It’s getting really bad. The software engineers I work with have been telling me that they now have their coding agents running 24/7 and it sucks for them because they never really clock out anymore. They know that if they don’t periodically check in and set the agent on to the next task, or solve some glitch, that it will only sit there for 8 hours until they come in next day and deal with it, and then they’ve lost that 8 hours. They’re able to do a lot with AI but it is not always fast. So they feel pressure to babysit their AI task flows all the time.
My thought was Jesus Christ what kind of energy is it consuming for these things to be running like that nonstop. I’ve stopped myself from using AI to look up one fact because it would be a waste of energy. But these guys have agents running agents running agents and they’re just crunching and crunching constantly.
It’s effective in terms of cranking out software. I’m talking about skilled senior engineers managing this directly. They know what they’re about. But at what cost?
At least they’re still trusting their software engineers to use it. My employer hired a bunch of people for their “AI taskforce”, the leader of which can’t even be bothered to use a password on his vibecoded SQL database full of sensitive company data. And you can bet your ass he’s getting paid three times as much as anyone in the IT department.
Best thing is that they’re expecting us to take care of their mistakes.
We went through that phase. A couple of vibe-coding douchenozzles had our management convinced that everyone can ship code now. They launched a whole initiative to get product managers and UX designers deploying. It failed. Then they dialed it back to “cosmetic fixes only” that aren’t worth an engineer’s time. Now they realize that having uneducated PMs using AI to ship code is actually slower than that PM asking an engineer to use AI to ship code. So we’re back to having distinct functions again. All that really matters is that someone in the chain is using AI to accelerate the process, and the engineers turn out to be so much better at it than anyone else that we now just let them work.
It’s effective in terms of cranking out software. I’m talking about skilled senior engineers managing this directly. They know what they’re about. But at what cost?
Those senior engineers became skilled by starting out as entry-level engineers who didn’t know all that stuff, but learned from the senior engineers before them (and by writing a lot of bugs that hopefully got caught by code reviews.) Now, companies are using AI as an excuse not to hire entry-level people.
15 years from now, we will find there are no mid-level people to promote, because they never got their entry-level job and are now waiting tables.
I said it was effective at cranking out software, not at training the next generation of engineers. However obviously the terms of engineering are changing so it would also be a mistake to automatically think we should train them exactly as we did before. Some people saw compilers as the same thing: it’s an abstraction layer! How is anyone going to know what’s actually happening in the CPU anymore?! Well, they don’t actually need to.
Using AI is actively harming people, stop using AI
It’s getting really bad. The software engineers I work with have been telling me that they now have their coding agents running 24/7 and it sucks for them because they never really clock out anymore. They know that if they don’t periodically check in and set the agent on to the next task, or solve some glitch, that it will only sit there for 8 hours until they come in next day and deal with it, and then they’ve lost that 8 hours. They’re able to do a lot with AI but it is not always fast. So they feel pressure to babysit their AI task flows all the time.
My thought was Jesus Christ what kind of energy is it consuming for these things to be running like that nonstop. I’ve stopped myself from using AI to look up one fact because it would be a waste of energy. But these guys have agents running agents running agents and they’re just crunching and crunching constantly.
It’s effective in terms of cranking out software. I’m talking about skilled senior engineers managing this directly. They know what they’re about. But at what cost?
At least they’re still trusting their software engineers to use it. My employer hired a bunch of people for their “AI taskforce”, the leader of which can’t even be bothered to use a password on his vibecoded SQL database full of sensitive company data. And you can bet your ass he’s getting paid three times as much as anyone in the IT department.
Best thing is that they’re expecting us to take care of their mistakes.
We went through that phase. A couple of vibe-coding douchenozzles had our management convinced that everyone can ship code now. They launched a whole initiative to get product managers and UX designers deploying. It failed. Then they dialed it back to “cosmetic fixes only” that aren’t worth an engineer’s time. Now they realize that having uneducated PMs using AI to ship code is actually slower than that PM asking an engineer to use AI to ship code. So we’re back to having distinct functions again. All that really matters is that someone in the chain is using AI to accelerate the process, and the engineers turn out to be so much better at it than anyone else that we now just let them work.
Those senior engineers became skilled by starting out as entry-level engineers who didn’t know all that stuff, but learned from the senior engineers before them (and by writing a lot of bugs that hopefully got caught by code reviews.) Now, companies are using AI as an excuse not to hire entry-level people.
15 years from now, we will find there are no mid-level people to promote, because they never got their entry-level job and are now waiting tables.
I said it was effective at cranking out software, not at training the next generation of engineers. However obviously the terms of engineering are changing so it would also be a mistake to automatically think we should train them exactly as we did before. Some people saw compilers as the same thing: it’s an abstraction layer! How is anyone going to know what’s actually happening in the CPU anymore?! Well, they don’t actually need to.
Okay but how does that effect next quarter? I need line up now