I really enjoyed this article and many the points you outlined for using AI feel true to me. We should be aware of how to engage AI where it’s most helpful and can more efficiently produce work or outcomes that might taken us longer to do.
Very good writing. I agree that we need to keep our eyes open while using AI because we need to keep ourselves in the loop of doing things. As a developer, I sometimes feel that we are passing too much work to the AI, which can lead to a loss of control over the results of our actions. For some things, it could be manageable; however, if I need to work on that result later, I need to have a very good knowledge of the technology so I can change the places where should be changed.
That's a very good essay, and it is important to keep this subject under discussion.
Great article, Luca. A key point when you say that we humans aren't as good at remembering things, but we're much smarter, has everything to do with that. We're very good at connecting things, and we become better at it when we keep reinforcing this habit. Another point I personally really like is building things, and handing that over to AI and losing the fun part doesn't seem so appealing, haha.
Really enjoyed reading this. Super well written. I often use a similar analogy to your 'Create Mode and Review Mode'. In fact, you come pretty close to it at the start of your post. I often talk about a "new technology" in food with the introduction of what used to be called "Convenience Food" in the post-war period with higher female employment and generally changing social habits. Today we call it "Processed Food" or highly processed food. And again the difference here is between going "full stack", i.e. shopping for the fresh ingredients, preparing, cooking and then eating or just going the easy "don't make me think" route of simply heating in the micro a "vellutata di carciofi"! Going 'full-stack' -> healthy eating habits (plus exercise of course) people more in shape, "don't make me think" -> obesity and overweight, good for Ozempic however!
The food metaphor is perfect! I hope this time we'll be faster at spotting the danger and not wait for >50% overweight rate (in US is 74%!) to do something about it.
Thank you Luca. This one got me thinking a lot. I ended up with more questions than answers and on this topic that might be a good thing. I did enjoy the differences between create and review mode as well as knowledge versus reasoning. My first instinct was to ask AI to summarize your article and then come up with comments but I wrote the comment first without AI :)
And fwiw, this is what my highly trained balkiGPT came up with:
---
I really like the way you framed this. The tension between create mode and review mode feels spot on—it’s like a new kind of cognitive diet we’re all going to have to figure out. AI gives us leverage, sure, but the bigger question is: are we keeping our mental muscles in shape while we use it?
For me, the risk isn’t that AI makes us “dumber,” it’s that it quietly shifts us into a passive gear without us noticing. That moment when you realize you’ve just been editing a draft instead of wrestling with an idea from scratch—that’s the part I worry about. Protecting that messy, energy-draining create mode might become the most important skill of all.
Insightful essay. One thing I was wondering though, you mention how AI is superior at Knowledge but Reasoning poor. But in its current state, AI has often been found to straight up give false and incorrect information. How do you work around that when offloading the Knowledge sector? And for sake of applicability, lets work under the assumption that we don't know when this will be sufficiently improved enough to no longer be an issue and want to look for solutions today. (So no, "well, it'll eventually in the future get to the point where that's not a problem")
I believe there is no way to 100% solve hallucinations. It's way better than it used to be, but the risk is still there.
In many cases, though, a small share of hallucinations is acceptable as a tradeoff when in return you get extremely cheap, always-available expertise. In some cases it's not and we'll have to limit AI autonomy
There's also the aspect of work independence here. In a normal corporate software position, business pushes delivery. I feel that we as developers are pushed -mostly maybe by ourselves- to make use of AI for creative work/reasoning to accelerate delivery. Some things we could do with our brains of course but would make it slower. And then comes in the laziness part to delegate once-hard thinking problems to AI for a first draft solution or some ideas.
I believe that making more use of AI for creative work is not necessarily bad, granted that you can spend the time you save on equally (or, ideally, more) creative activities.
Example: if you delegate some creative software dev to AI so you can own entire features by yourself and act as a mini PM, that looks good to me. But if you instead spend the rest of your time fixing AI bugs and writing tests — that's not good!
I really enjoyed this article and many the points you outlined for using AI feel true to me. We should be aware of how to engage AI where it’s most helpful and can more efficiently produce work or outcomes that might taken us longer to do.
Thank you Jack, it means a lot. I agree with you
Very good writing. I agree that we need to keep our eyes open while using AI because we need to keep ourselves in the loop of doing things. As a developer, I sometimes feel that we are passing too much work to the AI, which can lead to a loss of control over the results of our actions. For some things, it could be manageable; however, if I need to work on that result later, I need to have a very good knowledge of the technology so I can change the places where should be changed.
That's a very good essay, and it is important to keep this subject under discussion.
Great article, Luca. A key point when you say that we humans aren't as good at remembering things, but we're much smarter, has everything to do with that. We're very good at connecting things, and we become better at it when we keep reinforcing this habit. Another point I personally really like is building things, and handing that over to AI and losing the fun part doesn't seem so appealing, haha.
If it's fun we shouldn't give it up! At most, we can trade it for something else that is equally fun.
Ciao Luca,
Really enjoyed reading this. Super well written. I often use a similar analogy to your 'Create Mode and Review Mode'. In fact, you come pretty close to it at the start of your post. I often talk about a "new technology" in food with the introduction of what used to be called "Convenience Food" in the post-war period with higher female employment and generally changing social habits. Today we call it "Processed Food" or highly processed food. And again the difference here is between going "full stack", i.e. shopping for the fresh ingredients, preparing, cooking and then eating or just going the easy "don't make me think" route of simply heating in the micro a "vellutata di carciofi"! Going 'full-stack' -> healthy eating habits (plus exercise of course) people more in shape, "don't make me think" -> obesity and overweight, good for Ozempic however!
Nice one!
The food metaphor is perfect! I hope this time we'll be faster at spotting the danger and not wait for >50% overweight rate (in US is 74%!) to do something about it.
Not super optimistic about it though!
Thank you Luca. This one got me thinking a lot. I ended up with more questions than answers and on this topic that might be a good thing. I did enjoy the differences between create and review mode as well as knowledge versus reasoning. My first instinct was to ask AI to summarize your article and then come up with comments but I wrote the comment first without AI :)
And fwiw, this is what my highly trained balkiGPT came up with:
---
I really like the way you framed this. The tension between create mode and review mode feels spot on—it’s like a new kind of cognitive diet we’re all going to have to figure out. AI gives us leverage, sure, but the bigger question is: are we keeping our mental muscles in shape while we use it?
For me, the risk isn’t that AI makes us “dumber,” it’s that it quietly shifts us into a passive gear without us noticing. That moment when you realize you’ve just been editing a draft instead of wrestling with an idea from scratch—that’s the part I worry about. Protecting that messy, energy-draining create mode might become the most important skill of all.
Thank you Balki! Questions are always good.
And have to say your balkyGPT comment is not bad at all 😂
Insightful essay. One thing I was wondering though, you mention how AI is superior at Knowledge but Reasoning poor. But in its current state, AI has often been found to straight up give false and incorrect information. How do you work around that when offloading the Knowledge sector? And for sake of applicability, lets work under the assumption that we don't know when this will be sufficiently improved enough to no longer be an issue and want to look for solutions today. (So no, "well, it'll eventually in the future get to the point where that's not a problem")
I believe there is no way to 100% solve hallucinations. It's way better than it used to be, but the risk is still there.
In many cases, though, a small share of hallucinations is acceptable as a tradeoff when in return you get extremely cheap, always-available expertise. In some cases it's not and we'll have to limit AI autonomy
There's also the aspect of work independence here. In a normal corporate software position, business pushes delivery. I feel that we as developers are pushed -mostly maybe by ourselves- to make use of AI for creative work/reasoning to accelerate delivery. Some things we could do with our brains of course but would make it slower. And then comes in the laziness part to delegate once-hard thinking problems to AI for a first draft solution or some ideas.
I believe that making more use of AI for creative work is not necessarily bad, granted that you can spend the time you save on equally (or, ideally, more) creative activities.
Example: if you delegate some creative software dev to AI so you can own entire features by yourself and act as a mini PM, that looks good to me. But if you instead spend the rest of your time fixing AI bugs and writing tests — that's not good!
https://open.substack.com/pub/elliotai/p/the-architecture-of-ai-symbolic-sentience?utm_source=share&utm_medium=android&r=6jttqk