Soft Skills Engineering - Episode 342: Losing my job to AI and bad review season

Episode Date: February 6, 2023

In this episode, Dave and Jamison answer these questions: Hello Dave and Jamison, thanks for your great work. Your podcast has the bizarre magical property of making me look forward to long drives. Keep it up! I have been feeling anxiety over losing my job to AI, especially after the all the ChatGPT stuff from a few months back. I know that it definitely isn’t flawless but I know that this technology will just keep improving as time goes on. I am a software engineer with 2 years of experience. I can’t help but feeling like I will lose this amazing career in the near future. I left my old line of work a couple years back and am in my mid 30s, so switcyhing careers again is a dreadful thought. Is there anything you can suggest to ease my anxiety? Will being more social with my coworkers, or aiming towards management help reduce my chances of being automated? Any advice will be great, thanks. PS: If someone tries to replace your podcast with an AI generated one I will boycott them and stick with you. It’s review season! I am an IC software engineer, and I am required to document my impact for the last year. However, I work on an auxiliary team/new business team that is always trying to find new use cases for the existing product platform. If you look at the numbers, the impact is very low compared to the core business. Also, my team was disproportionally impacted by layoffs late last year. Lot of folks with institutional knowledge and good relationships with the core team were let go which disrupted our team and contributed to missed deadlines. How do I write my review for this bad year, with little to show for it?

it takes more than having chat gpt write your email rescinding your acceptance of a job offer to be a great engineer this is soft skills engineering episode 342 i'm your host dave smith i'm your host jameson dance soft skills engineering is a weekly advice podcast for software developers about all the non-technical stuff that goes into navigating this career such as how to use computers to write emails to human beings which will then be interpreted by computers and summarized so that the recipient can save time that's cybernetics i saw a great meme this week which was three images it was like the the image it was someone sending an email to someone else and the original image was like write an email saying i'm sorry for this or that and

then the ai generated this long flowy beautiful message like with deepest regrets and then on the other end the email came through and it was like summarize this email it's like joe says sorry it's like we're creating this like 100x data increase so that the ais can talk to each other and interpret for us yeah someone gets paid for all those tokens generated that's right that's right oh anyway i thought that was funny dave do you know what i want to do right now i do but i want you to say it okay that's also what i wanted to happen right now how i wanted you to reply so all is according to plan i want to shout out our patrons that contribute at the level that we do this every week i want to thank andrew realis connie lee valentin and datafold santa hopar

noah frazier loge kensi dodds jenny kim owen chardell craig motlin i love mavis the stochastic parrot alice jost land fair quill quinn go go go go go go wow i feel like it's been a while since i've tried yeah you pushed all the way through most of the letters yeah i feel like i owe it you know they're okay getting their money's worth kashokton ohio patreon.com.au were hiring ira chan monkey face emoji jonathan king testing is documenting.org oladapofadie will angel agner nick hathaway travis sanders brayden canes john grant bartek takowski cody sale nick cantar and philip john basile thank you thank you thank you we appreciate it we show our appreciation by saying these words and also by you know how in in cartoons when you snore the

snore noise is like it's like something like that it's like oh okay like it has a little music to it little tonality yeah yeah there's like the the snore noise i am aware of that and now i'm very eagerly awaiting how you're going to tie that into this connie lee that's what we do we we repeat these as our snore noises that's right yes i i forgot about that but i do do that yeah i mean it's been such a habit that yeah i mean i it's like breathing i don't even think about it just happens yeah yeah also if you contribute any dollar amount by going to soft skills.audio and click support clicking support us on patreon you'll get an invite to our slack team and we found a new thing to spend money on we bought the yachts and now we are publishing episodes on

youtube which has some editing and various costs associated with it but hopefully is a way to reach more folks so thank you to to people for supporting and and it helps the show thank you yep the yacht we just couldn't spend all the money on yachts we ran out we there was too much money left at the end of the yacht i mean once you have five the the marginal value of a sixth yacht is pretty low it's low it's low yeah oh shoot my other five yachts are all in the yacht shop good thing i have my sixth yacht that doesn't happen that often not often enough to justify the hundreds of millions of dollars should we read a question yes i would like to do that i also wanted to say we have over 900 people in the slack community now so if you join you get access to 934 amazing

people who smell great there's no question about that the smell the smell in there is just so aromatic it's great all right well that's also like joining makes you smell great if you don't already so yeah that's part of how we can guarantee that right okay so this question comes from an anonymous listener who says hello dave and jameson thanks for your great work your podcast has the bizarre magical property of making me look forward to long drives well said continuing i have been feeling anxiety over losing my job to AI, especially after all the chat GPT stuff from a few months back. I know that it definitely isn't flawless, but I know that this technology will just keep improving as time goes on. I'm a software engineer with two years of

experience and I can't help but feeling like I will lose this amazing career in the near future. I left my old line of work a couple years back and I'm in my mid thirties. So switching careers again is a dreadful thought is there anything you can suggest to ease my anxiety will being more social with my co-workers or aiming towards management help reduce my chances of being automated any advice would be great p.s if someone tries to replace your podcast with an ai generated one i will boycott them and stick with you how do you know that hasn't already happened yeah i was gonna say unless it's so good that it's imperceptible yeah oh this is a Great question. Very timely. I've been thinking about this question a lot from actually a very

different angle. Not, am I going to lose my job? But rather, how can I become much, much more productive by using AI-based tools? What's the answer? So far, no. I got my pink slip last week and they've replaced me with an AI. It wasn't a yes or no question, but the answer is definitely no. Right. How much more productive can I be? Answer, no. Well, it's, I mean, this is an extension of the classic problem of developers where if you spend enough time automating a thing, you'll become more productive at that thing. But then there's, it's got to, how long will it take to pay off? How much time can you waste looking at AI at work to become more productive? And how long will it take for that to make up for the fact that you wasted all that time? For every limerick that I ask chat GPT to generate, I become that much more productive at writing Java code. I do have one idea, which it's like, write this thing and make it rhyme.

That's 99% of my use case for using these large language models so far. Which is a really cool use case, actually. I asked it to write in iambic pentameter, and then I realized that, I mean, you have to verify these things because they lie seamlessly. Very, very well. I don't know what iambic pentameter is. I was like, I don't know. This looks iambic to me.

Yeah, it kind of has an iambic vibe. Yeah, I see the iambs in there. Whoever came up with the word iambic pentameter is just a legend, though. So I'm pretty sure it means stuff. Oh, yeah. It's something to do with five. There's like 10 syllables in each line and an alternating emphasis on each syllable or something like that. But I still couldn't tell if it was written that way or not.

Because I need an AI that I trust to tell me if it's written. That's right. You need another AI to check the work of the first AI. And they're both like, yep, looks good. Yeah. They have each other's backs in solidarity. They're actually the same entity under the hood. Do you think there might be a part of our audience that doesn't know what chat GPT is?

You know what? Probably worth assuming there's someone out there. Yeah. Do you want to explain what it is? Sure. So I'll pretend like I know what I'm talking about here. So we have this relatively new concept at scale of large language models using a technique that I believe they call stable diffusion or unstable infusion.

I don't know. It's something with fusion, cold fusion. That's what it is. Iambic stable fusion amateur, I believe, where they have trained these machine learning models on large corpuses, corpi, as the AI tells me, of human speech or human text, spoken language text, you know, and they have trained these things to now produce plausible responses to very common human questions or not common, but just naturally express human questions. Things like my favorite one I've done is, hey, write me a seven day meal plan, trying

to stay in under this many calories per day, three meals a day and using this percentage of protein, this percentage of carbohydrates, et cetera. And it will produce a plausible meal plan and it will fully understand your question in text, of course. And then you can say, now make a shopping list for that meal plan and it will do that. It's very, very, very, very impressive technology. The fear here is that this technology will be able to produce code that can replace human software developers, which I think is interesting and quite, you know, it's a possible outcome in the future. One of the ways that we've seen this incarnated is with a product called GitHub Copilot, which I've been using. It is also a GPT model, but an earlier

generation of GPT model from the current chat GPT that's taking the world by storm. And it will auto-complete large blocks of code. You can take a comment where you write a comment about what you want the code to do, and it will produce code that does that or attempts to do that. It's not perfect, as our question asker says, and it makes a lot of little mistakes, but it also saves you a ton of time time like searching documentation and things like that so i don't want to shill for github copilot but that i think it's those kinds of products and chat gpt in general that are making engineers wonder if they'll be replaced by technology was that a fair sum up jameson i was googling instead of listening but it sounded good did it sound plausible because that's all i'm

trained to do it gpt3 was trained on 45 terabytes of text which is a lot of text that's a lot of terabytes yeah sounds terrifying yeah they just kind of like threw the internet at it i have talked to folks that use github co-pilot specifically and have also played around myself with using gpt3 for code generation and it's it's incredible i don't know how else to describe it it can't take vague requirements like if you don't know pretty clearly what you want it won't figure out what you want very well but if you if you know like okay i need to reduce this array down to i don't know an object with these keys like if you could say the words in english of what you want the code to do it's pretty solid at writing code to do that thing yeah like it has

it has a lot of limitations of course but it's it seems to be on a growth trajectory that is going to become much more impressive in the future unless we top out here i guess but i don't think we are yeah but will it replace your job yeah and that that's a great question and i i my personal opinion is that like i don't want to wax too philosophical but i think that people who say no this will never replace me are wrong and people who say yes this will replace all engineers are also wrong. My prediction for GPT and AI-based software development technologies is that it will be a productivity boost to human beings. And there will also be a whole bunch of style of code that people just don't need to worry about writing anymore because the machines can write it for

them. So I think what will end up happening is we'll start working at a higher level of abstraction where we are generating instructions for machines that previously would get translated to machine code and now are going to get translated from english to machine to high level code languages and then and then to machine code so we're basically just adding another rung on the abstraction ladder for the for where we work i think and it'll be more productive and faster and less error prone so you already have heard this lament for a while about how programming today is just uh cobbling together these lego blocks and you don't really build stuff don't really understand what you're creating or it seems like if you are prone to that type of thinking

this this feels like a large step further in that direction where now you're not only you're not even like picking up the legos and plugging them together you're saying hey computer yeah pick up these legos and plug them together yeah i'm trying to think of some implications of that, which I suspect it will be a lot easier to get further without really understanding what is going on. But also, I mean, a thing with these language models in general is their idea of correctness is pretty different from a human's. It doesn't really, I don't know. Correctness isn't a thing. It's like, how do I put this? You need to know enough to check to see if the output is correct that's right because it'll look plausible like um if if you tell it write me this code in

a language i don't understand um it probably knows that that programming language better than you do and like if you've glanced at a tutorial then sure it looks i'm sure it'll look fine just like iambic pentameter i'm like yeah that looks like i am exactly yeah yeah sure that yeah like i would believe that writing that yeah so i wonder if it'll like does it require just as much investment to learn enough to be able to verify as it does to be able to create but you'll just be able to do more with that investment or will it like you'll create more stuff but you still need that background of of like you you need to have been able to write it yourself to check to see if it works you just won't be writing it yourself as much anymore and and that kind of that i believe

that what you have just described is the current state of the art today you can today ask these models to generate code, but it is never 100% correct, except for the most trivial cases. And so your verification is still needed, which requires a high level of training, which requires a human eye. So that is the state today. I think the question is, what should we be looking for to know that these systems have gotten so advanced that actually the human verification is no longer necessary, and in fact, just slows us down, slows the machine down, and introduces errors? It's kind of like the self-driving people say, all we have to do is be slightly better than the average human driver and we've won you know and i

think that's probably where ai for software engineers also takes over so i'm looking for indicators to see like all right is a non-developer non-trained not even self-taught software developer creating software applications that are getting usage whether for themselves or for a larger market of people that's that's like the tipping point for me is if that ever starts to show like it's going to happen, then I think jobs are on the line. Software engineering jobs are on the line. Because if people can produce software without software engineers, why wouldn't they? But I haven't seen that. And so far I've seen it be, it's almost like, look, I used to dig trenches with a shovel. Now I dig trenches with a tractor and I can dig a hundred

times more trench in a day. And so I'm actually more valuable, but I had to learn how to use a you know so yeah it's like it's still the it's still i'm still the trench digger you know i still dig trenches and regular people aren't out there digging trenches but if it ever if ever a day shows up where someone can just wave their arm and a trench appears with no one else with no training then now jobs are on the line right yeah i hope that wasn't too weird i was thinking about that and no no that makes a lot of sense to me i was thinking about like screwdrivers versus drills yeah it's the same type of thing of like yeah do more uh now i can yeah do more faster but i i don't know i mean if you needed a hundred times as long to install these screws doesn't that kind

of indirectly mean you need fewer people to man the drills could be or staff the drills yeah it could be and and that's okay i think in today's market that's okay because there's such there's so much unmet demand for software production right now, because there aren't enough people to produce it, that having all the people who do produce it be 10x or 100x more productive might actually bring this market into balance a little bit and have some of that demand be met. So personally, I'm optimistic. My short answer is no, I don't see any jobs on the line right now. But I do look forward to a day where we are all way more productive, because we're using AI assistance to generate our code faster and with fewer errors.

had an important insightful thing to say well i know i know one way to get that insight to come oh yeah yeah yeah even the hint of the next question i did it brought it back out what i have seen at least two examples of developers building systems on top of of these language models that basically say generate this code execute the generated code and then say here's the results like if there's an error fix the error oh so that's so cool that's and it's it seems to work fairly well if it's something that you can kind of iteratively try out yeah so that's a cool workflow that is so cool i love it and i think we should be embracing that as developers i think the absolute wrong thing to do at this time is to go oh ai bad hurt jobs you know cost money no like

bring it on let's let's see if we can make ourselves more productive the lisp people were right you just need to have like a repl based workflow and then you stick in ai where the they were right about having a repl they were wrong about the syntax yeah well i have i have one more thing just just just to reinforce my point that i think it's going to make us more productive is that i actually took a big block of code this week and i pasted it into chat gpt and i said explain this code for me and it produced a very succinct nice explanation of the code where it called out the most important functions in the code so that I could go look at those and then be like, aha, yes, it's right. And it kind of helped me focus. So instead of just reading

top to bottom line by line by line, it said, hey, this is a class interface. And these are the three most important methods. And here's what they do. And I'm like, aha, I could go kind of figure that out without having to be distracted by noise. It was like 700 lines of code or something, right? So it would have taken me some time to read all of it. And you don't actually read code line by line. It doesn't make sense if you do that, right? Because the call sites and call order is, is different. And so you actually, it's nice to have a starting point. So there you go, just one more example of how it's making me more productive, and saving me time to get a job done. And I think we're going to see a lot more of that. And I'm leaning into it. Actually, I've,

I've, I demoed some of these tools to my teams over the last couple of weeks, and I'm going to make it mandatory to use to have some of these tools installed in their IDEs. Because I'm like, just having it there is enough, it's worth a few bucks a month to have every developer save, even if it's just 10 minutes a month, I'm happy to pay 10 bucks for those 10 minutes. What I would love is the ability to just hand a code base to it and then interrogate it for things about the code base. Because I could see that working well with the existing systems if it's a self-contained chunk. Or if you write your whole code base in one file, maybe that's an argument. Imagine a world where you can point an AI at a giant legacy code base and say,

okay here's my code base the product manager has asked me to add a feature that looks like this and then just paste a link to the to the um product spec that the product manager gave you and say where should i go to start working on this and what what edge cases should i be aware of what failure modes should i be aware of and just imagine it being able to go do that research on a hundred thousand lines of code and then come back to you and say here are seven gotchas you should be aware of like this is a productivity booster huge so i'm i'm super on board you know i am totally plugged in and ready to go ready to embrace this future um and but i also don't think it's going to cost jobs i think it's going to make us more awesome my my boss has been talking about

cybernetics a lot recently as a way to think about this where i i can't remember the exact definition they used but i'll make one up good i hoped you would oh just you mean just like a chat gpt yeah exactly uh it's it's tools to augment human capability so you wouldn't you wouldn't say computer what should i do you would say like computer i want to accomplish this task help me do it faster better more accurately help me do the thing yeah not not like figure out what to do for me necessarily that's right although i guess that although why not have a fuzzy boundary computer you here are the things i like what should i do today for to maximize my happiness yeah all right now we must go to the next question i think you're up for this one yeah i'll read it this is from an

anonymous listener who says it's review season i am an ic software engineer ic stands for individual contributor so not manager and i'm required to document my impact for the past year however i work on an auxiliary team or a new business team that is always trying to find new use cases for the existing product platform if you look at the numbers the impact is very low compared to the core business also my team was disproportionately impacted by layoffs late last year lots of folks with institutional knowledge and good relationships with the core team were let go, which disrupted our team and contributed to missed deadlines. How do I write my review for this bad year with little to show for it?

Well, I've got some good news for you. There are these great AI-based systems that you may have heard of that can create plausible reviews. In fact, I saw a Twitter thread where someone said, I've been asked to justify my existence at my company. help i use chat gpt to fabricate examples of contributions i've made that are plausible and just technical enough that no one will investigate whether they're real and it came up with remarkably good descriptions of work for an engineer so if your ethics are a little

questionable that's one way you could go i used it to write my product manager a rhyming self-review it was pretty good oh my goodness didn't have a lot of context about their specific contributions but i don't know some self-reviews don't but it rhymed but it rhymed and that obviously was the part your product manager will remember exactly it's up to them to choose to decide whether they will use it or not but i provided them the opportunity okay i survived that's what you put on your self-review or your review or whatever, right? I survived. I made it. Everyone else was laid off. So I must be doing something right. I'm still here. Yeah. So they mentioned numbers impact low compared to the core business. This feels a little bit like maybe there's a

business with a, maybe they're kind of Google shaped where they have a single extremely strong source of revenue and are reveling in that but also kind of worried about potential disruption in the future of if you have one source of revenue then that going away means you're toast so they're kind of like hedging their bets a little bit i guess i don't know could be the impact is very low compared to the core business i i have questions and no answers but so it sounds like you have some idea of measured impact do you have an idea of expected impact because maybe that's fine yeah like maybe the the measured revenue generated is pretty low but that's on track with the targets that you're setting because the expectation is this will kind of ramp up slowly

over time that could be or maybe it's not maybe it's horrible you know that'd be that'd be good to know not just how did it do absolutely good to know so you cannot mention it in your review or mention it depending on how depending on what you find out i think in a world where your impact is limited to the business by virtue of your position in the organization, the way that I would focus my review is based on demonstrated proficiency and competency and contributions within that limited potential so that I look more high potential to my employer despite those limitations on my contributions. But I would not dwell on the limitations and share a bunch of reasons. Well, here's why I didn't make a big impact this year. Well,

these org chart or reorgs happened and people got laid off who were really important. That is not at all relevant to your performance within your sphere. And all you can be expected to do is perform well within the lane that you've been placed in by your leaders. Now, I'm going to wax philosophical here a little, but you've definitely been potentially the victim of some circumstances. But I hope that this lesson, you've taken away a lesson from this, which is that you have been in a low impact area and that has hurt you. And now that's coming clearly into focus. If you see things like this happening in the future, it's very good idea to try to navigate around them so that they don't become a limitation for you as hard as that can be so

get to a higher impact area or or see if there's work you can do that helps your area have higher impact kind of is that what you think so and easier said than done right yeah yeah i'm sure the business would be pumped if if uh your area didn't produce a lot of revenue and then you said i'll change it and then you did yeah they'd be happy about that so you said don't mention it i think it is worth i don't think you want to whine about it definitely not but i think it is worth saying something to to help provide the context for what you've delivered because it's possible that the folks reading your review haven't thought about it as much or it doesn't affect them as much true so i would still put in like a little blurb about like despite challenges

of i don't know you don't want to sound bitter but but acknowledging like my whole team was let go and you know like uh and i and i managed to do like this despite that yeah i so the way that i would couch i would count yes that's what you just said at the very end is exactly what i think is a valuable way to couch that which is not look at all these bad things that happened to me and i survive, but rather, I chose to operate professionally and supportive of the organization despite headwinds. And here's how I worked around each headwind. Here's how I demonstrated creativity in the face of restricted resources available. And that's maybe the spin you're looking for. Look how creative I am.

And the little sailboat. Yeah. Sorry, say again. Well, you said, look how creative I am. I said, draw a little picture of a sailboat with headwinds, which I think also would demonstrate how creative you are. Look at the sailboat. It's going forward. Yeah. how do they go forward when the wind is blowing backwards yeah it's incredible it's incredible that's what reviews need more of is illustrations that's right yeah i i like that i mean part of the

part of the possibility or danger here is you worked really hard and delivered great results on something that was canned and never shipped yeah and and i don't yeah i don't know how to account for that because well that's the wrong way to put it you know i'm just doing the time honor tradition of saying what you said but worse yeah you can you can still say look at all this great work i did with the caveat of like circumstances meant that we never launched the thing but like i delivered really well on it while we were working and maybe you don't even mention that it never launched just talk about all your great contributions to make it successful no seriously like yeah you all know about the famous project icarus right we don't it needs

no no mention of its impact needs to be made because we're all so familiar but look at my contributions to it it would not have been as successful as it wasn't without my contributions yeah i do think that's the right the right framing in this situation that's what i would do like i'm i'm thinking about your manager's position where your manager's thinking yeah i understand that a lot of team members got let go. You had limited access to resources, limited access to the core team. So how did you perform? And I think you need to highlight, look, I'm professional. I support this organization. I kept a positive attitude. I refuse to consider myself a victim. And you're basically painting yourself as a team player who pushes through challenges, regardless of where

they come from, whether they were in your control or not. And then that will hopefully paint a nice picture for the future so that you appear and hopefully you are a good contributing member of this team. Because guess who, guess who doesn't get a performance review like that? And that's the CEO of your company. Your CEO doesn't get the option to say, yeah, well, we had some headwinds. So what am I going to do? You know, someone else's fault. Like your CEO has to fight through those headwinds and come out the other side. So try to, I, when I do this kind of thing, I try to put myself in the mindset of the people that are receiving the review. And in this case, the mindset is i can't like i'm not going to reward excuses you know i like i would like to i feel

bad for you and i have empathy for you but at the end of the day your excuses aren't going to put your teammates salaries in their bank accounts you know so i want people who are willing to step up to the plate despite limited resources and challenges and show a really positive attitude and then hopefully motivate the whole team to do well so that that's where i would go with this that'll probably be a breath of fresh air to your manager frankly because i'm sure everyone else is complaining and frustrated and and having someone who's like look i'm here to win and i'm here to help us all be winning win together i think that would be really nice to hear maybe you're the only one left maybe there's no one else to compare the reviews against true yeah i like it good advice

dave great job you see an ai could never give that much nuance and context and and yeah it could though if you just said give more nuance in your answer and then okay now add some fake context that sounds like good advice but isn't all right we've done it we've answered the question okay all is well for now until next week when more questions will come yes like who's the person that held up atlas atlas is the guy who holds the world up right yeah is that you in this picture i think it's us together oh like one shoulder on each side yeah yeah like if we miss an episode then except the world constantly falls off and we have to pick it up again and and actually we're not we're not actually not holding it up it's holding us up is that yeah

that's okay all right let's get out of here what can people do if they want their own questions answer to soft skills.audio and click the ask a question button where you can fill out our handy dandy form thank you so much to everyone who does that we love reading your questions every week and we solemnly swear we are not training an ai to replace question generation but we reserve the right to do so in the future oh okay we solemnly swear that as of this moment we have not yet done it and that's only because we don't know how to yet all right thanks for listening we'll catch you next week