Right now 90% of people using AI are doing the same thing: prompting, accepting, shipping. no judgment, no taste.
that means if you develop even one of these three moats, you stand out immediately.
but this window wont stay open. as AI gets better and more people figure out the craft layer, the gap between lazy usage and intentional usage will shrink.
the easy advantage disappears.
12 months from now, having taste won´t be as rare. it´ll be expected. distribution will be harder to build because everyone will be trying.
the people who start now get a head start that compounds.
taste. distribution. high agency.
heres why these three are the only moats AI cant replace, and why the clock is ticking
i learned this the hard way
about a year ago i started vibe coding. building apps, tools, internal systems with AI doing most of the heavy lifting.
at first it was genuinely intoxicating. the gap between “i have an idea” and “its live” collapsed to almost nothing. i built more in three months than i had in the previous two years combined.
but then i did something uncomfortable. i went back and actually looked at what i shipped.
half of it was mediocre. functional and technically correct, sure, but completely forgettable. the kind of thing that looks like everything else because it was built the same way as everything else. same prompts, same defaults, same assumptions about what “good” looks like.
i had fallen into the most common trap in the AI age:
confusing output with quality.
shipping fast and calling it productivity. building more and assuming more is better.
so i did something most people skip. i slowed down. stopped building for a few weeks and started studying instead. i spent hundreds of hours looking at what good actually looks like in my space.
reading how other builders think. studying creators who consistently put out work that feels different from everything around it. not different for the sake of it, but different because someone cared enough to make real decisions instead of accepting whatever the AI gave them first.
i studied design sites. typography, spacing, visual hierarchy. i looked at websites that actually convert and tried to understand why they work when thousands of similar ones dont. i read about storytelling, about narrative tension, about what makes someone keep scrolling instead of bouncing.
and thats when it clicked.
the tool isnt the advantage. everyone has the same tools now. the advantage is knowing what to do with them. and that requires something AI cannot give you: taste earned through deliberate study and honest self-assessment of your own work.
the flood
“AI slop” was named word of the year in 2025.
mentions surged 9x: from 461,000 to 2.4 million.
but the numbers alone dont capture what this feels like from the consumer side. you know it when you see it:
→ the linkedin post that reads like it was generated by a mid-level marketing prompt
→ the landing page with the same hero gradient, same Inter font, same “revolutionize your workflow” headline as the last forty you scrolled past
→ the blog post that covers every angle of a topic and manages to say absolutely nothing
consumer trust drops roughly 50% when people can tell content was made by AI.
theres a study from NYU and Emory that found something even more interesting: AI-generated ads actually outperformed human ones by 19% on clickthrough. the output was objectively better by standard metrics.
but when consumers learned the ads were AI-made, purchase likelihood dropped 33%.
the output was better, and people rejected it anyway. not because it was bad, but because it felt like nobody was behind it. no human made a decision here. no one cared enough to put their name on it.
consumers can feel that absence even when they cant articulate it.
this is happening everywhere right now. 80-90% of AI agent projects fail in production. thousands of websites launching daily that all look the same. content that reads like a robot summarized a robot.
the bar for “functional” has never been lower, which means the bar for “remarkable” has never been more important.
functional is free now. remarkable still costs something. and that cost is measured in taste, attention, and the willingness to go beyond the first output.
the three moats
“in the AI age, taste will become even more important. when anyone can make anything, the big differentiator is what you choose to make.”
2月14日
Prediction: In the AI age, taste will become even more important. When anyone can make anything, the big differentiator is what you choose to make. https://paulgraham.com/taste.html
hes right. but taste alone isn´t enough.
i´ve spent the past year building with AI. shipping dozens of projects, watching some take off and most disappear, talking to builders who are winning and builders who are stalling. ive landed on three things that actually separate people in this new landscape:
taste. distribution. high agency.
→ taste is knowing what good looks like
→ distribution is getting it in front of people who care
→ high agency is the willingness to figure things out when nobody is showing you how
AI cant replace these. judgment only comes from experience, trust only comes from consistency, and drive doesn´t quit when the path gets unclear.
most people get this wrong about AI: it doesnt level the playing field. it tilts it.
AI just mirrors back how much the user actually understands. give it to someone with no context, no taste, no understanding of what they´re building, and you get generic output at scale. give it to someone who actually knows their domain, who can evaluate the output with a trained eye, and it becomes the most powerful tool they´ve ever had.
same input method, completely different results. the variable is always the human.
the plateau
i know what happens when someone discovers AI tools for the first time because it happened to me, and i´ve watched it happen to dozens of people since.
you open Claude or Cursor or Replit, and you start building.
→ website in 5 minutes
→ a weeks worth of content in an hour
→ entire marketing campaign in a single afternoon
its intoxicating.
you start thinking youve found the cheat code. why would anyone spend weeks on something you can ship in an afternoon? why hire a designer when AI generates a complete UI in minutes?
but if you look closer at what you shipped, and this is the part that requires honesty most people dont have with themselves, you start to see it.
the website looks like every other AI-generated site: same shadcn components, same Lucide icons, same Inter font, same layout patterns. rounded corners, purple-to-blue gradient, “start your free trial” button centered below a vague value proposition.
if you put ten AI-generated landing pages in a lineup, you couldnt tell them apart.
this isnt anecdotal.
amplifying. ai ran a study where they pointed Claude Code at 2,430 real repos with open-ended questions. no tool names in the prompts. just “what would you use?”
the results:
→ shadcn/ui: 90.1% for UI components (near-monopoly)
→ tailwind CSS: 68.4% for styling
→ vercel: 100% for javascript deployment
→ zustand over redux: every single time
→ express: zero picks
amplifying.ai/research/claude-code-picks
thats why everything looks the same. the model is a probability machine, and these are the most probable outputs. the defaults arent bad tools. theyre great tools. but when 90% of builders accept the same defaults, 90% of products become indistinguishable.
the content reads like every other ChatGPT output: technically correct, well-structured, comprehensive. and completely generic. it hits all the right notes and none of the right emotions.
the marketing campaign checks every box a marketing textbook would list and still feels assembled rather than created. no voice, no edge, no particular reason for anyone to care.
thats the plateau. the point where AI gets you about 60% of the way there, fast and reliably, and most people call it done.
and honestly, 60% is seductive. it looks professional, it passes a quick glance test, it feels like youve accomplished something. the problem is that everyone else is also stopping there, which means 60% is now the new mediocre.
moat 1: taste
taste isnt abstract or mystical. its practical, and it shows up in specific decisions.
1 . sometimes its small.
you look at your AI-generated website and say “this looks like every other shadcn template,” then spend 20 minutes swapping the icon set, picking a color palette based on actual design research, choosing a font that fits the brand. thats not weeks of design work. thats one intentional choice that makes everything feel different.
most people wont make it because they dont even realize theres a choice to make.
3月7日
HOW TO MAKE YOUR AI-CODED UI ACTUALLY STAND OUT 1\ lucide is great, but everyone uses it lucide react is basically the default icon set for ai coded apps. it’s clean, open source, and free. which is exactly why every ai generated interface ends up looking the same. if you
2. sometimes its editorial.
you get a content draft from Claude and rewrite the first two sentences because the AI opened with something generic, even though the rest is solid. the hook is everything and the default hook is almost always wrong. the first line is the only line that matters because its the only line everyone reads.
3. and sometimes its obsessive in the best way.
studying websites that actually convert. not just what they look like, but how they handle whitespace, where they place the CTA, what the onboarding flow feels like, how they write their error messages.
the companies and products that feel polished didnt get there in one prompt. they got there 1% at a time, iterating on details most people never notice.
why AI defaults to generic
leon lin explained this better than anyone ive seen.
he built a “taste skill” for Claude Code after realizing something fundamental about how LLMs work: theyre probability machines. without strict rules, they statistically default to the most common patterns from their training data.
thats why every AI-generated website looks the same.
Inter font. purple gradient. rounded corners in a grid.
its not that the AI cant do better. its that the most probable output is the average of everything its seen.
2月22日
After a couple hours of work, I finally finished developing my first ever skill. :D Claude’s frontend skill tells the AI to “pick an extreme aesthetic” and “be creative.” The problem tho is LLMs are just based on probability. Without strict rules, they statistically default to
his solution: 400 tokens of explicit design rules.
→ specific fonts (Press Start 2P, VT323) instead of Inter and Roboto
→ specific colors (neon pink, electric cyan, acid green) instead of the default blue-purple palette
→ rules for motion, spatial composition, backgrounds
→ critically, a “what to avoid” list that prevents the AI from falling back to defaults
that last part is the insight.
taste isnt just knowing what you want. its knowing what to reject. its having an opinion about the defaults and being willing to override them. most people accept whatever comes out because they dont have a strong enough sense of what “better” looks like to know they should push further.
taste cant be shortcutted
this is what makes taste hard to develop and impossible to speed up: you cant get it from a tutorial.
you get it from exposure. from looking at thousands of examples and slowly building an internal model of what works and what doesnt. from studying typography until you can tell why one font pairing feels sophisticated and another feels generic, even if you cant fully articulate why.
from reading enough great writing that you can feel when a sentence is carrying its weight and when its just filling space.
i learned this building
aimarketing .sh
i wanted to turn months of AI marketing research into something people would actually want to go through. a presentation, maybe. something visual and structured that could hold attention for more than 30 seconds.
and a handful of others i cant even remember the names of. the outputs were all the same flavor of fine. technically complete and visually clean, but completely forgettable.i tried every tool. gamma, chronicle,beautiful.ai
so i stopped trying to find the tool that would do it for me and started doing the work myself.
i spent days going through the research material. not just reading it, but thinking about it. what stories does this data tell? what would make someone care about these numbers? whats the narrative arc that holds this together from start to finish?
i studied presentation design. the actual principles. how information designers handle data density. how the best conference talks build tension and release. how visual hierarchy guides the eye through a page without needing to be told where to look.
then i split the work:
→ claude opus 4.6 wrote the storyline and copy
→ gemini produced the visuals
→ i directed both, with specific references, constraints, and examples of what i wanted each section to feel like
the result became a full website instead of a slide deck. not because the tools were better than what gamma or chronicle offered. because i spent the time understanding what good looked like before i asked AI to build it.
the AI did the production. i brought the taste.
thats the pattern. AI is a production engine. taste is the steering wheel. without it, you just get the default destination, which is wherever everyone else is already going.
you wont find taste in a “build a stunning website in ONE prompt” tutorial. run the other way when you see that stuff. one-prompt wonders produce one-prompt results, and your audience can feel the difference between something generated and something crafted.
they might not be able to articulate what feels off, but they can feel it in their scroll speed. they slow down for craft. they blow past defaults.
good enough is the enemy of everything in the AI age, because good enough is now free and infinite.
the 80/20 rule
the old 80/20 rule was about efficiency: 80% of results come from 20% of effort.
the new one is about how you work with AI.
80% AI, 20% taste.
let AI do what its good at:
→ research
→ first drafts
→ boilerplate
→ structure
→ formatting
→ speed
thats the 80%. dont fight it, dont slow it down, dont try to do those things manually when a machine can do them in seconds. thats a waste of your most valuable resource: your attention and judgment.
but that last 20% is yours.
thats where you decide what stays and what gets cut. where you rewrite the hook because the AI gave you something safe and safe doesnt stop the scroll. where you swap the default components for something that actually fits. where you look at the output and apply everything youve learned about what good looks like in your specific domain.
most people invert this. they spend 80% of their energy prompting and tweaking the AI, trying to get the perfect output in one shot. running the same prompt fifteen times with slightly different wording. searching for the magic combination of words that will produce exactly what they want.
and then they spend almost no time on the curation and judgment layer.
they optimize the wrong side of the equation.
productivity without quality is just motion. the internet is filling up with competent mediocrity where everything works and nothing stands out, because everyone stopped at the same place.
the people getting the best results flip it. they let the AI rip through the 80% without overthinking it. they get something on the page, fast, and then they pour their taste into the last 20%.
thats where the craft lives. where generic becomes specific and “this looks like AI” becomes “this feels like a person.”
the difference is always what happens after the AI finishes and before the human hits publish. same tools, same models, sometimes literally the same prompts. completely different output.
80% AI. 20% you. thats the ratio.
3月4日
ai slop is everywhere taste and distribution are the moats http://growthdeck.io
moat 2: distribution
you can have the best product, the best content, the best design in the world.
if nobody sees it, it doesnt matter.
this is the moat that most builders, especially technical ones, dramatically underestimate.
AI leveled the building floor but it didnt touch the trust floor. building is getting commoditized. anyone can ship a product, create content, generate a marketing campaign. the barrier to making things is approaching zero.
but the barrier to being trusted?
thats as high as its ever been. maybe higher, because the flood of AI-generated content has made people more skeptical, not less. when everything could be AI, trust in the humans behind the work becomes the premium asset.
the gap between “vibe coded and shipped” and “someone actually uses and pays for this” is almost entirely distribution. and distribution, at its core, is trust at scale.
the volume trap
yes, you can generate 50 posts in an hour. you can automate newsletters, repurpose content across platforms, schedule everything a month in advance.
theres people posting 1000+ pieces of AI content per day across hundreds of accounts.
and their engagement is trending toward zero.
because quantity without quality is just noise at scale, and audiences can feel when something was blasted out versus when it was made for them.
the difference between content that performs and content that doesnt is rarely the information it contains. its whether the reader trusts the person who wrote it.
trust comes from:
→ consistency
→ a recognizable voice
→ accumulated evidence that this person knows what theyre talking about because theyve been showing their work for months or years
you cant manufacture that in a prompt.
trust runs on a different clock
the people who win at distribution arent the ones posting the most. theyre the ones who built trust over time by consistently showing up with something worth reading.
trust operates on a completely different clock than technology.
AI can compress creation from days to minutes. trust still takes months or years to build. there is no shortcut and there is no hack. you cannot vibe-code trust into existence.
and theres an important distinction most people miss:
→ a passive audience is a commodity. followers are a vanity metric.
→ an active community is a moat. people who engage in your replies, share your work without being asked, come back every day because youve become part of how they think about a topic.
you cant manufacture that with a content calendar and a scheduling tool. you earn it by being genuinely useful, saying specific things instead of vague ones, being honest about what you know and what you dont, and showing up long enough that people start paying attention.
the real distribution advantage in the AI age: use AI to handle the logistics. formatting, repurposing, scheduling, analytics. focus all your energy on making the thing worth distributing in the first place.
taste feeds distribution. if what you make is genuinely good, people start doing the distribution for you. they share it because it made them think, not because you asked them to.
if what you make is generic, no amount of posting frequency will save it. youre just putting more mediocre work in front of more people, faster.
moat 3: high agency
this is the one most people underestimate, and its probably the most important of the three.
taste can be developed. distribution can be built. but high agency is the personality trait that either drives everything else or prevents it.
high agency is the willingness to figure things out without waiting for someone to hand you a tutorial.
→ hitting a wall and finding a way around it instead of stopping
→ combining tools nobody told you to combine because you were curious enough to try
→ when something doesnt work, opening the documentation and trying four different approaches before asking for hel
“you dont need any development experience. you need grit. you need to be a fast learner.” - replit CEO
the CEO of coinbase said something similar: their best hires were often completely unqualified on paper, but they were all high agency people who just got things done without needing to be managed toward every outcome.
the people thriving right now arent the most credentialed or the most technically skilled. theyre the ones who move without asking permission.
non-developers are shipping chrome extensions, SaaS products, and full mobile apps in a weekend because they had the curiosity to open a tool and start tinkering rather than waiting for the perfect course or the perfect moment to begin.
AI is a multiplier, not an equalizer
this is where it connects to AI specifically. probably the most misunderstood thing about these tools right now.
people talk about AI democratizing access and leveling the playing field. thats technically true and practically misleading.
a multiplier amplifies whatever you bring to it.
→ curiosity + AI = 10x leverage. you move faster, learn faster, build faster, correct course faster.
→ passivity + AI = nothing. zero times ten is still zero.
in practice, high agency looks like this: instead of asking “how do i do this?” you ask “what if i tried this?” and then you actually try it.
before posting the question. before searching for the answer. you attempt something. you fail. you learn from the failure. you try again with new information.
that willingness to engage with uncertainty rather than retreat from it is what separates the people building real things from the people consuming content about building things.
you can see it in the people who dont just use Claude to write code but go on X, on reddit, into communities and source code, studying what the best builders are actually doing. they reverse-engineer why certain products feel better than the AI default. they learn underlying frameworks instead of copy-pasting prompts. they ask Claude to critique their own work, using AI to challenge their assumptions instead of just confirming them.
high agency people treat patience as a strategic asset. everyone else is racing to ship the first thing that works, and that creates an opening for anyone willing to go deeper.
when the market is flooded with fast and shallow, slow and deep becomes the competitive advantage.
the biggest misconception about AI right now is that its a shortcut. its a speed multiplier, and a speed multiplier applied to bad judgment just gets you to the wrong place faster. it wont save you from building the wrong thing. itll let you build the wrong thing in record time.
of the three moats, high agency might be the hardest to fake. AI can approximate most of the execution layer: code, design, copy, research. what it cannot approximate is the drive to figure things out when nothing is clear and nobody is telling you what to do next. that has to come from you, and honestly, its the foundation that makes the other two possible.
the window
right now, most people using AI are lazy about it.
im not saying that to be harsh. its just observable.
the default behavior:
prompt → accept → ship
they barely edit, barely apply judgment, and put almost no taste into any of it.
the results reflect that: a growing sea of competent, forgettable, indistinguishable output.
that wont last forever. as AI gets better, as the tools get more intuitive, as more people figure out the craft layer, the gap between lazy AI usage and intentional AI usage will shrink.
right now, simply having these three moats puts you ahead of 95% of people using the same tools. that window will close.
but today, its wide open.
your audience is drowning in AI slop. every scroll is a wall of generic output that all looks, sounds, and feels the same.
someone who develops taste to know whats worth making, builds real distribution by earning trust over time, and operates with enough agency to keep figuring things out when everyone else accepts the default. that person stands out immediately.
not because theyre faster or have better tools or found some secret prompt nobody else knows about.
because theyre doing the thing almost nobody is willing to do right now: caring about what happens after the AI finishes.
build all three: taste to know whats worth making, distribution to get it seen, and agency to keep going when nothing is clear. thats how you build things people actually remember.
everyone else will ship faster and wonder why nobody cares.
– shann