Learning Library

← Back to Library

Taste: The Secret Skill for AI Success

Key Points

  • Success with AI in 2025 hinges on cultivating “taste”—the gut‑level sense of what’s right, valuable, and improvable—rather than just technical prompt‑engineering skills.
  • Taste is often seen as elitist (fashion, fine dining) but it’s actually a universal, experience‑based judgment that anyone can develop and apply across domains.
  • It originates from accumulated experience that forms strong, intuitive opinions, whether in hobbies like fantasy football, fashion, books, or any field you care about.
  • Developing taste means deliberately exposing yourself to diverse examples, reflecting on what feels right or wrong, and sharpening that internal compass as a practical skill.
  • Leveraging taste helps you navigate AI tools, teach others (including kids), and make strategic career choices by recognizing and acting on what truly resonates and adds value.

Full Transcript

# Taste: The Secret Skill for AI Success **Source:** [https://www.youtube.com/watch?v=A_Lv0Ze272g](https://www.youtube.com/watch?v=A_Lv0Ze272g) **Duration:** 00:15:45 ## Summary - Success with AI in 2025 hinges on cultivating “taste”—the gut‑level sense of what’s right, valuable, and improvable—rather than just technical prompt‑engineering skills. - Taste is often seen as elitist (fashion, fine dining) but it’s actually a universal, experience‑based judgment that anyone can develop and apply across domains. - It originates from accumulated experience that forms strong, intuitive opinions, whether in hobbies like fantasy football, fashion, books, or any field you care about. - Developing taste means deliberately exposing yourself to diverse examples, reflecting on what feels right or wrong, and sharpening that internal compass as a practical skill. - Leveraging taste helps you navigate AI tools, teach others (including kids), and make strategic career choices by recognizing and acting on what truly resonates and adds value. ## Sections - [00:00:00](https://www.youtube.com/watch?v=A_Lv0Ze272g&t=0s) **AI Success Requires Cultivated Taste** - The speaker argues that developing a refined sense of taste—traditionally seen as an elite, aesthetic skill—is essential for effectively collaborating with increasingly sophisticated AI systems and teaching this ability to others. - [00:03:06](https://www.youtube.com/watch?v=A_Lv0Ze272g&t=186s) **Evolving Taste Amid AI Automation** - The speaker explains how personal tastes and expertise evolve over a career, and how, as AI handles routine tasks, human judgment and adaptability become the valuable “taste” that remains. - [00:06:36](https://www.youtube.com/watch?v=A_Lv0Ze272g&t=396s) **AI Embedding to Capture Work Time** - The speaker argues that platforms like Claude, OpenAI, and Gemini are integrating AI directly into work workflows to monopolize users’ attention, while emphasizing that true career advancement still relies on human expertise, personal taste, and embodied experience that AI cannot replicate. - [00:10:03](https://www.youtube.com/watch?v=A_Lv0Ze272g&t=603s) **Embedding Human Taste in AI Interactions** - The speaker argues for giving language models precise, preference‑driven feedback—critiquing style, accuracy, and formatting—to let users shape outputs as AI systems become increasingly intelligent. - [00:14:12](https://www.youtube.com/watch?v=A_Lv0Ze272g&t=852s) **Cultivating Taste in AI** - The speaker urges users to apply personal judgment—“taste”—when interacting with AI tools, assessing what works, discarding what doesn’t, and treating AI as a flexible toolkit amid rapidly evolving model capabilities. ## Full Transcript
0:00I think we don't talk enough about what 0:02it really takes to be successful in 2025 0:05with AI. And I know, right, that's me 0:07saying that and I do prompts and this 0:08and that. But one of the things that 0:10hasn't really been mentioned that I 0:12haven't really talked a ton about that I 0:14see a fair bit of in actual AI 0:18interactions is the necessity and 0:20importance of taste. And so when we 0:23think of taste, like when I think of 0:24taste, I think, oh, someone who has 0:26fashion sense, right? someone who goes 0:28to the great restaurant and can order 0:29the right sushi and the French and food 0:31and this and that. And so people feel 0:33like it's off-putting. It feels very 0:35elitist. It feels like I couldn't have 0:38taste right here. I am in my little 0:39t-shirt like how could I have good 0:40taste? I want to make it accessible for 0:42you today because I think it is one of 0:44the skills that we most need not just as 0:49people who work with AI but as humans in 0:53the age of AI where we have to assume 0:56the models are getting better. The 0:58models are getting better and better and 0:59better and better and we have to figure 1:01out how to work with them as they scale 1:04in intelligence. We have to figure out 1:05if we're parents how to teach our kids 1:07or if we're teachers how to teach our 1:08kids about them in ways that make sense. 1:10Your secret weapon in all of that is 1:13taste. 1:15It's taste. How do you develop taste? I 1:18look I have seen the articles. If you 1:19Google for AI and taste, you see piles 1:22and piles of AIdriven crap for lack of a 1:25better term, right? Like just AI dril 1:27about the importance of taste and value 1:29of it. So what what matters is that you 1:32actually understand where taste comes 1:35from inside you. how you develop and 1:38sharpen it as a skill and then how you 1:41apply it. And that's really what I want 1:42to work through. I'm all about 1:44practical. So, first, where does where 1:46does taste come from inside of you? 1:48Taste is your gut knows best. Taste is 1:52the sense inside of you that something 1:55is right or something is wrong. Taste is 1:58the sense that something could be 2:00better. If you want to break it up even 2:02further, taste is where you start to 2:05accumulate enough experience in a 2:08particular area that you begin to have 2:10strong opinions. And this can happen 2:12outside of work. Like there are people 2:13who play fantasy football and they have 2:16very strong opinions on draft order and 2:18which players they're going to pick. 2:19They have taste in that area. There are 2:22people who have taste in widely known 2:26areas where taste is considered a skill 2:29set like clothes. I talked about 2:32clothes. That is considered an 2:33acceptable place for taste. And what I 2:36want to suggest to you is that taste is 2:38not just a skill in places where people 2:40honor it. Taste is actually something we 2:42all do every day. I would like to say 2:44that I have taste in books. I do a lot 2:46of book shopping. I do a lot of book 2:48collecting. Do reading. Taste is 2:51something that matters to me in that 2:52area. Not for everybody, but for me. And 2:55so when you're thinking about career 2:56pathing, when you're thinking about what 2:58you're good at, when you're thinking 2:59about what persists in the age of AI, 3:02think about it from a taste perspective. 3:04Think about what are the areas where you 3:06have experience and you can lean on 3:08them. And those can change, by the way, 3:10over the course of a career. If I go 3:12back decades in my career, the things 3:14that I thought I had taste about or had 3:16opinions about have changed 3:19unrecognizably 3:20since then, I have not had taste about 3:23any one thing for the decades I've been 3:26working. But my tastes have evolved. My 3:28I have transitioned into new areas as I 3:31have followed my curiosity. Why does 3:33this matter so much now? Why are we 3:36talking about this? Because taste is not 3:38a new skill. I just talked about this. 3:39I've seen this in people who were more 3:41senior than me long before AI became a 3:43thing. People had taste and used taste 3:45at work. We just didn't talk about it as 3:48such. It's a thing now because taste we 3:52are discovering is what is left when AI 3:56can do a lot of the grunt work. So, as 3:59an example from this week, when Claude 4:01can produce a workable discounted cash 4:04flow sheet, when Claude can produce a 4:07PDF that looks okay or a PowerPoint that 4:09looks okay in like one shot with just a 4:12little bit of context, what's left? 4:14What's left in the work stack that we 4:17do? And a lot of people look at this and 4:18they start to write those like panicky 4:20internet articles. They throw up their 4:21hands like, "Oh my god, this is the end 4:23of all things." No, no. Actually, this 4:26is a chance for us to transition. And 4:29we're very good at this. Humans are 4:30flexible tool users. I think that's one 4:32of the best definitions of humans I've 4:33ever heard. We're embodied flexible tool 4:35users. And we can flexibly use a 4:38different part of our skill set. And 4:40whereas before when you were producing 4:43workable value in the '9s in the 2000s a 4:46lot of the value was in time spent with 4:49other people physically and in work 4:52product that you produced by typing like 4:54it was physical creation of information 4:56and we were called knowledge workers was 4:58very fancy was the future of the world 5:00for about 10 years well that's changed 5:03that's not where the leverage is. That's 5:04not where the juice is anymore at work 5:07because I will tell you Chad GPT5 is 5:10faster at typing than anybody I know. 5:13And it's also got the option to generate 5:1520 ideas in the time it would take me to 5:17generate one. And because of the 5:19internet and high bandwidth connections, 5:21we can now do digital collaborations. We 5:23don't have to physically be in the same 5:24space, which by the way further enables 5:26AI collaboration because AI is 5:28disembodied. So if it's arriving over 5:30the internet, it sort of feels like a 5:31colleague, right? It pops up in Slack. 5:33You can have AI and Slack. This is all 5:35shifting our leverage as people. Our 5:38skill sets, the value of them is 5:41changing. And that's what drives a lot 5:42of the stress is because we have assumed 5:44if you're building your career, you 5:45assume like I read this book on product 5:48management. I'm going to get it out. I'm 5:49going to open it and like as soon as I 5:51get it out and I'm going to know the 5:53things I need to know or I did the 5:55projects. I I was the project man the 5:57technical project manager and I drove 5:59these big projects for multi-million 6:01dollar companies and I have the 6:02experience of managing 200 stakeholders 6:04in a room this and that right those are 6:06the things that we think give us the 6:08value and what I'm starting to realize 6:10is that taste is a word that has teeth 6:13that allows us to think about a wide 6:15range of related skill sets that AI is 6:18not seeming to take over like if you 6:21look at the strategy for model makers 6:24right now they are going after time 6:26spent at work. They're going after the 6:28work stack. So Claude wants you to be 6:31thinking in claude and then producing 6:33artifacts in Excel and Word and so on. 6:35Claude wants you to be thinking in 6:36claude at work and do team projects. 6:39That was another feature they 6:40emphasized. Open AAI wants you to be 6:42thinking in code in codeex. Similarly 6:45with cla code etc. Gemini wants you 6:47thinking in images in nano banana. And 6:50so I think of it as they are trying to 6:52capture more of your time the way 6:53Facebook tried to capture more of your 6:55time in the 2000. Tik Tok tried to 6:57capture more of your time in the 2020s. 7:00It's like what can we do to keep you 7:02engaged and keep you focused and in 7:03their case it's like intelligence in the 7:05work stack feels like a way to do that. 7:07Fine. What that means is your value lies 7:10in your ability to leverage those work 7:12primitives, the AI things that that can 7:14like produce that in ways that reflect 7:17your taste, in ways that reflect your 7:21expertise. And this can sometimes feel 7:23like it's it's a nod to gray hairs, 7:25right? It's it's a nod to the ability to 7:27have deep expertise. I have seen people 7:30who were just getting started in their 7:32career who have taste. They have figured 7:34out a particular corner that they are 7:36passionate about. They figured out 7:38something they have experience in and 7:39they figured out how to have taste in 7:41that area and they insist on it. That is 7:43a recipe for rapid career growth whether 7:46you're starting out or whether you're 7:47experienced because 7:49AI is not good at it. AI is really, 7:52really not good at it. And part of why 7:54it's not good at it is that AI isn't 7:57embodied. AI doesn't develop the kind of 8:00deep, nuanced, metabolized expertise 8:03that we get from living in society for 8:06decades before we become adults and go 8:09to work. There's no substitute for that. 8:11And so that is part of how we shape 8:13taste as creatures and that is part of 8:15what we bring 8:18that AI has trouble mimicking. And so if 8:21you've ever looked at something and you 8:23look at the work product that AI gives 8:24you and you're like it feels hollow. It 8:27feels artificial. I can't put my finger 8:29on it, but it just doesn't feel right. 8:32That's your taste speaking. That's your 8:34taste. Listen to it. And a lot of people 8:36then take that and they say, "Well, I'm 8:38going to just like throw out AI fluff 8:39and I'm going to not do this and I'm 8:40just going to have a pure human created 8:42world." It's like, I don't think that's 8:44the answer. I think the answer is having 8:46the taste to demand useful work. And 8:50this is important because increasingly 8:53our jobs are going to depend on our 8:56ability to demand useful, not perfect, 9:00but useful work. And so think about it. 9:02If your work is really defining what 9:05matters and what's high taste and what's 9:07good quality, and that's your job, can 9:09you do it? Do you get excited about 9:11doing that? Are you okay pushing back 9:14and not being overly differential with 9:16the model? One of the things I've seen I 9:17get privileged to see a lot of AI 9:19conversations because well people share 9:21them with me and I get on calls etc etc. 9:23A lot of the time people are overly 9:26differential with the model which I mean 9:28by that is they give the model too much 9:31rope. They give the model a lot of 9:33space. Now that shows up in prompting as 9:35you're not giving the model the guidance 9:36it needs. It shows up in building 9:38systems where you just talk to Claude or 9:40talk to chat GPT or talk to Gemini and 9:42you say design the system for me and 9:44then you just sort of trust it. But it 9:46also shows up in more individual 9:47interactions where you talk to the model 9:49and if the model gives you something, 9:51you just assume it's probably right. And 9:53if you're a skeptic and you see one 9:54wrong thing, then you just throw it all 9:56out. And I see those two sides flip 9:58really easily. They feel related. It 10:00feels like the same emotional response. 10:01It's either I'm going to trust the 10:03giving being to give me a good answer or 10:05I'm going to throw it out. Well, what if 10:07we just brought our taste into model 10:09interactions? What if when we were 10:11chatting with Chad GPT5 or Claude or 10:13your model of choice, we said, "I think 10:16you can do better. I don't like this 10:18particular piece. I like that particular 10:19piece." Like, we can be very specific, 10:21right? Like, I like the numbering you 10:22used here. Your phrasing is 10:24overdramatic. Or, I really, really don't 10:26appreciate it when you make up numbers. 10:28Two of these numbers are made up. The 10:30other 16 are, "Please tell me when you 10:33need me to get you more information." 10:35Or, "Please go and do research and come 10:36back and get the answers." We have a 10:39pallet of responses to work with and the 10:41choices that we make to sort of have 10:43those conversations to prompt are in 10:46many ways driven by taste by our 10:49willingness to have a gut opinion to 10:52have have a sense of like this could be 10:55better. And I emphasize that because I 10:58don't think we talk enough about how 11:00it's formed. So, I've talked about 11:01domain experience and being sort of in 11:03your in your work experience over time, 11:06understanding all the nuances or if 11:08you're new at work, understanding what 11:10you're passionate about and leaning on 11:12that for taste. And I've talked a little 11:14bit about how it applies. But if we look 11:16ahead, the third thing I want to call 11:19out is that taste is partly hard right 11:21now because models are gaining 11:24intelligence 11:25so so quickly. And so models are getting 11:28smarter and smarter and smarter and 11:30smarter on a time frame of months. We're 11:33not used to that. This is a new 11:35experience for the species. And if 11:37you're trying to apply taste, it feels 11:39jarring, right? It feels like, okay, but 11:41this new model, I have to get used to 11:43this model and this other new model 11:44comes along and very tired. 11:48The the simplest way to think about it 11:50is that we are moving into a world where 11:54you need to depend even more on taste, 11:59not less. And so when we were first 12:02working with Chat GPT two years ago, I 12:04think taste mattered less because chat 12:07GPT could do less of the workday. Claude 12:10wasn't even there. You you just had less 12:12options to do the whole workday in an 12:14AI. Now models have come along very 12:16quickly. more of the workday is there, 12:18more of our personal lives is there, 12:19health decisions are there, all of that 12:20stuff, and you need taste more. In other 12:24words, you could trust your own 12:26instincts cuz you were producing more of 12:27the work and it was all inside your head 12:29and like you had the meeting with 12:31colleagues, you metabolized it the way 12:32you usually do. You had a gut instinct. 12:34It came from the compost pile inside 12:36your brain and you were like, "This is 12:37the right way to go and then you wrote 12:38the thing." All of that happened in your 12:40head. Not much of it happened in Chad 12:42GPT. And gradually we've shifted that. 12:44So now more and more the meeting notes 12:46are in chat GPT you send them out but 12:48maybe you check them and the project 12:50plan might have come partly from Chad 12:51GPT partly from you and now increasingly 12:54the strategic analysis and like the way 12:55forward doing some thinking in an AI and 12:58some thinking not. This is why taste 12:59matters more because the thing that that 13:01still runs the central processing unit 13:03for all of this is still your taste. it 13:05is still your ability to say no thank 13:08you right like I don't think that's 13:09correct and not to say it in a binary 13:12way where you get frustrated and throw 13:13the model out but to actually recognize 13:15that in some ways the model is better I 13:17am fully convinced that GPT5 pro has 13:21thinking abilities in specific domains 13:24that vastly outpace me and that is true 13:26for most people I know if you're von 13:28Newman raise your hand obviously that's 13:30not true physics people will have will 13:31have a good laugh over that one but but 13:33for the most part GPT T5 Pro is really 13:36really smart and you have to take almost 13:39a like I talk about it as talking to the 13:40oracle. You sort of put your little prop 13:42together, you wait several minutes, it 13:43comes back with a response and you have 13:45to like interpret it, understand what it 13:46means. You still have to have taste. I 13:49have looked at GPT5 Pro responses and I 13:52have said, I see where you got that. I 13:54think it's correct given the inputs I 13:56gave you, but what I learned from this 13:57interaction is that you don't know the 13:59context I know in my head and that's why 14:00this feels off. And so instead of saying 14:03I should trust GPT5 Pro because it's 14:05smarter than me, I say good strategic 14:07analysis, I'm learning that I need to 14:09give you different kinds of inputs. 14:10That's a much more nuanced response. It 14:12doesn't denigrate GPT5 Pro is unhelpful. 14:14It's just understanding what the model 14:16can do and what it can't do and having 14:18some taste about it. And so as models 14:20get smarter, GPT6 may come along before 14:22the end of the year. You never know. 14:23There's been some hints. You need to 14:25lean in on taste. Lean in on the gut 14:30feeling. Lean in on your your ability to 14:32say this bit is good, this bit is not 14:35good. Look at all the ways that improves 14:38your experience with AI. Look at how it 14:40helps your prompting if you insist on 14:42taste. Look at how it helps your 14:44multi-turn conversations if you insist 14:46on taste. Look at how it helps you when 14:48you're reviewing other people's work 14:50that may have been assisted by AI. If 14:52you insist on taste, this doesn't give 14:54you a license to be annoying and say no 14:57AI. I keep I keep calling that out. AI 14:59is like a toolkit. You can go in and 15:01rate it for tools and come back and keep 15:04what you like, throw out what you don't 15:06on your tastes. I hope this has been 15:08helpful. I think this is one of the 15:09skills that I see clickbait on, but I 15:12don't see us having a meaningful 15:14conversation on where it comes from, how 15:16it works for early career folks, how it 15:17works for more mid and senior career 15:19folks, and above all, how we handle this 15:21in an age when intelligence is 15:23accelerating. And so, it feels like our 15:24relationship is evolving as we talk 15:26about it. I hope this has been a good 15:28sort of starter. Drop something in the 15:29comments. Let me know what you think. 15:31We're all learning to do taste together. 15:33We're flexible tool users. We're 15:34learning this skill that suddenly has, I 15:37would argue, 100x more value than it had 15:39in 2000. Uh so good luck out there. You 15:41also you also are a taste maker.