to add the floor under it. The skill-erosion test you give, can you still write and think and explain under pressure, assumes a muscle that got built in the first place. For an adult who already has it, attitude protects it. For a kid who never trained it, there is nothing yet to protect, and the tool fills the gap before the judgment ever forms. So attitude is the right posture for people who already think, and the harder problem sits one layer earlier: how a person builds the judgment before the tool is ever in their hand. That is the part I work on. I teach chemistry and physics, and I have spent over 20 years watching kids either build that muscle or quietly skip it, and AI makes skipping it free. You are writing the field guide for the operator. I am trying to build the operator before they sit down. Same wall, two sides of it. I wrote a book on this called Cognitive Sovereignty if it is useful: https://www.amazon.com/dp/B0GGX2WTSJ
Yes, this is such a useful addition, Syd! You’re right that “use AI with attitude” assumes there is already some judgment to protect, some trained ability to think under pressure. For kids, the risk forms before the muscle forms at all. That makes your work on cognitive sovereignty feel like the foundation underneath it. I'm definitely going to dig deeper into it. Thank you! 🤗
That means a lot, Karo, thank you. Foundation and field guide is exactly right, and I think the two have to travel together. Judgment with no attitude gets passive, attitude with no judgment gets reckless. Your readers are already standing where the muscle can still form. I'll keep building the floor, you keep arming the operator, and somewhere in the middle a kid gets to keep their own mind. Glad we found each other's corner of this.
I'm in. Late August into September is perfect, it gives us both room to do it right instead of fast. I love the idea of building it together, your operator and my foundation in one piece, the two halves of the same wall. Let's find the thread that only works because both of us are on it. I'll reach out as we get closer to lock the shape. Really glad we found each other's corner of this.
Most people are content to be passive users, but the danger is that we are being conditioned to accept the first output that looks decent. If you aren't building with the intent to understand the system, you aren't leveraging the technology, you are just being optimized by it.
Exactly. The part that trains behaviour is the most interesting one for me. Thank you for reading, Sharyph! 🤗
Thank you, Karo. I would be very interested in further elaboration on this framework, as it offers an excellent way to systematize thinking about AI.
Thank you, that’s the direction I want to take it. “Think critically about AI” is too vague to be useful. I'll definitely share more articles on this! Thank you, Certainly!
Sorry, autocorrect, I meant Certainty 🤗
I really like your approach and I'm happy that I already follow your suggestions. I'm building and I can see the limits just as you said. Congratulations
Thank you! I’m really happy it resonated with you!
The 'become the product' framing is the right one, and sharper than most AI-literacy pieces manage. I'd push it one step further though. Running an AI assistant across a few businesses daily, the thing that separates using AI from being used by it isn't awareness, it's friction. Deliberate friction. Approval queues, verification steps, a human sign-off before anything goes out. The people who get quietly turned into the product are the ones who stripped out all the friction because the demo felt like magic. Critical literacy without a workflow that enforces it is just a nice intention. You end up trusting confident output because checking it is slower than accepting it, and that's precisely the behaviour these tools are built to reward. Attitude is necessary. It isn't sufficient. The question I'd add to your series: what does critical AI literacy look like as a habit you can't skip, rather than a mindset you're meant to hold? Mindsets erode the moment the tool is convenient enough.
Thank you for reading, Chris! I agree with every word. Attitude only covers the mindset part.
I really appreciate this take, especially the "challenge" aspect of critical AI literacy. Lately, I'm finding a ton of value in pulling the thread with Claude in situations where I previously might have been quick to take its word as the truth. I'm simultaneously bumping up against some of AI's limitations and finding that these limitations can strengthen my own skills as long as I learn to recognize them and call them out.
So good that you're doing this. That's exactly it. Output = something to inspect. Thank you for reading and taking the time to comment, Sara! 🤗
Great read! I’m all-in on AI with attitude.
The guidance is right on spot but it remains hard for a lambda user to enforce.
As a starter, I’d recommend users to expend the tiny “Read” that shows you what your AI is doing and how it thinks (available in Claude Cowork). Then testing every new automation multiple times until you’re happy with the results.
I’m a product builder at heart and thinking about adding a control layer (fifth word in your list) inside my Claude OS.
If interested to brainstorm what it should do and look like, DM me!
Great to hear you're all in! Control and enforcement are really useful extensions of the idea, thank you! Feel free to share your writing with the rest of us in the community chat.🤗
And thank you for reading, Fabrice!
The "become the product" line is the right frame, and there's a builder's version that sits a layer below judgment: where the system actually runs. I got tired enough of the steering that I run my own AI stack on my own hardware ... on-prem k8s, the private data never leaves the building, the core on my own metal instead of someone's usage dashboard. Your human-only zones are the discipline; owning the substrate is that same instinct pushed down to the infrastructure. Use it hard, and own the floor it stands on.
Exactly, that's the builder move. Where we go deeper into also “who controls the runtime, the data path, the logs, the permissions, and the failure modes?” Thank you for reading and taking the time to comment, Jeff! 🤗
As to methods there may be a million and then some, but principles are few. The man who grasps principles can successfully select his own methods. The man who tries methods, ignoring principles, is sure to have trouble—Harrington Emerson
https://quoteinvestigator.com/2015/07/17/methods/
One thing I'd add to the "notice what the tool is doing to you" framing: the sneakiest version of this isn't in the outputs, it's in what questions you stop asking because the AI made the first answer feel sufficient. I've caught myself skipping the "is this even the right problem" step because the model gave me a confident, well-formatted wrong answer fast. Have you noticed critical AI literacy eroding faster for people who are otherwise strong critical thinkers offline, or does it track pretty evenly?
Great article! Critical AI literacy is four verbs:
- use, understand, evaluate, challenge
Most people stop at use.
Here is the part I want to push further.
The skill-erosion test she proposes is not a metaphor. It is a muscle test. Can you still draft the email cold. Can you still defend the answer under pressure. Can you reconstruct the reasoning without the chat window open.
If the answer is no, the tool is not augmenting you. It is eating you.
Run the test on yourself this week. One task. No model. See what is still yours.
Your point about becoming the product is incredibly sharp. Mainstream tech advice sells AI as a way to eliminate friction so we can move faster. But when we rush to automate all the friction in our workflows, we're in danger of accidentally automating our own thinking right out of the process. Convenience is a very expensive trade when the currency we are paying with is our own judgement.
Great post. I am not very polite to AI Agent: but I like it!
Great info. Thanks
https://substack.com/@johncharlesharman/note/c-282568315
I love how you explained this. I know a lot of people who definitely use it passively. I use it a little, but I'm geared to using an app someone made for the task rather than spit balling with the free open access that is far harder to point at what you want without the right 'box' to work in for proper context and framework. I don't have time to build it myself, so I research whose tools might help me with what I want to do regardless of AI or no AI in the app.
That makes total sense! Sometimes the smartest use of AI is choosing the right tool for the job. Thank you very much for reading and taking the time to comment, Athea 🤗
This lands, Karo, and I want to add the floor under it. The skill-erosion test you give, can you still write and think and explain under pressure, assumes a muscle that got built in the first place. For an adult who already has it, attitude protects it. For a kid who never trained it, there is nothing yet to protect, and the tool fills the gap before the judgment ever forms. So attitude is the right posture for people who already think, and the harder problem sits one layer earlier: how a person builds the judgment before the tool is ever in their hand. That is the part I work on. I teach chemistry and physics, and I have spent over 20 years watching kids either build that muscle or quietly skip it, and AI makes skipping it free. You are writing the field guide for the operator. I am trying to build the operator before they sit down. Same wall, two sides of it. I wrote a book on this called Cognitive Sovereignty if it is useful: https://www.amazon.com/dp/B0GGX2WTSJ
Yes, this is such a useful addition, Syd! You’re right that “use AI with attitude” assumes there is already some judgment to protect, some trained ability to think under pressure. For kids, the risk forms before the muscle forms at all. That makes your work on cognitive sovereignty feel like the foundation underneath it. I'm definitely going to dig deeper into it. Thank you! 🤗
That means a lot, Karo, thank you. Foundation and field guide is exactly right, and I think the two have to travel together. Judgment with no attitude gets passive, attitude with no judgment gets reckless. Your readers are already standing where the muscle can still form. I'll keep building the floor, you keep arming the operator, and somewhere in the middle a kid gets to keep their own mind. Glad we found each other's corner of this.
I love this! Such a great way to look at the problem, together. If you're up for it, we could do a collab post in late August/September 🤗
I'm in. Late August into September is perfect, it gives us both room to do it right instead of fast. I love the idea of building it together, your operator and my foundation in one piece, the two halves of the same wall. Let's find the thread that only works because both of us are on it. I'll reach out as we get closer to lock the shape. Really glad we found each other's corner of this.
Sounds wonderful!