Disagree with the overall argument. Human effort is still a moat. I've been spending the past couple of months creating a codebase that is almost entirely AI-generated. I've gotten way further than I would have otherwise at this pace, but it was still a lot of effort, and I still wasted time going down rabbit holes on features that didn't work out.
There's some truth in there that judgement is as important as ever, though I'm not sure I'd call it taste. I'm finding that you have to have an extremely clear product vision, along with an extremely clear language used to describe that product, for AI to be used effectively. Know your terms, know how you want your features to be split up into modules, know what you want the interfaces of those modules to be.
Without the above, you run into the same issue devs would run into before AI - the codebase becomes an incoherent mess, and even AI can't untangle it because the confusion gets embedded into its own context.
I feel like you're pretty strongly agreeing that taste is important: " I'm finding that you have to have an extremely clear product vision...""
Clear production vision that you're building the right thing in the right way -- this involves a lot of taste to get right. Good PMs have this. Good enginers have this. Visionary leaders have this....
The execution of using AI to generate the code and other artifacts, is a matter of skill. But without the taste that you're building the right thing, with the right features, in a revolutionary way that will be delightful to use....
I've looked at three non-engineer vibe-coded businesses in the past month, and can tell that without taste, they're building a pretty mediocre product at best. The founders don't see it yet. And like the article says, they're just setting themselves up for mediocrity. I think any really good PM would be able to improve all these apps I looked at almost immediately.
> I've gotten way further than I would have otherwise at this pace, but it was still a lot of effort, and I still wasted time going down rabbit holes on features that didn't work out.
By the time I'm done learning about the structure of the code that AI wrote, and reviewing it for correctness and completeness, it seems to be as much effort as if I had just written it myself. And I fear that will continue to be the reality until AIs can be trusted.
You make a really salient point about having a clear vision and using clear language. Patrick Zgambo says that working with AI is spellcasting; you just need to know the magic words. The more I work with AI tools, the more I agree.
Now, figuring out those words? That's the hard part.
Today: Ask AI to "do the thing", manual review because don't trust the AI
Tomorrow: Ask AI to "do the thing"
I'm just getting started on my AI journey. It didn't take long before I upgraded from the $17 a month claude plan to the $100 a month plan and I can see myself picking the $200 a month plan soon. This is for hobby projects.
At the moment I'm reviewing most of the code for what I'm working on, and I have tests and review those too. But, seeing how good it is (sometimes), I can imagine a future where the AI itself has both the tech chops and the taste and I can just say "Maybe me an app to edit photos" and it will spit out a user friendly clone of photoshop with good UX.
We already kind of see this with music - it's able to spit out "Bangers". How long until it can spit out hit rom-coms, crime shows, recipes, apps? I don't think the answer is "never". I think more likely the answer is in N years where N is probably a single digit.
It's leaning in a good direction, but the author clearly lacks the language and understanding to articulate the actual problem, or a solution. They simply dont know what they dont know.
> Human effort is still a moat.
Also slightly off the mark. If I sat one down with all the equipment and supplies to make a pair of pants, the majority of you (by a massive margin) are going to produce a terrible pair of pants.
Thats not due to lack of effort, rather lack of skill.
> judgement is as important as ever,
Not important, critical. And it is a product of skill and experience.
Usability (a word often unused), cost, utility, are all the things that people want in a product. Reliability is a requirement: to quote the social network "we dont crash". And if you want to keep pace, maintainability.
> issue devs would run into before AI - the codebase becomes an incoherent mess
The big ball of mud (https://www.laputan.org/mud/ ) is 27 years old, and still applies. But all code bases have a tendency to acquire cruft (from edge cases) that don't have good in line explanations, that lack durable artifacts. Find me an old code base and I bet you that we can find a comment referencing a bug number in a system that no longer exists.
We might as an industry need to be honest that we need to be better librarians and archivists as well.
That having been said, the article should get credit, it is at least trying to start to have the conversations that we should be having and are not.
I think you're missing the point. Effort is a moat now because centaurs (human+AI) still beat AIs, but that gap gets smaller every year (and will ostensibly be closed).
The goal is to replicate human labor, and they're closing that gap. Once they do (maybe decades, but probably will happen), then only that "special something" will remain. Taste, vision... We shall all become Rick Rubins.
> Without the above, you run into the same issue devs would run into before AI - the codebase becomes an incoherent mess, and even AI can't untangle it because the confusion gets embedded into its own context.
We have a term for this and it is called "Comprehension Debt" [0] [1].
I'm continually fascinated by the huge differences in individual ability to produce successful results with AI. I always assumed that one of the benefits of AI was "anyone can do this". Then I realized a lot of people I interact with don't really understand the problem they're trying to solve all that well, and have some irrational belief that they can get AI to brute force their way to a solution.
For me I don't even use the more powerful models (just Sonnet 4.6) and have yet to have a project not come out fairly successful in a short period of time. This includes graded live coding examples for interviews, so there is at least some objective measurement that these are functional.
Strangely I find traditional software engineers, especially experienced ones, are generally the worst at achieving success. They often treat working with an agent too much like software engineering and end up building bad software rather than useful solutions to the core problem.
This seems more telling on the artist who, I guess, believes that if you have taste in any field, it will manifest itself as wearing stylish clothes. I see their most recent blog post is analyzing luxury brands, so I think I'm on point here.
> One of the most useful things about AI is also one of the most humbling: it reveals how clear your own judgment actually is. If your critique stays vague, your taste is still underdeveloped. If your critique becomes precise, your judgment is stronger than the model output. You can then use the model well instead of being led by it.
Something I find that teams get wrong with agentic coding: they start by reverse engineering docs from an existing codebase.
This is a mistake.
Instead, the right train of thought is: "what would perfect code look like?" and then meticulously describe to the LLM what "perfect" is to shape every line that gets generated.
This exercise is hard for some folks to grasp because they've never thought much about what well-constructed code or architectures looks like; they have no "taste" and thus no ability to precisely dictate the framework for "perfect" (yes, there is some subjectivity that reflects taste).
> Instead, the right train of thought is: "what would perfect code look like?" and then meticulously describe to the LLM what "perfect" is to shape every line that gets generated.
I think this goes against what a lot of developers want AI to be (not me, to be clear).
I've worked in too many large codebases where no one can point to any _single file or class_ and label it "correct," ("the right way") yet management is amazed when the lack of a "North Star" means the codebase is full of overlapping, piecemeal patterns that are lucky to work together at all.
> Instead, the right train of thought is: "what would perfect code look like?"
That's the classic 2nd-system effect - "let's rewrite it from scratch, now that we know what we want". And you throw away all the hard-learned lessons.
> Instead, the right train of thought is: "what would perfect code look like?" and then meticulously describe to the LLM what "perfect" is to shape every line that gets generated.
I don't think there's perfect code.
Code is automation - it automates human effort and humans themselves have error, hence not perfect.
So as long as code meets or exceeds the human output, it's "good enough" and meets expectations. That's what a typical customer cares about.
A customer will happily choose a tent made of tarp and plastic sticks that's available at their budget, right now when it's raining outside, over an architectural marvel that will be available sometime in the future at some unknown pricepoint.
Put another way, I don't think if you built CharlieAI CharlieGPT today, where the only differentiating factor over ChatGPT was that CharlieGPT was written using perfect code, you would have any meaningful edge.
I am yet to see any evidence where everything else being equal, one company had an edge over another simply due to superior code.
Infact, I have overwhelming evidence of companies that had better code succumb and vanish against companies that had very little, if any code, because those dollars were instead invested in better customer discovery, segmentation and analytics ("what should we build?", "if we did one thing that would give our customers an unfair advantage, what would be that thing?")
Software history is full of perfect OS, editors, frameworks, protocols that is lost over time because a provably inferior option won marketshare.
You are using a software controlled SMPS to power your device right now. You have 0 idea what the quality of that code is. All you care about is whether that SMPS drains your battery prematurely and heats up your device unnecessarily. It's extremely unlikely that such an efficient, low overhead control system was written using well abstracted modules. It's more likely that control system is full of gotos and repeated violations of DRY that would make a perfectionist shudder and cry.
I think there is a parallel to what happened to watch market with Quartz crisis. The same way Quartz has led to decline of Swiss movements, LLMs are going to have a huge effect on developer market. I hypothesize that in future there will be a micro segment which care about quality, taste, exclusivity etc the same way the luxury watch makers found a niche. My perspective is that this "taste" or "quality" will not be a moat. Instead, it will be a niche where only a small segment would care about it.
Trying to bring my nose for AI up to standard -- care to share what you're smelling? For me it's:
- Short, declarative sentences, stating grandiose yet vague claims, in a high school vocabulary: "Taste becomes useful when it moves from vibe to diagnosis."
- Absence of references (let alone web links) to real-world examples.
- Em-dashes, gone. No semicolons, but 23 full colons. As instructed by prompt?
It’s an unbelievable lack of self awareness. I tried to give it the benefit of the doubt because surely no one would stoop to that level, but 5 paragraphs in and I’m certain it is AI written.
If you're properly bitter-lesson-pilled then why wouldn't better models continue to develop and improve taste and discernment when it comes to design, development, and just better thinking overall?
I think that would imply the creation of AGI (i.e. something as intelligent or more intelligent than mankind), which many consider to be science fiction at this point.
> bitter-lesson-pilled
The "bitter lesson" doesn't imply that AGI is coming, all it says is that letting AIs learn on their own yields better results than directly teaching them things.
They do improve, but the general creativity and sparkle we see with increasing scale comes mostly from scaling up pretraining/parameter-size, so it's quite slow and expensive compared to the speed (and decreasing cost) people have come to take for granted in math/coding in small cheap models. Hence the reaction to GPT-4.5: exactly as much better taste and discernment as it should have had based on scaling laws, yet regarded almost universally as a colossal failure. It was as unpopular as the original GPT-3 was when the paper was released, because people look at the log-esque gains from scaling up 10x or 100x and are disappointed. "Is that all?! What has the Bitter Lesson or scaling done for me lately?"
So, you can expect coding skills to continue to outpace the native LLM taste.
I think the author addresses this in saying that since AI output is statistically plausible by design its unlikely to improve in this area. Why do you think AI will get better in this way?
Regardless of how good the tools get, third-party tooling can never be a product differentiator unless you somehow manage to have exclusive access. Otherwise, everyone else out there can and will use the same tools you are. It's more a hedonic treadmill than a moat.
Speed and distribution aren't a long-run moat because they are something AI can canabalize in a platform. Eventually they will coexist on your distribution base and offer it at a lower cost than you. Its a mote if it holds up before you exit at a high valuation... which a lot are setup to do.
Taste: that's interesting. There is an argument there. It's hard to keep in the long-run and requires a lot of reinvestment in new talent
Proprietary data: Yes, very much so.
Trade Craft: Your new shiney system will still have to adhere to methods of of old clunky real world systems. Example, evidence for court. Methods for investigations. This is going to be industry specific, but you'd be surprised how many there are. This is long-term.
Those who have the moat should focus on short burts of meaningful changes as they will rely heavily on gaining trust in established systems. In those places its more about trusting whats going on than doing it faster and better, so you want trust + faster and/or better.
IMO, taste has always been one of the strongest moats because we struggle to define what good taste even is. We know it when we see it, but other than pointing to examples, we can’t really describe it in general terms. I still remember a line from Paul Graham’s Hackers and Painters where he was describing the difficulty of hiring software engineers. He says he was talking with a colleague after an interview and remarked (I’m paraphrasing), “I know he can write code. But does he have taste?” Taste is something we all want our colleagues and products and services to have, but defining it is really difficult. And yes, I fully agree with the writer that it’s important more than ever in this age of AI where generation is cheap.
I think "taste" is definitely an overused meme at this point, its like tech twitter discovered this word in 2024 and never stopped using it (same with "agency", "high leverage", etc).
Having read the article, I think I see the author's argument (*). I think "taste" here in an engineering context basically just comes down to an innate feeling of what engineering or product directions are right or wrong. I think this is different from the type of "taste" most people here are talking about, though I'm sure product "taste" specifically is somewhat correlated with your overall "taste." Engineering "taste" seems more correlated with experience building systems and/or strong intuitions about the fundamentals. I think this is a little different from the totally subjective, "vibes based taste" that you might think of in the context of design or art.
Now where I disagree is that
1. "taste" is a defensible moat
2. "taste" is "ai-proof" to some extent
"Taste" is only defensible to the extent that knowing what to do and cutting off the _right_ cruft is essential to moving faster. Moving faster and out executing is the real "moat" there. And obviously any cognitive task, including something as nebulous as "taste," can in theory be done by a sufficiently good AI. Clarity of thought when communicating with AI is, imo, not "taste."
Talking specifically about engineering - the article talks about product constraints and tradeoffs. I'd argue that these are actually _data_ problems, and once you solve those, tradeoffs and solving for constraints go from being a judgement call to being a "correct" solution. That is to say, if you provide more information to your AI about your business context, the less judgement _you_ as the implementer need to give. This thinking is in line with what other people here have already said (real moats are data, distribution, execution speed).
I think there's something a bit more interesting to say about the user empathy part, since it could be difficult for LLMs to truly put themselves in users shows when designing some interactive surfaces. But I'm sure that can be "solved" too, or at least, it can be done with far less human labor than it already takes.
In general though, tech people are some of the least tasteful people, so its always funny to see posts like this.
After a decade with tech people I can confidently say that most of them have zero taste because they have little to no exposure to the world outside of their bubble.
It's frankly pathetic to see how techno-optimists think that innovations like driverless cars will simply be happy pills to be swallowed by the masses who make a fractional amount of money to them.
As a species we have quite literally killed each other for less.
I use AI for code and we review that code and write tests ourselves first which the AI cannot touch. For writing we hardly ever do, unless we know the requester of something is incompetent and will never read it anyway; then it is a waste of time to do anything, but they expect something substantial and nice looking to tick a few boxes. It is great for that; a large bank with 40 layers of management, all equally incompetent, asked for a 'all encompassing technical document vault'; one of them sent an 'expectation document' which contained so much garbage as to show they did not even know what they were asking, but 1000s of pages was the expectation. So sure, claude will write that in an hour, notebooklm will add 100 slidedecks for juiceness. At first sight it looks amazing; its probably mostly accurate as well, but who knows; they will never ever read it; no one will. We got the 20m+ (with many opportunities to grow much larger) project. Before that was only in reach of the huge consultants (where everyone in those management levels worked before probably) who we used to lose against. Slop has its purpose.
> AI and LLMs have changed one thing very quickly: competent output is now cheap.
If you're working on something not truly novel, sure.
If you're using LLMs to assist in e.g. Mathematics work on as-yet-unproven problems, then this is hardly the case.
Hell, if we just stick to the software domain: Gemini3-DeepThink, GPT-5.4pro, and Opus 4.6 perform pretty "meh" writing CUDA C++ code for Hopper & Blackwell.
And I'm not talking about poorly-spec'd problems. I'm talking about mapping straightforward mathematics in annotated WolframLanguage files to WGMMA with TMA.
I am not sure you set it up right. Did you have a runnable WolframLanguage file so it can compare results? Did you give it H100 / H200 access to compile and then iterate?
My experience is that once you have these two, it does amazing kernel work (Codex-5.4).
It doesn't have to be anything so extreme as novel work. The frontier of models still struggle when faced with moderately complex semantics. They've gotten quite good at gluing dependencies together, but it was a rather disappointing nothingburger watching Claude choke on a large xterm project I tried to give him. Spent a month getting absolutely nowhere, just building stuff out until it was so broken the codebase had to be reset and he'd start over from square 1. We've come a long way in certain aspects, but honestly we're just as far away from the silver bullet as we were 3 years ago (for the shit I care about). I'm already bundling up for the next winter.
lol the unfortunate truth is that hundreds of billions and trillions will be spent to learn a single truth:
Taste cannot simply be bought nor can you bring products that add value into the world through sheer will of training machines.
Ah, the classic "we'll ship production to China and just do design and marketing in US, because we have taste on what to build, and China doesn't". That worked really well...
If you replace "China" with some other countries with large offshore engineering centres, your statement would still hold true today.
China managed to copy, improve, and localise for their Chinese-reading market, then oust competition through good use of the Great Firewall (though I wonder if that specifically was an unintended consequence).
Many other countries, especially the English-speaking ones that don't have a great firewall to keep their market buying locally, still need to compete with US tech giants for "taste".
Lots of techies hate things that are popular with the rest of humanity. You see lots of nagging, complaining, and disconnected from reality takes. Hate for Instagram, "Dropbox will never work", "pop culture sucks", etc.
I'll make a mean joke: a lot of y'all better learn a trade. Plumbing, perhaps. I kid, of course, but I also wonder if it might turn out to be the eventual reality.
This cope is insane. Even simple projects generated by Claude are riddled with bugs. And there’s no way in hell it could generate a larger scoped project without a lot of manual human intervention. But yea, TODO apps and trivial calculators are effectively “solved”. Same with leetcode. I guess that’s probably the limit of many people’s imagination these days.
This reads like cope. If taste were a real moat, designers and art directors would be the highest paid people in tech. They arent. Execution speed, distribution, and capital are moats. Taste is a tiebreaker at best. The market consistently rewards "good enough, shipped fast" over "exquisite, shipped late".
It’s not that straightforward. Art directors and designers get paid to visually communicate things the business wants to communicate— anything from brand vibes, to directing people to click on a “buy me” button, to the state of an interface. Most designers in tech companies aren’t even the ones that design things like branding — that’s done by specialists in extremely well-compensated studios, and corporate designers are stuck following their guidelines. Taste is nearly irrelevant to an interface designer, for example.
Taste is cheap. Taste (or a rudimentary version of it at least) is something you start with at the beginning of your career. Taste is the thing that tells you "this is fucking cool", or "I don't know why but this just looks right". LLM's are not going to replicate that because it's not a human and taste isn't something you can make. Now - MAKING something that "looks right" is hard, and because LLM's are churning out the middle - the middle is moving somewhere else. Just like rich people during the summer.
There are subheadings every 3 paragraphs and enough bullets to reload a machine gun.
There are also neither any sources nor any personal anecdotes. Everything feels generic.
> Over time, this changes how you work. You stop admiring polish for its own sake. You get faster at spotting empty specificity, borrowed tone, and fake confidence.
“Empty specificity, borrowed tone and fake confidence” describe the article itself.
It's getting bad here; I've seen at least three obviously AI-written "anti-AI" or "AI critical" pieces hit and remain on the front page in the last week. I can't help but think about Bill Hicks on marketing: "Everyone here who’s in marketing is now thinkin' the same thing: 'Oh, cool. Bill's going for that anti-marketing dollar. That's a huge market.'"
No - at face value, our work has diminished value. The entire supply and demand economics of our careers is changing in the blink of an eye.
There are people trying to figure out what this means and where to create value. "Taste is the only moat" is one such hypothesis. "Senior engineers will be fine" is another.
Everything is super frothy right now and we're in for a wild 2026.
What you notice
What you reject
How precisely you can explain what feels wrong
I think it's just as important, if not more, to be able to explain what is right and what you accept. Having a well defined acceptance criteria also fits into existing project management frameworks. These criteria are generally based on asking users. The article mentions, You do not get a spreadsheet that tells you which sentence will make a customer care, which feature is worth a month of engineering time, or which design crosses the line from polished to forgettable. And this is why you talk to your customers.
I agree with the author and I think this is turning everyone into an investor. How I view (financial) investing as a career is that it is less manual and more taste oriented. You put your stake in the things you feel will work out and taste here just means the judgement required to make good calls. A person with good taste would have a better idea of capital allocation.
What AI is doing is making all of us investors instead of doers. "Doing" is no longer something praiseworthy - what will become praiseworthy is how your taste has turned out in hindsight.
I'm seeing this at work. More or less everyone can do tasks well. But what's harder now is the more subtle task of taking bets and seeing it work over a few months or years.
Disagree with the overall argument. Human effort is still a moat. I've been spending the past couple of months creating a codebase that is almost entirely AI-generated. I've gotten way further than I would have otherwise at this pace, but it was still a lot of effort, and I still wasted time going down rabbit holes on features that didn't work out.
There's some truth in there that judgement is as important as ever, though I'm not sure I'd call it taste. I'm finding that you have to have an extremely clear product vision, along with an extremely clear language used to describe that product, for AI to be used effectively. Know your terms, know how you want your features to be split up into modules, know what you want the interfaces of those modules to be.
Without the above, you run into the same issue devs would run into before AI - the codebase becomes an incoherent mess, and even AI can't untangle it because the confusion gets embedded into its own context.
I feel like you're pretty strongly agreeing that taste is important: " I'm finding that you have to have an extremely clear product vision...""
Clear production vision that you're building the right thing in the right way -- this involves a lot of taste to get right. Good PMs have this. Good enginers have this. Visionary leaders have this....
The execution of using AI to generate the code and other artifacts, is a matter of skill. But without the taste that you're building the right thing, with the right features, in a revolutionary way that will be delightful to use....
I've looked at three non-engineer vibe-coded businesses in the past month, and can tell that without taste, they're building a pretty mediocre product at best. The founders don't see it yet. And like the article says, they're just setting themselves up for mediocrity. I think any really good PM would be able to improve all these apps I looked at almost immediately.
> I've gotten way further than I would have otherwise at this pace, but it was still a lot of effort, and I still wasted time going down rabbit holes on features that didn't work out.
By the time I'm done learning about the structure of the code that AI wrote, and reviewing it for correctness and completeness, it seems to be as much effort as if I had just written it myself. And I fear that will continue to be the reality until AIs can be trusted.
You make a really salient point about having a clear vision and using clear language. Patrick Zgambo says that working with AI is spellcasting; you just need to know the magic words. The more I work with AI tools, the more I agree.
Now, figuring out those words? That's the hard part.
Isn't this a temporary situation though.
Today: Ask AI to "do the thing", manual review because don't trust the AI
Tomorrow: Ask AI to "do the thing"
I'm just getting started on my AI journey. It didn't take long before I upgraded from the $17 a month claude plan to the $100 a month plan and I can see myself picking the $200 a month plan soon. This is for hobby projects.
At the moment I'm reviewing most of the code for what I'm working on, and I have tests and review those too. But, seeing how good it is (sometimes), I can imagine a future where the AI itself has both the tech chops and the taste and I can just say "Maybe me an app to edit photos" and it will spit out a user friendly clone of photoshop with good UX.
We already kind of see this with music - it's able to spit out "Bangers". How long until it can spit out hit rom-coms, crime shows, recipes, apps? I don't think the answer is "never". I think more likely the answer is in N years where N is probably a single digit.
> Disagree with the overall argument.
It's leaning in a good direction, but the author clearly lacks the language and understanding to articulate the actual problem, or a solution. They simply dont know what they dont know.
> Human effort is still a moat.
Also slightly off the mark. If I sat one down with all the equipment and supplies to make a pair of pants, the majority of you (by a massive margin) are going to produce a terrible pair of pants.
Thats not due to lack of effort, rather lack of skill.
> judgement is as important as ever,
Not important, critical. And it is a product of skill and experience.
Usability (a word often unused), cost, utility, are all the things that people want in a product. Reliability is a requirement: to quote the social network "we dont crash". And if you want to keep pace, maintainability.
> issue devs would run into before AI - the codebase becomes an incoherent mess
The big ball of mud (https://www.laputan.org/mud/ ) is 27 years old, and still applies. But all code bases have a tendency to acquire cruft (from edge cases) that don't have good in line explanations, that lack durable artifacts. Find me an old code base and I bet you that we can find a comment referencing a bug number in a system that no longer exists.
We might as an industry need to be honest that we need to be better librarians and archivists as well.
That having been said, the article should get credit, it is at least trying to start to have the conversations that we should be having and are not.
I think you're missing the point. Effort is a moat now because centaurs (human+AI) still beat AIs, but that gap gets smaller every year (and will ostensibly be closed).
The goal is to replicate human labor, and they're closing that gap. Once they do (maybe decades, but probably will happen), then only that "special something" will remain. Taste, vision... We shall all become Rick Rubins.
Until 2045, when they ship RubinGPT
> Without the above, you run into the same issue devs would run into before AI - the codebase becomes an incoherent mess, and even AI can't untangle it because the confusion gets embedded into its own context.
We have a term for this and it is called "Comprehension Debt" [0] [1].
[0] https://arxiv.org/abs/2512.08942
[1] https://medium.com/@addyosmani/comprehension-debt-the-hidden...
> ... for AI to be used effectively.
I'm continually fascinated by the huge differences in individual ability to produce successful results with AI. I always assumed that one of the benefits of AI was "anyone can do this". Then I realized a lot of people I interact with don't really understand the problem they're trying to solve all that well, and have some irrational belief that they can get AI to brute force their way to a solution.
For me I don't even use the more powerful models (just Sonnet 4.6) and have yet to have a project not come out fairly successful in a short period of time. This includes graded live coding examples for interviews, so there is at least some objective measurement that these are functional.
Strangely I find traditional software engineers, especially experienced ones, are generally the worst at achieving success. They often treat working with an agent too much like software engineering and end up building bad software rather than useful solutions to the core problem.
https://x.com/netcapgirl/status/2024140332963705342?s=46
evergreen.
I was expecting a circa 1993 rambling essay, pal.
usually when you call something "evergreen" it's not 2 months old
Is the joke that the guy is drinking bad coffee?
Sounds like I’d better run out and buy an Arcteryx vest.
This seems more telling on the artist who, I guess, believes that if you have taste in any field, it will manifest itself as wearing stylish clothes. I see their most recent blog post is analyzing luxury brands, so I think I'm on point here.
What's evergreen about it?
This is a mistake.
Instead, the right train of thought is: "what would perfect code look like?" and then meticulously describe to the LLM what "perfect" is to shape every line that gets generated.
This exercise is hard for some folks to grasp because they've never thought much about what well-constructed code or architectures looks like; they have no "taste" and thus no ability to precisely dictate the framework for "perfect" (yes, there is some subjectivity that reflects taste).
> Instead, the right train of thought is: "what would perfect code look like?" and then meticulously describe to the LLM what "perfect" is to shape every line that gets generated.
I think this goes against what a lot of developers want AI to be (not me, to be clear).
I've worked in too many large codebases where no one can point to any _single file or class_ and label it "correct," ("the right way") yet management is amazed when the lack of a "North Star" means the codebase is full of overlapping, piecemeal patterns that are lucky to work together at all.
> Instead, the right train of thought is: "what would perfect code look like?"
That's the classic 2nd-system effect - "let's rewrite it from scratch, now that we know what we want". And you throw away all the hard-learned lessons.
https://en.wikipedia.org/wiki/Second-system_effect
> Instead, the right train of thought is: "what would perfect code look like?" and then meticulously describe to the LLM what "perfect" is to shape every line that gets generated.
I don't think there's perfect code.
Code is automation - it automates human effort and humans themselves have error, hence not perfect.
So as long as code meets or exceeds the human output, it's "good enough" and meets expectations. That's what a typical customer cares about.
A customer will happily choose a tent made of tarp and plastic sticks that's available at their budget, right now when it's raining outside, over an architectural marvel that will be available sometime in the future at some unknown pricepoint.
Put another way, I don't think if you built CharlieAI CharlieGPT today, where the only differentiating factor over ChatGPT was that CharlieGPT was written using perfect code, you would have any meaningful edge.
I am yet to see any evidence where everything else being equal, one company had an edge over another simply due to superior code.
Infact, I have overwhelming evidence of companies that had better code succumb and vanish against companies that had very little, if any code, because those dollars were instead invested in better customer discovery, segmentation and analytics ("what should we build?", "if we did one thing that would give our customers an unfair advantage, what would be that thing?")
Software history is full of perfect OS, editors, frameworks, protocols that is lost over time because a provably inferior option won marketshare.
You are using a software controlled SMPS to power your device right now. You have 0 idea what the quality of that code is. All you care about is whether that SMPS drains your battery prematurely and heats up your device unnecessarily. It's extremely unlikely that such an efficient, low overhead control system was written using well abstracted modules. It's more likely that control system is full of gotos and repeated violations of DRY that would make a perfectionist shudder and cry.
I think there is a parallel to what happened to watch market with Quartz crisis. The same way Quartz has led to decline of Swiss movements, LLMs are going to have a huge effect on developer market. I hypothesize that in future there will be a micro segment which care about quality, taste, exclusivity etc the same way the luxury watch makers found a niche. My perspective is that this "taste" or "quality" will not be a moat. Instead, it will be a niche where only a small segment would care about it.
(edit: typos)
Didn't PG write a post about this like a month ago?
Title: Good Taste the Only Real Moat Left
Followed by an entire AI generated fluff piece https://www.pangram.com/history/347cd632-809c-4775-b457-d9bc...
Flagged
It is profoundly ironic that this article is AI generated.
Seriously. Very Claude-y vibes from this post. I guess the value of human effort doesn’t extend to writing your own blog posts
Trying to bring my nose for AI up to standard -- care to share what you're smelling? For me it's:
- Short, declarative sentences, stating grandiose yet vague claims, in a high school vocabulary: "Taste becomes useful when it moves from vibe to diagnosis."
- Absence of references (let alone web links) to real-world examples.
- Em-dashes, gone. No semicolons, but 23 full colons. As instructed by prompt?
It’s an unbelievable lack of self awareness. I tried to give it the benefit of the doubt because surely no one would stoop to that level, but 5 paragraphs in and I’m certain it is AI written.
Looks like the comments on this article are too.
Was looking for this comment. How can people still read AI slop like this?
He has taste. The LLM knows that and creates a tasteful article. /s
If you're properly bitter-lesson-pilled then why wouldn't better models continue to develop and improve taste and discernment when it comes to design, development, and just better thinking overall?
I think that would imply the creation of AGI (i.e. something as intelligent or more intelligent than mankind), which many consider to be science fiction at this point.
> bitter-lesson-pilled
The "bitter lesson" doesn't imply that AGI is coming, all it says is that letting AIs learn on their own yields better results than directly teaching them things.
They do improve, but the general creativity and sparkle we see with increasing scale comes mostly from scaling up pretraining/parameter-size, so it's quite slow and expensive compared to the speed (and decreasing cost) people have come to take for granted in math/coding in small cheap models. Hence the reaction to GPT-4.5: exactly as much better taste and discernment as it should have had based on scaling laws, yet regarded almost universally as a colossal failure. It was as unpopular as the original GPT-3 was when the paper was released, because people look at the log-esque gains from scaling up 10x or 100x and are disappointed. "Is that all?! What has the Bitter Lesson or scaling done for me lately?"
So, you can expect coding skills to continue to outpace the native LLM taste.
I think the author addresses this in saying that since AI output is statistically plausible by design its unlikely to improve in this area. Why do you think AI will get better in this way?
At least in part because some of Taste is fashion.
most (all?) models are fundamentally a regression toward the mean. Good taste is rarely, if ever, residing in the mean.
Regardless of how good the tools get, third-party tooling can never be a product differentiator unless you somehow manage to have exclusive access. Otherwise, everyone else out there can and will use the same tools you are. It's more a hedonic treadmill than a moat.
Think about moats in long term vs short term:
Speed and distribution aren't a long-run moat because they are something AI can canabalize in a platform. Eventually they will coexist on your distribution base and offer it at a lower cost than you. Its a mote if it holds up before you exit at a high valuation... which a lot are setup to do.
Taste: that's interesting. There is an argument there. It's hard to keep in the long-run and requires a lot of reinvestment in new talent
Proprietary data: Yes, very much so.
Trade Craft: Your new shiney system will still have to adhere to methods of of old clunky real world systems. Example, evidence for court. Methods for investigations. This is going to be industry specific, but you'd be surprised how many there are. This is long-term.
Those who have the moat should focus on short burts of meaningful changes as they will rely heavily on gaining trust in established systems. In those places its more about trusting whats going on than doing it faster and better, so you want trust + faster and/or better.
Good judgment and effort has always been the "real moat" - in arts, music, science, food, product...
There's always been ways to "flatten the middle" - by outsourcing, by using pre-packaged goods, with industrialization...
So yeah we've always loved handcrafted, exquisite things; there's never been a "moat" in middle
It doesn't mean you can't make a good living without a moat though
Does anyone remember this quote by Why the Lucky Stiff: “When you don't create things, you become defined by your tastes rather than ability.”
IMO, taste has always been one of the strongest moats because we struggle to define what good taste even is. We know it when we see it, but other than pointing to examples, we can’t really describe it in general terms. I still remember a line from Paul Graham’s Hackers and Painters where he was describing the difficulty of hiring software engineers. He says he was talking with a colleague after an interview and remarked (I’m paraphrasing), “I know he can write code. But does he have taste?” Taste is something we all want our colleagues and products and services to have, but defining it is really difficult. And yes, I fully agree with the writer that it’s important more than ever in this age of AI where generation is cheap.
Ever since Rick Rubin has been on his book tour, he has become the patron saint of Product Manger Tech Twitter.
It buried the more important point, one tech hasn't learned yet.
Taste may be kind of important because it helps toward the truly important thing, which is skin-in-the-game.
But also, with the right skin-in-the-game, you don't even need "taste." You just need real life consequences, which we don't do enough in tech.
I'm sure this can be solved by A/B tasting. ;)
I already disagree with the first line: competent output is not cheap. At least if defined as a final product.
- Just think about scientific research. Lots of data analysis results are not cheap to get.
- Even vibe coding is difficult: you need to think very hard about what you want.
What is cheaper now are some building blocks. We just have a new definition of building blocks. But putting the blocks is still hard.
I think "taste" is definitely an overused meme at this point, its like tech twitter discovered this word in 2024 and never stopped using it (same with "agency", "high leverage", etc).
Having read the article, I think I see the author's argument (*). I think "taste" here in an engineering context basically just comes down to an innate feeling of what engineering or product directions are right or wrong. I think this is different from the type of "taste" most people here are talking about, though I'm sure product "taste" specifically is somewhat correlated with your overall "taste." Engineering "taste" seems more correlated with experience building systems and/or strong intuitions about the fundamentals. I think this is a little different from the totally subjective, "vibes based taste" that you might think of in the context of design or art.
Now where I disagree is that
1. "taste" is a defensible moat
2. "taste" is "ai-proof" to some extent
"Taste" is only defensible to the extent that knowing what to do and cutting off the _right_ cruft is essential to moving faster. Moving faster and out executing is the real "moat" there. And obviously any cognitive task, including something as nebulous as "taste," can in theory be done by a sufficiently good AI. Clarity of thought when communicating with AI is, imo, not "taste."
Talking specifically about engineering - the article talks about product constraints and tradeoffs. I'd argue that these are actually _data_ problems, and once you solve those, tradeoffs and solving for constraints go from being a judgement call to being a "correct" solution. That is to say, if you provide more information to your AI about your business context, the less judgement _you_ as the implementer need to give. This thinking is in line with what other people here have already said (real moats are data, distribution, execution speed).
I think there's something a bit more interesting to say about the user empathy part, since it could be difficult for LLMs to truly put themselves in users shows when designing some interactive surfaces. But I'm sure that can be "solved" too, or at least, it can be done with far less human labor than it already takes.
In general though, tech people are some of the least tasteful people, so its always funny to see posts like this.
No one was discussing 'taste' until pg's article.
After a decade with tech people I can confidently say that most of them have zero taste because they have little to no exposure to the world outside of their bubble.
It's frankly pathetic to see how techno-optimists think that innovations like driverless cars will simply be happy pills to be swallowed by the masses who make a fractional amount of money to them.
As a species we have quite literally killed each other for less.
Ah, Steve Jobs vs Bill Gates. Designer vs 41 shades of blue. This is nothing new. There's space for everybody.
>AI and LLMs have changed one thing very quickly: competent output is now cheap.
Already wrong.
The only real moat is care. It was, it is, it will be.
Designers and product managers were showing Steve Jobs all the fancy things the new app for writing DVDs could do.
Steve Jobs stopped them, drew a square on the whiteboard and said “anything the user drags into this square gets written to the DVD” - that is taste!
Reminds me of PG's classic essay, "Taste for Makers" (2002): https://paulgraham.com/taste.html
> Good Taste the Only Real Moat Left > YC startups are doomed
I use AI for code and we review that code and write tests ourselves first which the AI cannot touch. For writing we hardly ever do, unless we know the requester of something is incompetent and will never read it anyway; then it is a waste of time to do anything, but they expect something substantial and nice looking to tick a few boxes. It is great for that; a large bank with 40 layers of management, all equally incompetent, asked for a 'all encompassing technical document vault'; one of them sent an 'expectation document' which contained so much garbage as to show they did not even know what they were asking, but 1000s of pages was the expectation. So sure, claude will write that in an hour, notebooklm will add 100 slidedecks for juiceness. At first sight it looks amazing; its probably mostly accurate as well, but who knows; they will never ever read it; no one will. We got the 20m+ (with many opportunities to grow much larger) project. Before that was only in reach of the huge consultants (where everyone in those management levels worked before probably) who we used to lose against. Slop has its purpose.
Try using a coding agent to write an efficient GPU kernel. I guess they might get good at it soon, but they definitely aren't there yet.
I had a very complex cuda kernel and codex cli managed to improve the throughout 20x.
> AI and LLMs have changed one thing very quickly: competent output is now cheap.
If you're working on something not truly novel, sure.
If you're using LLMs to assist in e.g. Mathematics work on as-yet-unproven problems, then this is hardly the case.
Hell, if we just stick to the software domain: Gemini3-DeepThink, GPT-5.4pro, and Opus 4.6 perform pretty "meh" writing CUDA C++ code for Hopper & Blackwell.
And I'm not talking about poorly-spec'd problems. I'm talking about mapping straightforward mathematics in annotated WolframLanguage files to WGMMA with TMA.
I am not sure you set it up right. Did you have a runnable WolframLanguage file so it can compare results? Did you give it H100 / H200 access to compile and then iterate?
My experience is that once you have these two, it does amazing kernel work (Codex-5.4).
It doesn't have to be anything so extreme as novel work. The frontier of models still struggle when faced with moderately complex semantics. They've gotten quite good at gluing dependencies together, but it was a rather disappointing nothingburger watching Claude choke on a large xterm project I tried to give him. Spent a month getting absolutely nowhere, just building stuff out until it was so broken the codebase had to be reset and he'd start over from square 1. We've come a long way in certain aspects, but honestly we're just as far away from the silver bullet as we were 3 years ago (for the shit I care about). I'm already bundling up for the next winter.
Rick Rubin said it best.
https://youtu.be/jg1WUOxY6Cg?si=0ajVvgKnyuSz0e2Y
no share Id link: https://youtu.be/jg1WUOxY6Cg
lol the unfortunate truth is that hundreds of billions and trillions will be spent to learn a single truth: Taste cannot simply be bought nor can you bring products that add value into the world through sheer will of training machines.
Ah, the classic "we'll ship production to China and just do design and marketing in US, because we have taste on what to build, and China doesn't". That worked really well...
If you replace "China" with some other countries with large offshore engineering centres, your statement would still hold true today.
China managed to copy, improve, and localise for their Chinese-reading market, then oust competition through good use of the Great Firewall (though I wonder if that specifically was an unintended consequence).
Many other countries, especially the English-speaking ones that don't have a great firewall to keep their market buying locally, still need to compete with US tech giants for "taste".
And if anybody knows about good taste, it's techies, right?
Some of the worst taste and worst opinions.
Lots of techies hate things that are popular with the rest of humanity. You see lots of nagging, complaining, and disconnected from reality takes. Hate for Instagram, "Dropbox will never work", "pop culture sucks", etc.
I'll make a mean joke: a lot of y'all better learn a trade. Plumbing, perhaps. I kid, of course, but I also wonder if it might turn out to be the eventual reality.
This cope is insane. Even simple projects generated by Claude are riddled with bugs. And there’s no way in hell it could generate a larger scoped project without a lot of manual human intervention. But yea, TODO apps and trivial calculators are effectively “solved”. Same with leetcode. I guess that’s probably the limit of many people’s imagination these days.
This reads like cope. If taste were a real moat, designers and art directors would be the highest paid people in tech. They arent. Execution speed, distribution, and capital are moats. Taste is a tiebreaker at best. The market consistently rewards "good enough, shipped fast" over "exquisite, shipped late".
It’s not that straightforward. Art directors and designers get paid to visually communicate things the business wants to communicate— anything from brand vibes, to directing people to click on a “buy me” button, to the state of an interface. Most designers in tech companies aren’t even the ones that design things like branding — that’s done by specialists in extremely well-compensated studios, and corporate designers are stuck following their guidelines. Taste is nearly irrelevant to an interface designer, for example.
I maintain the opinion that courage is the only moat and always will be
> A practical loop for training taste
Taste is cheap. Taste (or a rudimentary version of it at least) is something you start with at the beginning of your career. Taste is the thing that tells you "this is fucking cool", or "I don't know why but this just looks right". LLM's are not going to replicate that because it's not a human and taste isn't something you can make. Now - MAKING something that "looks right" is hard, and because LLM's are churning out the middle - the middle is moving somewhere else. Just like rich people during the summer.
Extremely ironic piece of slop.
I agree.
For those who didn’t read the article…
There are subheadings every 3 paragraphs and enough bullets to reload a machine gun.
There are also neither any sources nor any personal anecdotes. Everything feels generic.
> Over time, this changes how you work. You stop admiring polish for its own sake. You get faster at spotting empty specificity, borrowed tone, and fake confidence.
“Empty specificity, borrowed tone and fake confidence” describe the article itself.
It's getting bad here; I've seen at least three obviously AI-written "anti-AI" or "AI critical" pieces hit and remain on the front page in the last week. I can't help but think about Bill Hicks on marketing: "Everyone here who’s in marketing is now thinkin' the same thing: 'Oh, cool. Bill's going for that anti-marketing dollar. That's a huge market.'"
Yeah I feel like we’re getting pranked here
No - at face value, our work has diminished value. The entire supply and demand economics of our careers is changing in the blink of an eye.
There are people trying to figure out what this means and where to create value. "Taste is the only moat" is one such hypothesis. "Senior engineers will be fine" is another.
Everything is super frothy right now and we're in for a wild 2026.
Related: https://blog.kinglycrow.com/no-skill-no-taste/
Discussion: https://news.ycombinator.com/item?id=47089907
Article assumed as absolute truth, without explanation, that competent systems can be effortlessly implemented.
If one disagrees with that's statement, there is nothing of value to extract from this article.
Words are cheap, bullet point are cheap.
Taste shows up in three places:
I think it's just as important, if not more, to be able to explain what is right and what you accept. Having a well defined acceptance criteria also fits into existing project management frameworks. These criteria are generally based on asking users. The article mentions, You do not get a spreadsheet that tells you which sentence will make a customer care, which feature is worth a month of engineering time, or which design crosses the line from polished to forgettable. And this is why you talk to your customers.
I agree with the author and I think this is turning everyone into an investor. How I view (financial) investing as a career is that it is less manual and more taste oriented. You put your stake in the things you feel will work out and taste here just means the judgement required to make good calls. A person with good taste would have a better idea of capital allocation.
What AI is doing is making all of us investors instead of doers. "Doing" is no longer something praiseworthy - what will become praiseworthy is how your taste has turned out in hindsight.
I'm seeing this at work. More or less everyone can do tasks well. But what's harder now is the more subtle task of taking bets and seeing it work over a few months or years.
> That is why so much AI-generated work feels familiar:
This was already a complaint people had before Ai. Like when logos and landing pages all used to look the same. Or coffee shops all looking the same.
Or cars, or apartment buildings, or houses, or....
It has been for a while. Hollywood and other outlets didn’t need AI tools to create abysmal slop.
Well, nope. There are three real moats left in software:
Distribution, Data (Proprietary) and Iteration Speed.
Very successful companies have all three: Stripe, Meta, Google, Amazon.
The moat of all four of those companies is simply infrastructure, partnerships, and plain-old name recognition.
Data and iteration speed aren't moats. I don't know what you mean by "distribution".
taste isn't a moat at all because it's so variable, in fact this stuff will start dictating what taste is through broad proliferation
you already see it on facebook with all the ai generated meme sharing... taste is being eroded there
The new world order is what not to build...
I dont buy the authors argument. Not much has changed imo. Mediocre slop has always been the easiest thing to generate.