William Roush

I'm tired of this stupid ass shit, fuck off. This person has been posting nothing but this, and twitter keeps being like "maybe you want to read another 'you're holding the phone wrong' posts again" No, I fucking don't.

I will 1,000% admit this AI makes okay marketing copy because I am so bad at making marketing copy Will a human that is skilled do better? Absolutely. For projects that I can't hire a human for yet because they're too immature, does AI get my foot in the door? Yes I'm not very good at visual layouts AI will give me something basic that I can roll with, a designer that is experienced and skilled in this will 100% wipe the floor with my design and 100% wipe the floor with the AI's design I think AI tooling is very useful where I can't justify the costs to hire somebody yet but it gets me into that MVP stage Because it's better at certain skills that I just absolutely suck at I can identify that and I'm willing to admit it At the skills I'm really good at it's very easy to identify where it falls short, and it's absolute insanity to pretend that that only exists where I'm an expert. But sometimes I don't need high performing or sometimes have to deal with the fact that I can't afford high performing until we get a little traction I think it has its places, assuming that the subsidy cliff doesn't make it financially unreasonable to continue to use these tools. But there is a lot of hype and a lot of stupidity going around

@hustlin_heev This is why we use separate apps for our dev and test envs, eek Though Shopify needs to support that flow better

shopify is really killing me because i made 1 bad api call while i was testing something just 1, and they flagged my app for 14 days and put these warnings all over it my customers are getting pissed off, and i think one might churn.... i really think u guys need to reserve this flag for people who call the bad api like 5x or something cuz just once and this punishment is not cool @liam_at_shopify

@adahstwt Not one word but my experience: Claude is down again

I'm a vibe coder, scare me with one word.

@PreciousBa82157 @catalinmpit It's not just known, research says we can't fix it Didn't you know random people on Twitter are smarter than the research teams that build these systems

@catalinmpit i've started blocking them,they're so fucking annoying atp, it's literally known LLMS hallucinate no matter how smart they're, they act like they're in this LLM worshipping cult for some reason

The “LLMs and Agents can’t be wrong. It’s the human who’s wrong” type of people are the most insufferable.

@tarasshyn @catalinmpit Lolol holy shit

@catalinmpit I felt that what could be ambiguous about this command was that it acted so stupidly. After that, I created custom MCPs and read-only keys for all APIs that I use and that allow permission control, and this is the only way Claude now interacts with my app.

@catalinmpit I especially like the duality of "I never check the output" and "the output is correct" No you're just letting it write problems. You're letting it write 50x the tests with 10% the coverage of real flows. Etc etc

@MikegazaurusR @JamesWard 100% Also: it's usually the same people

@StrangeWill @JamesWard Crypto bros? What about the Web3 shills? What about the Blockchain shills? What about the IoT shills? We're at like what, the fifth IT grift in the span of about 16 years?

"Developers now only review code" "Developers now only write specs" "The spec is the source of truth" "The code is the source of truth" "Code is now free" "Programming language doesn't matter" "LLMs will just write assembly" The only thing that is actually true: Everyone has an opinion but not a crystal ball.

@theo Smaller pieces, check and correct as it goes.

Just let Opus go for over an hour on a new feature. When it was done, I asked how I can test it. 20 minutes later, it realized I can't test it because it did the whole thing entirely wrong. Idk how you guys use this model every day for real work 🙃

What am I reviewing tests with? Claude has been great at generating a ton of tests (looks good on metrics) while missing critical paths or in some cases just making up results. Problem is: if you're not reviewing output, you can't vouch for it being right to begin with, as we've been -- we're catching it do a lot of things it shouldn't.

@0xKofi software engineers don't have to (upfront verifiability & tests). Data scientists have to (methodology is easy to get barely wrong and numbers can always "look" good).

I've been seeing a lot of folk on the timeline say IDEs are dying. But I don't get it. Are you guys not reviewing the code your agents generate 😅

He's just greasing the wheels for the subsidy crunch -- when Claude goes from $200/mo to $10k/mo. you have to shift the narrative from "it's cheap" to "no it's worth twice as much as the average senior engineer is getting paid", problem is outside of SV insanity circles, engineers are *not* being paid $500k/yr. $250k/yr in tokens is *insane*, but that's what they're hoping for to get their money back. These are the same people that were smoking crack and handing out 6-figure salaries to juniors during the last big tech crunch, we found out *QUICKLY* that the number is not justified and the market lurched hard.

Software engineers will not be trusted to spend 50% of their salary on variable opex costs with no guarantee of productivity, unless they are executive level. this is a pipe dream to sell GPUs

Knoll's Law of Media Accuracy is pretty well applied to AI. The people who usually are like "OMG this is so good" at their own job are usually massively shit at their job lol. I've had some luck with it on programming, but it's made some *massively* boneheaded mistakes, even with clear instructions. Does that mean it's useless? No. Does that mean I should verify its output and adjust accordingly? Yes.

It's crazy how AI is really good at the stuff I don't know anything about and total dog shit at the stuff I do.

@celispj These asshats are jamming up the queue for app approval too, SMH.

This is a first, app dev straight up copying our name for some free app store juice. Ridiculous.

There's quite a few reasons: A lot of us have seen massive software quality *decreases* over the past couple of years, user experience is in the toilet almost across the board. I've been working with a few teams that have leaned on AI a lot in the past couple of years: - One of them is responsible for my company to have to do a platform rewrite because they listened to an LLM on how to design a data pipeline - Another (that I'm an investor) has expanded their team 3x the size, gone through multiple iterations of AI "revolutions", are down massively on projected revenue, are 2 years behind on key product developments -- I'm hoping this pans out but I'm not holding my breath at this point. - I've had 2+ projects brought to us because they're stuck in AI hell of "I can't fix this trivial bug because it keeps breaking other things to fix it", but because the entire code base was developed by crap engineers it's a nightmare to try to work with, we turned them away. - We've had multiple engineers projects go in the toilet, hallucinated approaches we've told them wouldn't work that got approved by management, then months later ended up in the garbage because -- surprise surprise, they didn't work. - Multiple engineers that couldn't explain their work and would have to ask the LLM about what they're doing. It enables bad teams to be *much* worse, to a crazy degree, the reduced friction has meant a lot more distraction, it also allows people that would normally sound like idiots to be confidently incorrect on direction. It can take a good engineer and make them better, but most people are completely incapable of determining which engineers are good anymore (they were bad at it before, now it's nearly impossible) At the same time I've been using it myself, and I've had to step in the way a few times *just today* to remind Claude to not do absolutely insane shit. I'm not even doing super complicated things, engineers that aren't taking the time to review stuff (most because they're lazy) are making an absolute shit show of stuff. Ultimately the biggest point: Pragmatically people aren't believing it because they're having *worse* experience, worse software, worse team members, more frustration, more fragility, more garbage code. People that will ship 40+ products to make $4k/mo, that's entirely unmaintainable, 30+ of those products are going to get 0 support, if any patches, it's crazy, and that's before the subsidy crunch. It may be all 40.

@rez0__ @Unipoopr @0xTib3rius Yeah, for simple contracts that are pretty much a "solved" problem (you can buy them online), sure. I've had frontier LLMs suggest some really ass legal advice before.

@Unipoopr @0xTib3rius "would it hold up in court?" is the better question

Lawyers being rated a 9/10 risk at being replaced by AI tells you everything you need to know about this "research".

"Writing code" and "being developers" are different things, developers who's jobs were just to blindly turn requirements into code? I guess, but they always sucked and never should have been employed in the first place. Engineers like that are endlessly frustrating. It's not like JavaScript was hard, I have 14 languages under my belt, it just kind of happened, I've picked up new languages in a day, my skill isn't in something that people can learn that quickly.

people keep saying AI replacing developers is just hype. meanwhile the creator of Node.js says: “This has been said a thousand times before, but allow me to add my own voice: the era of humans writing code is over. Disturbing for those of us who identify as SWEs, but no less true. That’s not to say SWEs don’t have work to do, but writing syntax directly is not it.” maybe we should listen.

It isn't just Claude, it's all of them. I'm watching my Claude spend go↗️as I use it The reality is it's going to be a bloodbath when these companies want a profit, oh Claude Code is going to charge me $60k/yr/dev? Nah fuck that noise, I'll move to another platform. There is practically zero friction to jump ship to another model, I just cancelled ChatGPT for Claude, I'm sure I'll end up on something else the moment Claude tries to pull a dumb move. Google -- funny enough, may be the one to outlast.

@TheNFTCPA I don't know how you can trust it, I do a lot of our accounting, I've had frontier models make massive mistakes. There's going to be a lot of audits and people getting in trouble this year, but sucks to suck. Stop trusting the machine blindly.

Today's episode of AI taking over Tax. Client: Your draft says 9k refund, AI told me 111K refund. You did something wrong. Me: Oh boy. send me what you are seeing. Me: This is wrong, just wrong. You don't qualify for any of the tax credits, you don't have 3 kids... #TaxTwitter