I just used ai to do three things that were a bit outside my skill and comfort zone. I’m pretty sure there are lots of people using it to good effect. Actually two things that were totally outside and one thing that is well within, but would have taken me eight hours and with me directing it took the ai about 12 min.
Most software engineers I know use some amount of ai assistance in coding.
Pretends is a pretty strong word here. A lot of people actually use it to help them do their work.
Didn’t read the article, and I’m an engineer who is using too many genai tools personally and professionally, but the title spoke to me: I am 100% pretending that my professional artifacts came from genai tools.
Part of it is that “what good looks like” - from leadership - looks a certain different way right now, thanks to LLMs. The other part is that knowing isn’t enough in a large org, you have to show.
We are embracing this SO HARD, I need my communication to look like this, and most importantly my teams communication needs to implicitly show that we are bought in (to accompany the explicit proof, measured in token kpis - not kidding).
Not in the US, but I exaggerate my use of AI to the CEO because he has fully taken on board the idea that AI can do everything, except presumably his own job.
I do use AI, but I make out it's a bigger part of my workflow than it is to appease him.
I'm trying to figure out ways to placate the boss by fitting AI into the edges of my workflow.
"Look at these commits and do a review" to find stupid stuff like forgotten exception throws or nulls. Or "Critique this documentation for a different audience"
I don't want to get into the "let the machine write the code" phase because that's the task I enjoy most, and it swaps my efforts into review, which I'm bluntly less confident about (having to follow and second-guess an architecture without being "there" when it was evolved increases the chances I'll miss stuff)
The stuff that AI demos well at-- "refactor 5,000 lines of code", "build a new client from ground level" are simply not what my team works on; we end up doing things where building a prompt to actually make the change we want-- and only the change we want-- takes longer than writing the code itself. 80% of the time is the debugging and planning.
Some use AI detection to be sure that you're not cheating. Others use AI detection to make sure you're doing your job. "This is terrible, bring it back to me when it's slop!"
We recently had a poll at work about what we were using AI for, and there was only a single vote for "I'm not using any AI for ky work", and that was mine.
Maybe I should share this article in that chat; then I might seem less alone.
Yeah I do this too. I'm involved in a GenAI (copilot) project. And I know we check how often it's used so I just say hello to it every day. Because I can't be seen to be at the bottom of the list.
I don't think genai as currently included in ms office is very useful. Simple things it can do like writing emails but in such a stupid formal way that people instantly know it's not me because it can't clone my style. It's more work to tell it how to write an email than to do it myself. And I can read fast so I don't need summarisation of emails.
More advanced things like complex Excel modifications, things I could actually use help with, work badly. From "there's been a problem, please try again later" to just doing completely stupid stuff. Creating a PowerPoint? Yeah it can do that but it's a one shot thing. I can't say "ok nice start but change xyz". It'll just tell me to do it myself in a passive aggressive way. I can restart the generation but it's like rolling the dice because every time it gives a completely different design. It's cool for demonstrations, completely useless for actual work. Everything else just costs me more time than doing it myself especially because I have to verify to make sure it didn't hallucinate.
The only thing I do see value in is finding stuff back. Look through my emails and find that dude I spoke to about buying new company phones two years ago, that kinda stuff.
There's some promising new features coming like the researcher (reasoning engine) and analyst, and it's slowly starting to become somewhat useful. I'm sure it'll get there.
But right now I don't bother with it. So I just fake it. But it's a bit disheartening when everyone wants to go full steam ahead with something rusty that's still so rough around the edges. But everyone is crazy about getting on board this train.
I do have a big AI server at home. While it's even worse performing, I don't need to trust big tech to use it which is a must-have for any kind of personal use for me whatsoever.
11 comments
[ 15.7 ms ] story [ 411 ms ] threadMost software engineers I know use some amount of ai assistance in coding.
Pretends is a pretty strong word here. A lot of people actually use it to help them do their work.
Part of it is that “what good looks like” - from leadership - looks a certain different way right now, thanks to LLMs. The other part is that knowing isn’t enough in a large org, you have to show.
We are embracing this SO HARD, I need my communication to look like this, and most importantly my teams communication needs to implicitly show that we are bought in (to accompany the explicit proof, measured in token kpis - not kidding).
I do use AI, but I make out it's a bigger part of my workflow than it is to appease him.
"Look at these commits and do a review" to find stupid stuff like forgotten exception throws or nulls. Or "Critique this documentation for a different audience"
I don't want to get into the "let the machine write the code" phase because that's the task I enjoy most, and it swaps my efforts into review, which I'm bluntly less confident about (having to follow and second-guess an architecture without being "there" when it was evolved increases the chances I'll miss stuff)
The stuff that AI demos well at-- "refactor 5,000 lines of code", "build a new client from ground level" are simply not what my team works on; we end up doing things where building a prompt to actually make the change we want-- and only the change we want-- takes longer than writing the code itself. 80% of the time is the debugging and planning.
Maybe I should share this article in that chat; then I might seem less alone.
I don't think genai as currently included in ms office is very useful. Simple things it can do like writing emails but in such a stupid formal way that people instantly know it's not me because it can't clone my style. It's more work to tell it how to write an email than to do it myself. And I can read fast so I don't need summarisation of emails.
More advanced things like complex Excel modifications, things I could actually use help with, work badly. From "there's been a problem, please try again later" to just doing completely stupid stuff. Creating a PowerPoint? Yeah it can do that but it's a one shot thing. I can't say "ok nice start but change xyz". It'll just tell me to do it myself in a passive aggressive way. I can restart the generation but it's like rolling the dice because every time it gives a completely different design. It's cool for demonstrations, completely useless for actual work. Everything else just costs me more time than doing it myself especially because I have to verify to make sure it didn't hallucinate.
The only thing I do see value in is finding stuff back. Look through my emails and find that dude I spoke to about buying new company phones two years ago, that kinda stuff.
There's some promising new features coming like the researcher (reasoning engine) and analyst, and it's slowly starting to become somewhat useful. I'm sure it'll get there.
But right now I don't bother with it. So I just fake it. But it's a bit disheartening when everyone wants to go full steam ahead with something rusty that's still so rough around the edges. But everyone is crazy about getting on board this train.
I do have a big AI server at home. While it's even worse performing, I don't need to trust big tech to use it which is a must-have for any kind of personal use for me whatsoever.