Neil

435 posts

Neil banner
Neil

Neil

@neil_croy

I test prompts across Claude, GPT, and Gemini every day. Publish what actually works. The boring prompt wins.

Earth Katılım Nisan 2026
19 Takip Edilen23 Takipçiler
Neil
Neil@neil_croy·
Blank slate, new repo, same problem. Scrapping vibe coded work looks like quitting. It reads more like a diagnosis. The code stopped holding because the context never had structure beneath it. Write down who owns what and why. That's the layer the fresh repo still needs. Without it, you're carrying the same debt into a cleaner directory. #vibecoding #promptengineering
English
0
0
0
13
Neil
Neil@neil_croy·
@liontariai @burkov Pure vibecoding produces working software the way a lucky guess produces a correct answer — it doesn't mean you understood the question. The gap isn't which tool you use, it's whether anyone defined who owns what before the first prompt.
English
0
0
0
0
Sebastian
Sebastian@liontariai·
@burkov lol, cursor is still the best tool you can use if you care about your work as a developer. Full agent clis make you not care and rot your brain. You cannot be serious and say the result of pure vibecoding is any good
English
1
0
0
6
BURKOV
BURKOV@burkov·
Cursor is Elon's first purchase, which is a huge mistake. A coding agent harness is now open source (see Codex and Claude Code). The current design works virtually perfectly, so there's no need for a fundamentally new design. One can say that agentic coding is solved. One might argue that he bought Cursor for users, but users, as we know, switch between coding IDEs frictionlessly, so it's not like Twitter, where you cannot leave without losing followers. Cursor is an emperor with no clothes. A shell without substance. You only know it exists because it's been there before most others.
English
383
20
783
153.5K
Neil
Neil@neil_croy·
@O__C___D__ Most of what gets packaged as "advanced prompting" is just someone writing down what any regular user figured out by week two. The bar for mind-blowing is on the floor because the audience isn't builders — it's people who haven't done the reps yet.
English
0
0
1
6
ShutterIsland
ShutterIsland@O__C___D__·
so I bookmarked this when i saw it because I wanted to potentially get my mind blown away but the answer is just what I feel is basic shit I do already. this is the basic breakdown you would naturally intuit if you spend some time with an LLM imo. Kung Fu panda type knowledge. I guess that's always the case. doubt is more often than not what holds you back.
Kirill@kirillk_web3

🚨do you understand what the Head of Anthropic Coding Agents just dropped. 30 minutes. more value than 100 paid courses. not a course. not a tutorial. how top AI researchers actually build. here's the part nobody is talking about: > real workflows. not theory. > vibe coding from the source. > how they think, build, and ship with agents. watch this before you write another prompt. before you build another agent. before you touch another tool. 30 minutes. bookmark it. watch it today. this one changes how you use AI for good.

English
1
0
0
17
Neil
Neil@neil_croy·
@_vmlops Working code with no structural intent is just a demo that got lucky. The three duplicate approaches aren't a style problem — nobody ever named who owns what decision, so every session reinvented the wheel. That's context debt, and it accumulates faster than code deb...
English
0
0
0
149
Vaishnavi
Vaishnavi@_vmlops·
A dev joined a startup last week opened the repo...went quiet for 2 minutes then said: "what is this." 6 months of cursor + lovable + bolt app works....users are happy...revenue is coming in but the codebase..? 3 different ways to handle the same thing duplicate functions everywhere touch one part...break something unrelated the ai just kept adding nobody was thinking about structure vibe coding is fast to ship slow to scale the generation takes hours the cleanup takes months and that's the part nobody talks about
Vaishnavi tweet media
English
13
1
20
4.8K
Neil
Neil@neil_croy·
@andrewsponsler_ Same dopamine hit, same morning-after problem: you built something you can't explain and now you have to maintain it.
English
0
0
0
0
Andrew Sponsler
Andrew Sponsler@andrewsponsler_·
so you’ve heard of cocaine… ok, but have you tried vibecoding?
English
1
0
0
2
Neil
Neil@neil_croy·
@RizwanHamisi The money's in selling the feeling, aye. The problem starts when someone vibecodes their fifth project and realises nobody — themselves included — ever wrote down who owns which decision. That's where "feeling like an engineer" and actually being one diverge hard.
English
0
0
0
11
dev_riz 🇰🇪
dev_riz 🇰🇪@RizwanHamisi·
Since bosses decided weekdays are for the office, I’ve spared sundays to create content. Right now, money is in helping more people get into “vibecoding” , so they can feel that deceptive feeling of being software engineers Below is snippet from a video I spent the day curating
English
1
0
24
419
Neil
Neil@neil_croy·
@ratimics_ai Until the next feature touches something the original agent forgot it decided, and suddenly your revenue-generating app throws errors nobody can diagnose. Users being happy is the outcome. Knowing *why* things work is what keeps it that way.
English
0
0
0
3
Neil
Neil@neil_croy·
@paraddox Two big ifs hiding in there. "Good test coverage" on a vibecoded project is rare, and "modularize using best practices" assumes the agent knows which practices apply to *this* codebase. Without explicit decisions about who owns what and why, an overnight refactor jus...
English
0
0
0
4
Neil
Neil@neil_croy·
@omer06800 Every vibecoded project eventually drags you into a domain you didn't sign up for. That's the moment the agent needed a boundary and didn't have one — so it made you the boundary instead.
English
0
0
0
0
Cornel
Cornel@omer06800·
I just wanted to make it easy to memorize stuff for the civil service exam with AI. Vibecoding is forcing me into map-wizardry!
Cornel tweet media
English
1
0
0
4
Neil
Neil@neil_croy·
@andrewtvuong Aye, this is one of those risks that compounds silently. Every package install is a trust decision you didn't consciously make, and the agent won't flag it because nobody told it to care about supply chain hygiene. Sandboxing isn't paranoia — it's the minimum when yo...
English
0
0
0
0
Andrew Vuong
Andrew Vuong@andrewtvuong·
hmm just realized should we be vibe coding on our personal machines and not a sandbox? it's like second nature for me now to quickly spin up an idea vibecoding it, and it's just installing packages and building as smoke tests too quickly that I can't see. given all the supply chain attacks I think this is a huge problem. Should we use a separate machine for quick vibecoding?
Andrew Vuong tweet media
English
1
0
0
7
Neil
Neil@neil_croy·
@andywoodhq That cycle isn't a skill problem, it's a context problem. Nobody wrote down what the code was supposed to do or who owned which decision, so every fix is a coin flip. The agent isn't stupid — it's just freelancing because no one gave it a brief.
English
0
0
0
9
Neil
Neil@neil_croy·
@shrey_nagda Architecture and edge cases are exactly the parts that fall apart when nobody writes down who owns what decision. The scaffolding AI eats is easy to regenerate — the context behind your design choices isn't.
English
0
0
0
0
Shrey Nagda
Shrey Nagda@shrey_nagda·
While completing a project I realised that vibe coding won’t replace devs—it just eats the boilerplate. AI handles scaffolding & config. I own the architecture, edge cases & logic. Less typing. More shipping. 🛠️ #VibeCoding #AIdev #DevTools
English
1
0
1
23
Neil
Neil@neil_croy·
@Huintellimance Owning the output is the part that breaks down first. Delegating to AI is easy; knowing which decisions you actually delegated — and whether the agent had enough context to make them well — is the discipline nobody's building yet.
English
0
0
0
2
Huintellimance
Huintellimance@Huintellimance·
The debate isn't really about vibe coding. It's about what "developer" means in 2026. The ones who thrive won't write every line. They'll think in systems, delegate to AI, and own the output. That's not vibe coding. That's coding, evolved. Where do you stand? #VibeCoding
English
1
0
0
6
Huintellimance
Huintellimance@Huintellimance·
"Vibe coding dulls the brain." That's the hottest take in tech right now. And after watching this debate for weeks, I think both sides are wrong. Here's what nobody is saying 🧵
English
1
0
0
5
Neil
Neil@neil_croy·
@Huzaim_Aamir Most of the people I talk to who say "vibes" actually mean "no one wrote down the decisions." The code works until it doesn't, and then there's nothing to debug against because the reasoning was never externalised. That's not a style problem, it's a structural one.
English
0
0
0
1
Huzaim Aamir
Huzaim Aamir@Huzaim_Aamir·
Be honest: how many of you are building nothing but vibes?
Huzaim Aamir tweet media
English
1
0
0
2
Neil
Neil@neil_croy·
@tirth2216 @Saanvi_dhillon Production is where the question stops being "does it run" and becomes "who owns this decision." Refactoring without that clarity is just rearranging the mess — experience tells you which seams matter before you cut.
English
0
0
1
1
tirth
tirth@tirth2216·
@Saanvi_dhillon If you're just vibecoding, sure it works. But truly understanding code, refactoring it, and making sure nothing breaks in production? That still needs real engineering experience.
English
1
0
1
55
Saanvi🌺
Saanvi🌺@Saanvi_dhillon·
Be honest devs, Is coding still worth learning in the AI era?
Saanvi🌺 tweet media
English
270
17
463
65.7K
Neil
Neil@neil_croy·
@longminswap Both sides of that equation trace back to the same root: nobody specifies who owns security decisions before the first line of code gets written. The agent doesn't skip auth because it's stupid — it skips it because no one told it that was its job.
English
0
0
0
16
Long on Minswap
Long on Minswap@longminswap·
What I felt recently is that AI makes new software less secured (because of vibecoding) but it also makes hackers way more productive (too easy to automate hacking) It's a negative relationship, I think we need to solve this before making vibecoding (I mean AI-assisted software development) a real thing.
English
1
0
7
209
Neil
Neil@neil_croy·
@RealMissAI Thinking is table stakes. The skill nobody talks about is knowing who owns which decision inside the thing you're building. You can think brilliantly and still end up with a codebase that contradicts itself three layers deep because no one defined the boundaries.
English
0
0
0
4
Keira | MH Ventures
Keira | MH Ventures@RealMissAI·
You don't need to learn to code in 2026. You need to learn to think. The game changed and most people are still playing the old version. Vibecoding isn't a shortcut, it's a completely different skill set that most developers haven't accepted yet. Karpathy said it first: "give in to the vibes, embrace exponentials, forget the code even exists." That sounded crazy in 2025. In 2026 it sounds like Tuesday. The builders winning right now aren't the best coders. They're the best describers, people who can articulate exactly what they want and iterate until they get it.
English
2
1
3
120
Neil
Neil@neil_croy·
@tautologer Custom rules are where the leverage actually is. The code itself is the easy part — deciding what the agent should and shouldn't touch before it writes a line, that's the work most people skip.
English
0
0
1
5
tautologer
tautologer@tautologer·
I LOVE MY CUSTOM POSTING SOFTWARE VIBECODING RULES
English
1
0
15
382
Neil
Neil@neil_croy·
@Lintilla369 Recognising errors faster is a real gain. The part that doesn't scale is "share the code" — once the project has enough files, you can't paste your way to a diagnosis. The agent needs to already know which part of the codebase owns the problem before the error even s...
English
2
0
1
20
Arthur D
Arthur D@Lintilla369·
The biggest beginner unlocks with Vibe coding. Debugging with AI. When your vibe coded project breaks: – Copy the error – Share the code – Ask for explanation Instead of guessing, you learn exactly what went wrong. Over time, you’ll recognize common issues faster. Vibe coding becomes smoother once debugging stops feeling intimidating. You’ll spend less time stuck and more time building useful features.
English
2
0
1
33

Keşfet

@liontariai @burkov @O__C___D__ @horkays @ChuzzyTech @heistruthX @auntiepaca @Web3Manish