I'm starting to notice AI power users communicating in a shorthand tech-speak I find unintelligible. I genuinely cannot follow the individual points in this article or what the overall arc is.
Is this something that was already part of web dev or is it new?
I regard myself as pro-AI, and I find this article is all over the place. It's just trying to promote their SaaS. It's clear they didn't even read what the LLM generated before posting it, as evidenced by this line:
"99.9% uptime (CloudWatch, Feb to Sep, insert your real number)"
This user also appears to have a Reddit account from 3 years ago where they claim to be a half human cyborg.
> I genuinely cannot follow the individual points in this article or what the overall arc is.
It's just slop. Acknowledge, move on, and find something better to spend your time on.
We're in a weird uncanny valley state of AI writing where it's barely comprehensible, but makes people feel productive so they shove it everywhere, including where it's not needed.
At some point, either:
1. People will realize that this is a waste of time
2. Models will get better and this will be remembered as a blip in history
It's not the users doing the communication, it's their AIs.
And you, as a human, were not meant to read it - you're supposed to instead feed it back into your own AI agent and have it create a groundbreaking new AI startup based on it. Give it your Cloudflare credentials, sit back, sip a latte, and within an hour or two you'll have a beautiful, fully functional website, complete with fancy fonts, 30 different "No X. No Y. Just Z" taglines, and a $20 monthly subscription, ready to show off to investors. You'll be a highly successful entrepreneur by the end of the week.
My workflow these days is basically ultracode + PRs. I do control the features that go but barely need to check the code due to the amount of checks: lint, TS, Local AI visual Model for visual check for screen frame comparison (Mac M5 48GB mem), Pixel/Screenshot regression test for basic UI, code coverage/c8 to avoid trash code without tests, etc. etc.
Essentially the tests are just for the AI to remember what they did (as in the description + break if they change smth unintentionally).
I think that’s because across experience levels there really isn’t much to teach regarding “how to use AI”.
The only grind is still just do you know the subject matter, tools and frameworks, etc. because that’s the knowledge that lets you look at the output and say “this is fine” or “this is absolutely the worst implementation I’ve ever seen try again but do X”.
In my experience adding on multiple layers of controls etc aside from basic skills is essentially as useful as a ring of salt and a live offering.
Does this implementation really require this robust of infra / setup?
The entire stack feels overkill for a repo that has 2 followers and 29 stars.
I also am probably not the target demographic though, I just download things that I want to preserve and completely forget about them in a folder called 0_archive.
I just feel like in 2013 some fella would have had this project running on an old laptop in his garage hooked up to FTP and the uptime would be comparable.
I also love when AI adds tidbits like this:
> and staging has to pass before prod deploys.
Thanks for the clarity on that, Claude, you read my mind.
Text is human written. The repository popularity shows dev interest not product usage, if this was closed source it would have been 0 stars and 0 followers.
Maybe most devs are not interested in how to build stable infra and just vibe code
I don't think so. I have lieer + notmuch setup with a few AI generated Python scripts that will mark things as read. It also took about 5 minutes of fiddling with notmuch cli and jq to extract SNS message bodies from emails (for instance).
I think a thin layer on notmuch would get you pretty close without so much infrastructure.
On the web scraping side, scrapling seems to pretty much just work. There hasn't been a website yet it won't load in some mode--even Google SERP.
Well, what I found "AI native software" to be is "usable heedlessly via scripts/CLI with UI being optional", because that's how LLM agents naturally interact with software.
Also, I'm really not sure this bookmark project/website/extension would require 6 months of Claude with 6 of them running continuously to build.
I read the article and it sounds kinda-sorta good at face value but the more I read and look into readplace.com the less things make sense and the more "ick" grows. Oh well just a puff piece... It's a big pet peeve of mine when someone uses a "CEO" title where it's probably very likely there's no Board or any other officers (or is a side project).
Author comes off as diligent, intelligent and dedicated... but might gain some credibility by having real humans read and proof the copy.
Ai native is only a term for the present moment. There is still skill in using ai. We are fast approaching where you ask the ai, and it does the task, there is no wrangling it
13,000 tests. That tells you everything you need to know. It doesn't matter how experienced you say you are, if AI is doing all the work including writing and running tests because I know no human is actually reviewing these. I've recently found tests written by Fable that don't actually do anything in a very large codebase. My point is not to advocate for doing any of this manually, but to highlight one of the problems brought about by pumping out so much code.
I hate to sound cynical, but I don't know if bragging about being AI-native is something any of us should be doing. We're really just pulling a lever on a slot machine and crossing our fingers.
> I've recently found tests written by Fable that don't actually do anything in a very large codebase.
tbf, I've seen these kind of tests as well at $dayjob, written by humans before gpt was a thing. That doesn't invalid your point. But it's still funny that we expect the machine to be perfect, before it can do something for us.
>But it's still funny that we expect the machine to be perfect, before it can do something for us.
We expect machines to be perfect at the things they should be perfect at, and when they aren't perfect we consider that a malfunction and try to make them closer to perfect. A calculator that can't do math correctly every time is a bad calculator. And until LLMs came along computers were just complicated calculators.
Now, however, we build the calculator with an LLM and just accept that it sometimes gets confuzzled and makes numbers up because "humans are no better at math."
It's not funny, it's the way things used to work and it's objectively worse that things can't work that way anymore.
You need to ensure the whole system is optimised for LLM. Getting human written code and trying to run like this will be a failure since you'll be missing essential guidelines particularly on the domain-specific design fundamentals that are acquired in the beginning in a human feedback loop before the automation really kicks in
I'm starting to notice AI power users communicating in a shorthand tech-speak I find unintelligible. I genuinely cannot follow the individual points in this article or what the overall arc is.
Is this something that was already part of web dev or is it new?
I regard myself as pro-AI, and I find this article is all over the place. It's just trying to promote their SaaS. It's clear they didn't even read what the LLM generated before posting it, as evidenced by this line:
"99.9% uptime (CloudWatch, Feb to Sep, insert your real number)"
This user also appears to have a Reddit account from 3 years ago where they claim to be a half human cyborg.
https://www.reddit.com/user/fagnerbrack/comments/195jgst/faq...
I think that's all I need to know, personally.
Bruh, there's 3 paragraphs I got info from the codebase using AI, the bullet points and the insights section
Everything else was written by hand and I made sure to not use AI even for structure (which I usually do)
It's settled. Either I use AI or not people will always say the same thing
I find wine helps. Lots of wine.
> I genuinely cannot follow the individual points in this article or what the overall arc is.
It's just slop. Acknowledge, move on, and find something better to spend your time on.
We're in a weird uncanny valley state of AI writing where it's barely comprehensible, but makes people feel productive so they shove it everywhere, including where it's not needed.
At some point, either:
1. People will realize that this is a waste of time
2. Models will get better and this will be remembered as a blip in history
I think people are making skills to have AI generate text that sounds like them but AI--especially Claude--has a hard time not sounding like AI.
It's not the users doing the communication, it's their AIs.
And you, as a human, were not meant to read it - you're supposed to instead feed it back into your own AI agent and have it create a groundbreaking new AI startup based on it. Give it your Cloudflare credentials, sit back, sip a latte, and within an hour or two you'll have a beautiful, fully functional website, complete with fancy fonts, 30 different "No X. No Y. Just Z" taglines, and a $20 monthly subscription, ready to show off to investors. You'll be a highly successful entrepreneur by the end of the week.
Doing drugs sounds both more rewarding and productive.
Might even be cheaper!
I didn't find her article hard to understand. Maybe you have a reading comprehension issue?
Seems more like AI-in-the-loop more than AI-native.
I feel like this post didn't teach me anything I didn't already know or have an intuition about.
My workflow these days is basically ultracode + PRs. I do control the features that go but barely need to check the code due to the amount of checks: lint, TS, Local AI visual Model for visual check for screen frame comparison (Mac M5 48GB mem), Pixel/Screenshot regression test for basic UI, code coverage/c8 to avoid trash code without tests, etc. etc.
Essentially the tests are just for the AI to remember what they did (as in the description + break if they change smth unintentionally).
I think that’s because across experience levels there really isn’t much to teach regarding “how to use AI”.
The only grind is still just do you know the subject matter, tools and frameworks, etc. because that’s the knowledge that lets you look at the output and say “this is fine” or “this is absolutely the worst implementation I’ve ever seen try again but do X”.
In my experience adding on multiple layers of controls etc aside from basic skills is essentially as useful as a ring of salt and a live offering.
Looking for a job? This sounds like personal marketing speech rather than a blog worth reading, no new insights no nothing
no shit, it's some medium backed shit website.
Wish i could filter ALL of these, not just those with the medium website
Does this implementation really require this robust of infra / setup?
The entire stack feels overkill for a repo that has 2 followers and 29 stars.
I also am probably not the target demographic though, I just download things that I want to preserve and completely forget about them in a folder called 0_archive.
I just feel like in 2013 some fella would have had this project running on an old laptop in his garage hooked up to FTP and the uptime would be comparable.
I also love when AI adds tidbits like this:
> and staging has to pass before prod deploys.
Thanks for the clarity on that, Claude, you read my mind.
Text is human written. The repository popularity shows dev interest not product usage, if this was closed source it would have been 0 stars and 0 followers.
Maybe most devs are not interested in how to build stable infra and just vibe code
I don't think so. I have lieer + notmuch setup with a few AI generated Python scripts that will mark things as read. It also took about 5 minutes of fiddling with notmuch cli and jq to extract SNS message bodies from emails (for instance).
I think a thin layer on notmuch would get you pretty close without so much infrastructure.
On the web scraping side, scrapling seems to pretty much just work. There hasn't been a website yet it won't load in some mode--even Google SERP.
Well, what I found "AI native software" to be is "usable heedlessly via scripts/CLI with UI being optional", because that's how LLM agents naturally interact with software.
Also, I'm really not sure this bookmark project/website/extension would require 6 months of Claude with 6 of them running continuously to build.
> 99.9% uptime (CloudWatch, Feb to Sep, insert your real number)
Did we forget to insert the real number?
I read the article and it sounds kinda-sorta good at face value but the more I read and look into readplace.com the less things make sense and the more "ick" grows. Oh well just a puff piece... It's a big pet peeve of mine when someone uses a "CEO" title where it's probably very likely there's no Board or any other officers (or is a side project).
Author comes off as diligent, intelligent and dedicated... but might gain some credibility by having real humans read and proof the copy.
If humans proof the copy it sounds AI generated and people complain. I tried already.
Ai native is only a term for the present moment. There is still skill in using ai. We are fast approaching where you ask the ai, and it does the task, there is no wrangling it
13,000 tests. That tells you everything you need to know. It doesn't matter how experienced you say you are, if AI is doing all the work including writing and running tests because I know no human is actually reviewing these. I've recently found tests written by Fable that don't actually do anything in a very large codebase. My point is not to advocate for doing any of this manually, but to highlight one of the problems brought about by pumping out so much code.
I hate to sound cynical, but I don't know if bragging about being AI-native is something any of us should be doing. We're really just pulling a lever on a slot machine and crossing our fingers.
> I've recently found tests written by Fable that don't actually do anything in a very large codebase.
tbf, I've seen these kind of tests as well at $dayjob, written by humans before gpt was a thing. That doesn't invalid your point. But it's still funny that we expect the machine to be perfect, before it can do something for us.
>But it's still funny that we expect the machine to be perfect, before it can do something for us.
We expect machines to be perfect at the things they should be perfect at, and when they aren't perfect we consider that a malfunction and try to make them closer to perfect. A calculator that can't do math correctly every time is a bad calculator. And until LLMs came along computers were just complicated calculators.
Now, however, we build the calculator with an LLM and just accept that it sometimes gets confuzzled and makes numbers up because "humans are no better at math."
It's not funny, it's the way things used to work and it's objectively worse that things can't work that way anymore.
You need to ensure the whole system is optimised for LLM. Getting human written code and trying to run like this will be a failure since you'll be missing essential guidelines particularly on the domain-specific design fundamentals that are acquired in the beginning in a human feedback loop before the automation really kicks in
>I don't know if bragging about being AI-native is something any of us should be doing
Me neither but we now have a likert scale on our performance review for it...
So what does it look like? This is a load of booster talk tuned to reel in the devs who appreciate *****.
Getting a 502 error. I had to come to the comments to check whether that was the joke.
[dead]