I've been using AI for a year now, and what changed for me was this: iām no longer amazed by what it could produce and started noticing what it got wrong.
And what it got wrong, most of the time, was the judgment call. It's technically fine, but it still doesn't know what to leave out.
That's the thing nobody talks about ā editing is harder than generating. Knowing what to cut requires you to hold the whole in your head, understand the reader, and have opinions about what matters. That's not something you can delegate. It only gets cleaner the more work you've actually done.
The line "AI has read about those things. You lived them" is the one that hits hardest for me. I work with people in career transitions, and the gap between knowing a concept and having been through something is exactly what they struggle to articulate ā and exactly what clients keep paying for. Where I'd add a complication: most people don't know what their taste actually is until they're forced to explain it out loud. That's the real work ā not using AI better, but first excavating what you actually know and why you know it.
"Take responsibility for outcomes that matter" .... yes, Mr. Hassid, your statement captures my Northstar for AI and what it means to be a professional focused on purpose-driven service. I will be subscribing to your substack in the morning to thank you for the incredible collection of guides I have cataloged over the past month from your posts on LinkedIn. Best, Scott
Great piece, thank you! It's funny. I just did a presentation last week with the same argument, about judgment and how it's build through gaining experience with solving messy problems. I thought I came up with this all on my own but of course I did not. I was sounding off my ideas and reflections to Claude. It was giving me articles to read. I am curious how our extensive reliance on AI (even assuming human is in the loop throughout) will impact our collective creativity and intellectual output. Are we getting homogenized a bit?
And there is the other angle: in many roles, 70% of work were tasks that can now be performed by or with AI and only 30% was taste and judgement. This is flipping now to 70% of day to day being judgment. Quite intense when you think about it.
Sure. I think it is very apt to compare the costs of performing cognitive tasks using AI versus human intelligenceāhighlighting that you will be left out of the game if you don't utilize itābut I haven't been able to verify the figure of 6K that you cited. Authoritative sources only report cost reductions reaching 900 (Epoch AIāthough that is already substantial) or up to 280 (Stanford Institute for Human-Centered AI).
Maybe once you could make a demo of how you use your AI to make this substack feel so human and real. This will be a great demo for us to understand it in real life!
Can you point me to one of your articles that will show me how to get started from the ground up with Cowork. I regularly use claude but donāt feel like Iām getting the most from it. And Iām getting a new computer next week which will be a fresh start
Iāve consistently used ChatGPT with all of my settings in place and chat history built up for years now. How would you recommend using Claude and CoWork without having to spend years of it learning truly who you are, your beliefs, your personal experiences, and even how you operate and think? These things are not something you can just put into a prompt. You could say Iāve developed a relationship with my ChatGPT, whom I can argue and disagree with, yet it still seems to understand me and why I think the way I do.
"What happens when the thing you used to sell becomes almost free?"
This is the question most consultants haven't asked themselves yet (even though their clients already have.)
Most are still selling the deliverable. The strategy doc. The funnel build. The audit.
All things AI does "fine" for $20 a month now.
The ones still charging real money are selling something else entirely - judgment under pressure, taste, the willingness to push back. None of which fits on a Fiverr listing.
what survives is the part that was always invisible on the invoice: knowing which version of the strategy is wrong even though it looks right, telling the CEO what he doesn't want to hear, catching the assumptions.
Right, and here's the uncomfortable bit: that invisible part was always what you were paying for. The deliverable was just the receipt. AI didn't change the value, it stripped away the prop we used to justify it. The whole profession now has to learn to sell the thing the document used to stand in for.
Jevons Paradox is the right frame, but it cuts deeper than you take it. Salesforce reported $800 million in Agentforce ARR, 29,000 deals, nearly 20 trillion tokens served this year. https://thesynthesisai.substack.com/p/the-efficiency-trap. They executed the "taste plus systems" playbook at corporate scale and the market still punished them. Cheap intelligence compresses entire categories, including the one where you package and sell judgment. Building taste into reusable context files is solid advice right up until everyone has done it, at which point taste-as-a-file becomes the new average.
I think one of the biggest shifts happening underneath all of this is that information production is becoming commoditised far faster than judgement, coherence, and authorship.
AI can generate enormous volumes of āgood enoughā output now. What still remains difficult is preserving the thinking structure underneath the output itself.
The founders and experts who stay differentiated will probably be the ones who can externalise judgement, standards, philosophy, decision logic, and contextual understanding into governed systems without flattening themselves into generic language in the process.
Thatās a very different problem from simply learning prompts or increasing productivity.
And I suspect most businesses still havenāt fully realised the distinction yet.
Thatās such an amazing reminder and quite a solid take focusing on how humans are being tagged to the AI cost, deteriorating the original impact they bring to deliver quality outcomes!!
I've been using AI for a year now, and what changed for me was this: iām no longer amazed by what it could produce and started noticing what it got wrong.
And what it got wrong, most of the time, was the judgment call. It's technically fine, but it still doesn't know what to leave out.
So I agree with you, Ruben, humans will win.
its only right: be critical, be selective, be the one to steer it
AI produces and has zero responsibility for the outcome
That's the thing nobody talks about ā editing is harder than generating. Knowing what to cut requires you to hold the whole in your head, understand the reader, and have opinions about what matters. That's not something you can delegate. It only gets cleaner the more work you've actually done.
100% agree - I've learned to pause, reflect, extract and refine. That's a new mode of thinking I've developed and it works vs getting bogged down.
Each pass sharpens the lens.
Thnx for an insightful SS @Ruben. It's knowing when, where, how and why you're using the tool you're using to fulfill a brief right?
yes, feeding it enough of yourself that the output doesn't look like everyone else's output from the same tool :)
New to the experiment. Arenāt we in the end training AI to be better at mimicking human thought every time we input our lived experience into AI?
The line "AI has read about those things. You lived them" is the one that hits hardest for me. I work with people in career transitions, and the gap between knowing a concept and having been through something is exactly what they struggle to articulate ā and exactly what clients keep paying for. Where I'd add a complication: most people don't know what their taste actually is until they're forced to explain it out loud. That's the real work ā not using AI better, but first excavating what you actually know and why you know it.
iāve shared an interview prompt before, 100 questions to extract the taste :) it takes a good 2 hours maximum
you can find it here if youāre interested: https://ruben.substack.com/p/youre-just-a-text-file
"Take responsibility for outcomes that matter" .... yes, Mr. Hassid, your statement captures my Northstar for AI and what it means to be a professional focused on purpose-driven service. I will be subscribing to your substack in the morning to thank you for the incredible collection of guides I have cataloged over the past month from your posts on LinkedIn. Best, Scott
appreciate it, Scott :) is there any guide youāre looking for? maybe i can help you locate the newsletter
Great piece, thank you! It's funny. I just did a presentation last week with the same argument, about judgment and how it's build through gaining experience with solving messy problems. I thought I came up with this all on my own but of course I did not. I was sounding off my ideas and reflections to Claude. It was giving me articles to read. I am curious how our extensive reliance on AI (even assuming human is in the loop throughout) will impact our collective creativity and intellectual output. Are we getting homogenized a bit?
i see it in newsletters, in decks, in LinkedIn posts, comments, everyone literally sounds the same now
people let AI homogenize them - they approve the first output, they stop at good enough, they lose the reps that build taste in the first place
I also bet on judgment and I call the change: the 70/30 flip, where 70% of our work is judgment. https://danaaonofriesei.substack.com/p/the-7030-flip-what-ai-is-actually?r=1wmkip&utm_medium=ios
i agree - AI gives the 70%, our taste gives another 30% :)
And there is the other angle: in many roles, 70% of work were tasks that can now be performed by or with AI and only 30% was taste and judgement. This is flipping now to 70% of day to day being judgment. Quite intense when you think about it.
Brilliant!
appreciate it :) what made it brilliant, if i may ask?
Sure. I think it is very apt to compare the costs of performing cognitive tasks using AI versus human intelligenceāhighlighting that you will be left out of the game if you don't utilize itābut I haven't been able to verify the figure of 6K that you cited. Authoritative sources only report cost reductions reaching 900 (Epoch AIāthough that is already substantial) or up to 280 (Stanford Institute for Human-Centered AI).
Love this honest article. Keep this coming.
Maybe once you could make a demo of how you use your AI to make this substack feel so human and real. This will be a great demo for us to understand it in real life!
thank you, iāll definitely keep this in mind :)
I love this framing, thank you for sharing.
iām glad you loved it - what are you working on right now? :)
This is an important article. Thanks for this.
youāre welcome :) i figured it would be nice to have a little breather from how-to guides
This article is so reassuring in knowing I'm not as far behind as I thought while also being inspired to take it further. Bookmarking this!
thereās nowhere to go but up - 20 mins a day, thatās all it takes :)
Can you point me to one of your articles that will show me how to get started from the ground up with Cowork. I regularly use claude but donāt feel like Iām getting the most from it. And Iām getting a new computer next week which will be a fresh start
sure - hereās my complete Cowork guide: https://ruben.substack.com/p/claude-cowork-20
Amazing thanks š
Iāve consistently used ChatGPT with all of my settings in place and chat history built up for years now. How would you recommend using Claude and CoWork without having to spend years of it learning truly who you are, your beliefs, your personal experiences, and even how you operate and think? These things are not something you can just put into a prompt. You could say Iāve developed a relationship with my ChatGPT, whom I can argue and disagree with, yet it still seems to understand me and why I think the way I do.
do a taste interview with Claude, i talked about it in this blog https://ruben.substack.com/p/youre-just-a-text-file everything you know can be extracted into one file :)
P.S: this file is the core of all of my task, its how Claude knows how to write and sound like me
Thank youā¦ Iāve nicknamed you THE AI Genius! I honestly have learned more from you than anyone on AI. The way you present it is absolutely amazing! š¤©
"What happens when the thing you used to sell becomes almost free?"
This is the question most consultants haven't asked themselves yet (even though their clients already have.)
Most are still selling the deliverable. The strategy doc. The funnel build. The audit.
All things AI does "fine" for $20 a month now.
The ones still charging real money are selling something else entirely - judgment under pressure, taste, the willingness to push back. None of which fits on a Fiverr listing.
what survives is the part that was always invisible on the invoice: knowing which version of the strategy is wrong even though it looks right, telling the CEO what he doesn't want to hear, catching the assumptions.
Right, and here's the uncomfortable bit: that invisible part was always what you were paying for. The deliverable was just the receipt. AI didn't change the value, it stripped away the prop we used to justify it. The whole profession now has to learn to sell the thing the document used to stand in for.
Jevons Paradox is the right frame, but it cuts deeper than you take it. Salesforce reported $800 million in Agentforce ARR, 29,000 deals, nearly 20 trillion tokens served this year. https://thesynthesisai.substack.com/p/the-efficiency-trap. They executed the "taste plus systems" playbook at corporate scale and the market still punished them. Cheap intelligence compresses entire categories, including the one where you package and sell judgment. Building taste into reusable context files is solid advice right up until everyone has done it, at which point taste-as-a-file becomes the new average.
the file is just the floor
what salesforce can't sell you is the bad calls that taught you what to ignore
I think one of the biggest shifts happening underneath all of this is that information production is becoming commoditised far faster than judgement, coherence, and authorship.
AI can generate enormous volumes of āgood enoughā output now. What still remains difficult is preserving the thinking structure underneath the output itself.
The founders and experts who stay differentiated will probably be the ones who can externalise judgement, standards, philosophy, decision logic, and contextual understanding into governed systems without flattening themselves into generic language in the process.
Thatās a very different problem from simply learning prompts or increasing productivity.
And I suspect most businesses still havenāt fully realised the distinction yet.
the thinking lived in people's heads because keeping people around was easier
now the output is free and the thinking still isn't written down anywhere
Thatās such an amazing reminder and quite a solid take focusing on how humans are being tagged to the AI cost, deteriorating the original impact they bring to deliver quality outcomes!!
quality outcomes still depend on a human knowing what quality looks like
judgment is what stays expensive