... I uh, I've done this before. Sorta.
Relatively new app, needed iteration about live performance and tweaks. Pipelines left the blast radius pretty minimal and out of the hands of "AI doing stuff", but AI would write code, push it, wait for the deploy, review the data, make adjustments and do it again.
On a live site with customers? Ehh, too high risk. A mistake could cost a ton, while I'm asleep lol.
Basically it boils down to basic business risk analysis: at a point your risk gets high enough that you'd have a person in the loop anyway. Because "the computer did it" doesn't bring back a million dollars in lost revenue.
Yes and no, ugh this is a difficult one.
Pre-AI, lots of companies justified *large* teams, we were never one of those, and it kind of clicked with AI.
With most dev teams, you needed to have a significant number of people, with a significant number of them being just "coders", people who effectively just slammed code all day, who couldn't care more about the architecture, who purely lived on the quality of their code. Not deliverables (they didn't design them), not system design (that was the Architect's job), just the ability to write code.
I *never* thought this was an acceptable position, but this position was a major one throughout most decent sized dev teams (for a lot of reasons, it bolstered startup team sizes which always looked good for the manager of the IT team, they would wax poetic about the size of the team they managed).
This is really just the market finding out what we knew the whole time: a significant, probably majority, easily supermajority of "Engineers" were just coders, and with AI that position is basically gone.
@GameCornerPlay1@Worshipperfx I've seen some tooling that's like syntax aware. There are tools and methods that either eek out some edge-cases that can be automated, or just throw it at an LLM.
Problem is, there is no magical way to go "two people edited the same file in the same place, wtf do you want?"
It doesn't *always* but I find when it's a big problem, it usually is. God classes/files or somewhere where people keep clobbering each other.
These days though I find what few conflicts we run into to be trivially handled by AI, the low number of conflicts due to task breakdown and team makeup probably help.
People keep acting like it's an issue on social media but I find it less of a problem then it's ever been.
The blast radius you just described is generally inaccurate though, the runtime still has access to whatever's currently executing which is going to typically be things like database access
An RCE that is exploited due to say a sloppy node package that gives you access to eval is going to typically give you access to the database for the website because that is typically not running isolated in most application designs
What's the likelihood that somebody with absolutely no experience can own that security stack on their own and know to update all their dependencies on their workers and keep track of all the cve?
Remember a lot of people vibe coding this type of stuff don't even know to use GitHub so tools like dependency bot aren't generally available or are very ignored
And you still have to interact with it and update those just like you have to do on WordPress
Right but most WordPress hosts isolate processes most of the time these days, for WordPress generally people aren't worried about crossing boundaries and additional exploits they're worried about the data that the process has access to which is the same security threat you're going to have under a worker on cloudflare
The OS is pretty irrelevant when people are manipulating and dumping your database which can happen just as well on a worker
The original argument is that you don't have to care I'm saying you do, but now you're responsible for the code instead of just clicking a patch button
Except most WordPress installations don't manage their own infrastructure these days so that also matches
PHP and WordPress is a tire fire but you're still responsible for told you deploy if you're putting dynamic code on workers
It doesn't matter when sensitive data gets destroyed or leaked or modified if it's WordPress or a worker in a container.
And the kind of people that make WordPress websites? Yeah I'm not going to trust them to not create exploits, I've seen llms drop exploits in, and they don't have the experience or skill set to catch it.
IDC, I deal with UIs all the time. Just stop lacking basic functionality. I had this issue with Mantle, I have this issue with Shopify. I've had this issue with https://t.co/xvEk8CHJ64.
Terminal obsession is a common phase a lot of engineers go through, I did like over a decade ago.
The problems and frustration come from poor system design, not whether the UI is a web page or a CLI.
It's the pragmatic take.
1) Most of the time, people making these posts run an AI company
2) They rarely are any significant size, chasing $20 and $40/mo subscriptions on piles of labor building and maintaining tools
3) It's mostly about the social media attention
4) Actual companies don't have time to fuck around with this, or the cost of letting some random guy vibe a platform and fuck up their data (when the person should be doing their main job).
On the open landscape: almost no one is doing this, and even the overhead of just determining what you want to build, migrating data, and maintaining that platform is typically more of a problem then it's worth vs. just paying another company to run it.
With every new fad, it's popular to effectively be an idiot online, these are the kinds of people that were screaming NFTs were the future. They're desperately hoping to be on the next bandwagon early so they can win big on it, but this isn't how you do that.
@samongaro_ It would be like refusing to use Intellisense, or work with Node. It's random hills to die on that I can't really justify, we're a high performing team and I don't have room for silly behavior.