Develop Yourself
To change careers and land your first job as a Software Engineer, you need more than just great software development skills - you need to develop yourself.
Welcome to the podcast that helps you develop your skills, your habits, your network and more, all in hopes of becoming a thriving Software Engineer.
Develop Yourself
#248 - The AI Productivity Myth: Reality vs. Hype
Check out this beginner project to use AI with HTML and JS 👇
I'm no AI hater but I am getting so tired of the dialogue online I read from people who claim that AI is making developers 2, 3 or 10x more productive.
Google's CEO recently came out with a sobering statistic:
AI has made their engineers 10% more productive. (source)
10%.
This is still massive but no where near the gains we were promised or what our managers are expecting.
Let's explore this myth of productivity and what you, as a developer, can and cannot do with AI.
Shameless Plugs
🧑💻 Join Parsity - For career changers who want to pivot into software.
✉️ Got a question you want answered on the pod? Drop it here
Zubin's LinkedIn (ex-lawyer, former Googler, Brian-look-a-like)
Welcome to the Develop Yourself podcast, where we teach you everything you need to land your first job as a software developer by learning to develop yourself, your skills, your network and more. I'm Brian, your host. Sometimes I wonder if I'm just using AI wrong, because I'll see software engineers online talking about how they've 10x'd their productivity, how no one should learn to code, because now one developer can do the job of a whole team, and in some ways, they're right. Like I'm on a team of software developers where it's like me and a couple other people and we are vibe coding our way through a prototype, building it really, really fast. What used to take me a week now takes me a couple hours.
Speaker 1:Now, the key word there is prototype. Prototype is essentially just a proof of concept, something that you can slap together. That kind of shows people yeah, here's what I'm thinking, and you used to just draw this out. You might have a designer literally draw this out using a wireframing tool or Figma, or maybe literally hand-drawn things if you're living in the Stone Age, I don't know but nowadays, because of AI tools like V0 or Cursor or whatever, you can basically scaffold a web app really really quickly and for a lot of people that are looking at this and thinking, oh my God, what does this mean for me and my career? And I'll be honest, this was something that kind of freaked me out. I'm a mostly front-end developer, mostly on the web, and I'm thinking, holy smokes, this thing is coding not only good-looking code, but the code it generates is actually pretty good, and I started using vibe coding a ton at this new position I'm in. I was using AI tools previously at the company that I was at that I left, but it was nothing like what we're doing here and I was getting a little bit nervous. I was feeling a little bit like, well, wow, what value do I have to offer?
Speaker 1:Then reality sunk in the moment we moved from prototype or the moment we tried to add to the original prototype. I just saw how brittle and broken things got to add to the original prototype. I just saw how brittle and broken things got. And then I saw how we had diminishing returns on just vibe coding everything or asking AI to do everything. And then meanwhile we have CEOs and non-technical people saying, hey, why are things so slow? We just got through everything like in a day or two days or a week. We built this amazing looking multi-page website and now you're saying it's going to take a week to integrate an API or something. This is where reality meets the fantasy, and it's not pretty.
Speaker 1:And I see that a lot of developers haven't really met reality yet, because maybe they're not working on projects that require some level of complexity, or maybe they're not working within a team, maybe they're not deploying these apps, maybe they're just way, way smarter than me. That could also be very true. But let me tell you something I've worked with a lot of smart people. I'm working with a lead data scientist, a researcher, working with another data scientist, working with a CTO. I worked with a principal level engineer from Amazon recently and I'm telling you, none of them have 10x'd their output using AI tools. So are all of us just dumb?
Speaker 1:And then I spoke to a guy, dickie, who I've had on the show. He came to give a little talk at Parsity. We do these tech talks once a month and while he's during the tech talk, I asked him say hey, how do you feel like as a new coder about AI? And I was genuinely curious. And he's like well, you know, we're using it at work. We've been kind of told we have to use it. I'm like this is an interesting theme, right, and he's like but it's just not really speeding my work up. They keep thinking it's going to make super effective and the reality is it makes certain things really fast, but it makes other things either really slow or is just completely useless.
Speaker 1:And I even have a YouTube video that people hate. So much. People hate when I talk bad about AI, as if I hate AI. I don't hate AI. Let me make that clear.
Speaker 1:It's a tool that is incredibly useful. It has, honestly, completely changed the way I write code. Let me be very, very clear about that. It's like having a really, really dumb junior developer that you can cuss at. It won't get mad at you, they won't cry or tell HR or anything. I can cuss out my little junior developer and they won't even talk back. It's wonderful in that regard, but it's like a really, really dumb junior developer.
Speaker 1:That being said, it costs like 20 bucks a month. So, yeah, I mean, companies obviously want to try to cut corners wherever they can, but the reality is these tools it's aren't that great, and even Dickie is a pretty early career developer, just got his first job. He's over 40, by the way gotten to tech after 40 self taught, didn't go to Parsity. I still love his story and wanted him to come and speak to people at Parsity about what it's like getting into tech, but anyway, he's saying he's feeling the same thing and I'm like man, you know what? There's something going on here, and then this piece of news came out which I think you're really going to find interesting.
Speaker 1:Sundar Pichai, who is the CEO of Google, was recently on a podcast and he said are you ready for this? That AI has increased developer productivity by around 10%. This was not what people expected or probably wanted to hear. Now, this is Google. This is the CEO of Google. He is very, very incentivized to give a really high number right and the fact he said, increased by about 10% this should be sobering to many people out there who think that AI is the doom, the end of all white-collar work and software engineering work in specific. I found this really interesting because people are telling you no, no, no, it's just a replacement for software. No, it'll 10x your workflow. Now, meanwhile, google, who ruthlessly tracks these metrics obviously they're a massive corporation they want to know how much this is improving their software engineering output. So 10%, now, what do they plan to do that their engineers can now produce about 10% more code, they're going to hire more engineers. Why? He's like, well, now we can actually expand what our software engineering teams are able to do, so we're actually going to hire more of them.
Speaker 1:Now, this might not make sense because you're thinking, wait, what about all the layoffs and stuff like that? Why is it so hard to get a job if it's only increased by 10% and this guy's going to hire more? Well, something's got to be off. The math is not math and I feel you.
Speaker 1:Layoffs are complex. The news loves to talk about layoffs because for so long, software engineers were like untouchable. Have you seen those stupid TikToks or Instagram reels where it's like day in the life and it's some dude or woman or whatever drinking coffee like in the middle of the day, barely going to any meetings, writing like zero code, and then going home at like 2 pm and there's like I make half a million dollars. People hate that, right. I mean, they're easy to hate. So now that the tides have turned and tech is finally filling the layoffs, people are kind of dunking on software engineers. I don't really blame them, but there's so many complex factors that are going into this. And there's one major factor that no one's really talked about Interest rates.
Speaker 1:Right, when interest rates were low, right, companies were over hiring like crazy. I was part of a layoff and I'm a software engineer. But guess what? The layoff that I was part of was actually in a marketing organization, and when you look at the layoffs that have happened, look how many people are actually in software and of those people in software, how many were middle managers. There's a lot of bloat at big companies. So this is also a big part of layoffs overhiring during the pandemic, interest rates not being favorable and bloat.
Speaker 1:So this creates the perfect atmosphere for companies who are looking for a quick win. Not like companies are short-sighted or anything like that. Not like they'd overhire during a crazy once-in-a-lifetime event and be shocked that that once-in-a-lifetime event wouldn't translate into sustained and consistent growth. Not like they would ever do that, right? So AI is a very convenient excuse.
Speaker 1:It's like, oh, we're just going to reinvest in AI or oh, we're just replacing people with AI. Ibm did a survey and they've said something like 16% of software projects that are incorporating AI have, like, gone out to production. I don't exactly know what that means. But seeing 16% of like all projects that are AI related actually making it to the finish line, that's really really bad right. This is the problem. We've been sold on this dream of AI. There are multiple billion dollar companies and trillion dollar valuations happening that are telling us this is going to be the biggest game changer of our lives. I still think it is Like I'm not Mr Doom and Gloom and I'm also not Mr Like, oh, it's going to be nothing. I'm like this is a game changer.
Speaker 1:It changes the way that I fundamentally work. I use it constantly. I also know there's a lot of people that don't use AI at all or that their workplace has completely banned it. The reality is, we're going to use it. It's a wonderful tool. I pay for multiple services. I'm not some AI hater. I also have an incentive to tell you that it's not the worst thing ever because, let's be honest, I own a coding mentorship program. I am a software engineer. I'm hoping it doesn't take over my job. But at this current time, based on my usage and what I'm reading and what I'm talking to other developers and learning, I don't see this as reality. In fact, I see it as complete and utter fantasy.
Speaker 1:But then I run across articles like this from this principal engineer who worked at Meta and he was saying AI has made me so effective. I'm doing the work of three to 10 engineers. I have no reason to doubt this guy. I actually think he's pretty impressive. I'm not going to name and shame him because I don't think he deserves any shame, but I do feel like there are outliers out there. This top 0.1% engineer, this guy who admittedly was like the most productive engineer at Meta, right, like literally, quite objectively, he produced more code than any other singular engineer at that particular company. That is amazing. That puts him in like the top 0.0001% of all software engineers. So when he says, hey, I am doing the work of three to 10 engineers, I'm like well, dude, you are already doing the work of three to 10 engineers.
Speaker 1:And then when he says, hey, don't go to a boot camp, don't learn to code, I just see this and I hate this word. I see this as the ultimate form of gatekeeping. I'm like what purpose does it serve to tell people, hey, don't learn this skill that could change your life and go code. Also, like the idea that we just don't need coders is ridiculous. It's incredibly silly.
Speaker 1:In the San Francisco Bay Area, I'm seeing salaries rise to levels that I really haven't seen for a long time. I'm kind of shocked, actually, even though I know the overall trend is pretty stagnant or even down in some places. Let's be honest, having people write, react and get paid $300,000 a year at some companies, or a quarter million, which was a real thing. I mean, this was a real thing. At some big tech companies, they were paying like $300,000 for senior React engineers that really didn't know much beyond React. It was a very magical, stupid time, but we're past that time, right?
Speaker 1:So what do you do now? Here's a bit of a reality check for you, right? If you're new to the industry or maybe you're not coding yet and you're learning, here's what a lot of developers are feeling. We're feeling burnt out, we're feeling confused and we're feeling a little bit dumber. Honestly, right, we're using these tools like cursor. We're being told by management, leadership, people that generally don't have any technical expertise. We're being told use these tools, get faster, because that's what the company wants. That's what these tools have said. They've expressed this. We've seen the videos, we've seen the benchmarks, we've seen all the fancy marketing and we know, we know these tools are going to 10X your work and we're like using these tools and being like, eh, not really.
Speaker 1:And we're having this awkward conversation now in a lot of organizations about what's real, what's not, what can be done, what can't be done. There's a lot of distrust that I'm seeing from different organizations, including my own, between leadership and developers and being like, well, we've seen what we saw on YouTube or we've seen the Joe Rogan podcast or heard Mark Zuckerberg or whatever, and they're getting fed so much information and they're not using the same tools that we are. So they're thinking why can't you do it? And some of us are thinking the same thing. They're thinking maybe it's me right.
Speaker 1:So if you're new to coding, I want to tell you a few things that AI can and cannot do very well in the places where you need to get really really good as a developer, which you've always had to be good at. Let's be honest, these are not new skills at all. These are not new at all. Let me reiterate that these are things that you should be learning in any good program or course, school, whatever. Ai is really really good at writing quick boilerplate. Remember AI. These large language models have been trained on the entire body of the internet, which includes tons and tons of code. It is going to match patterns. It has seen a million websites. It's seen a billion Hello World projects.
Speaker 1:When you ask it to do that, it's like oh yeah, simple, it can explain syntax. If you're learning about functions, let's say you're in a program like dev 30, which we offer at Parsons, that teaches like basic JavaScript, which I actually think is even more relevant now than it was a few years ago. When it was created that same course, I was almost thinking about abandoning it, but now I'm like no, this course is actually more relevant than it's ever been, because you have to know how to write code and understand what it does at a fundamental level before you use large language models or cursor or other AI tools. But if you wanted to explain the syntax and you're like what does this function do? I don't get how this for loop works. I don't get what this variable declaration means. I don't get like just basic JavaScript stuff. I don't know how to set up my environment. Hey, I don't know how to get code that is in my, you know, vs code or cursor or whatever. I want to see that in the browser how do I do that?
Speaker 1:It's great at explaining these really simple problems that you're going to run into and, as a beginner, all your problems are simple and all your problems have been seen a billion times. The answers are out there. And now with a large language model like chat, gpt or cursor, the answers are out there. And now, with a large language model like ChatGPT or Cursor, you can find those answers more easily. It's great for those kinds of things. Here's what it's terrible for.
Speaker 1:Let me give you an example of a fairly complex problem we had to solve at work, and this will be what you'll likely experience when you get hired as a software developer. Right, somebody will come to you right with a problem and they're not going to say here's how to solve it with code. They'll say something like hey, dev team, we need a system that can look over a bunch of videos for a certain influencer, categorize them, and then we want to be able to search those videos by genre. And you're like, okay, well, how do I do that? It's like well, yeah, you tell me how to do that. Now you could certainly ask ChatGPT does it know what kind of systems you have? Does it know what kind of security you have? Does it know the kind of expertise on your team you have? No, it does not. So you can tell it these things and it can give you an answer, and it can even start making code for you.
Speaker 1:This code is often bloated, will not scale well, meaning it won't really work once you try to add more and more complexity to it, and it just generally won't know how to deploy it or get it to work in a way that you think it should actually work. It often gives you the most convoluted answers. The more complex the problem, the more convoluted and spaghetti-like the actual answer is. This is a problem that is just well known. Right. When it looks at like common problems, it gives you a common answer, and that's really really good. When it looks at a novel problem, it gives you something that it tries to match and it's like well, here's kind of what you want. Like you could, you could build this entire service on Kubernetes and Docker and use an AI tool in the.
Speaker 1:It does give you these really awful answers, so you'll need to think through these yourself. You say, well, okay, well, what's our time constraint? Okay, how many people on the team do we have. How could we do this? Do we need to scrape the service? Is there an API we can use? How can we download this smartly? How can we do concurrent downloads? Because maybe we're downloading a million. Maybe we don't need to worry about these things.
Speaker 1:This is where interesting things happen when you're solving the problem before writing the code. This is where most of the thinking happens. This is where most of the meetings happen. Most of your time spent as a developer is not writing code, which I think is the biggest myth out there for junior developers and people learning to code and also management. They're thinking don't you just code all day long? Not really right. You can. You certainly can be coding all day long. I was at a startup where I would code hours of the day. At another company, I was in meetings most of the day and I would code a few hours of the day. There was a study done. It said developers generally code, I think, like an hour and a half a day or less. I think it was actually much less than that.
Speaker 1:So you're doing a lot of thinking. You're thinking, okay, how can this work? You're maybe doing a quick script to see if it's going to work. You're saying can this method work? Okay, it does. Okay, well, how can it work for now, for 100,000 videos? How can we run this in the cloud so it doesn't have to depend on me pushing a button on my computer? Oh, you know, we're getting rate limited or black. Now these are more interesting issues we have to come up with Now. We want to download from another service. Can we do that? So now you're dealing with actual problems.
Speaker 1:So, when it comes to doing these multi-service systems, like systems that might interact with the database you have, it might interact with another downstream API. It's very, very rare that you'll work for any company, even a very small one, that has something that lives in isolation. A web app will typically have a backend. That backend will typically have a database. That database will have some sort of security layer. That security layer may interact with another API that logs all your errors to another service that will trigger an alert that will call you on your phone at midnight to tell you that the service is down. These things are not easily chat GPT-able, and the more complex the system is, the less useful your little AI tool is. And when it comes to error handling, writing maintainable code, think about the incentives of the AI, which is likely to produce as much code as possible.
Speaker 1:Sometimes I've asked for things and I'm like why would you write a 500-line file? And I will literally say, if you just did it like this, you'd have a much better answer. And they'll say, oh, you're right, you're right, I'll just rewrite it. And it will rewrite it or it'll take the code I suggest and it'll have a much more efficient, easier-to-follow answer. I have zero clue why these LLMs and I'm talking about cursor and I'm using Claude Code, I'm using Sonnet, I'm using a lot of things. I'm even having a bit of model fatigue, because one of the things I hate the most is when people say, oh, just use this other model. I'm like I don't want to just keep using other models. I want to use one thing that works good enough and I want it to be able to do what I tell it. So this is why you still need to learn in the right way, right?
Speaker 1:Ai is an amazing tool, it is a game-changing tool, bar none, right? But you need to learn how to write and read code. Think in terms of system data flows, how to debug across multiple layers. This is the kind of stuff we teach at Parsity how to build large projects, how to build something and take it from your head into code and deploy it. This will naturally make you come across a lot of the difficult errors that you're likely to see in a production or professional environment, and it'll give you like a little bit of that like glimpse into what it's like to be a software developer. Right, this isn't something you can just get from reading a book.
Speaker 1:Software development in general is a really messy occupation. It's not something that's super conceptual or like research oriented. There's certainly research aspects to it, but the majority of your day is like figuring stuff out, writing some janky code, proving that it works, and then finally thinking how can we write this code as a team, where each one of us takes small bits of a problem and then does our own little piece and then puts them together so that we can have something working. This is what makes software interesting. It also was one of the reasons why people talk so much about communication, and it's exactly why I think that more of us need to stand up and just say the truth, right? Let's be honest, if you're using AI tools and you're a senior software developer, maybe you're even a junior developer listening to this and you're seeing that it's not as helpful as you want it to be. That's okay. It's not bashing AI to say that it's not working the way we expect. It's just saying it's a tool.
Speaker 1:Everybody hates the tools they use, right? There's never been a tool that's been popular and just loved to death by everybody. Everybody hates the popular stuff. It's why people hate JavaScript. It's why people hate React. Ai is becoming yet another tool that we all have grown to love and then hate. So I hope that maybe gives you a little bit more calm, and also to know a little bit about what's going on in the world of software from a senior software developer's perspective that's working in the San Francisco Bay Area Really the heart of tech in the world. Honestly, I mean, maybe it's hyperbole, but whatever, this is the place where the big tech giants go to play.
Speaker 1:And lastly, if you don't want to miss out on the AI hype train because I get it you want to learn AI. You want to use AI, whatever that means, right, but you're thinking I want to learn Python or I better. You know, maybe I'll become a data scientist or machine learning engineer. No, no, no, no, no, no, no. If you actually want to understand how to use AI to build something, I've created this free little like course you could call it. I don't know what to call it. It's a video and it has some code samples that you can use to use HTML CSS, has some code samples that you can use to use HTML, css, javascript and a free API Gemini from Google that you can use to make like a really basic looking chat GPT-like little project. I think it's really cool. It's also super duper simple. If you're new to learning code, this is a wonderful way to just get started. If you want it, just click on the link in the show notes and it's all yours.
Speaker 1:Last favor I have from you is that I want to start doing more with this podcast. If you're listening right now, it means you're pretty engaged. You're one of the listeners that I really want to speak to. So if you have questions, there's a form in the show notes that I'd like you to fill out and just say, like, what you want to know. And if you want to leave your name, great, I'll shout you out, because I want to add one more episode per week, and this episode is going to be called office hours and every week I want to do office hours and I want to answer the burning questions you have, because I think I know what you want to hear, because I talked to a lot of people in your shoes that are either learning to code, or maybe you're in your first job, or maybe you're further along in your career. There's people that listen to this show in all different stages of their career and I hope to bring a little something for all of them.
Speaker 1:But I'd really love to hear what you want to ask. It can be anything like hey, I'm getting ready for an interview. You know how would I prepare for this? How would you prepare for this if you were me? Or hey, I'm thinking about going to this one to negotiate my salary. What are some ways you've seen to do that? I've seen a lot of stuff. I've been in tech for 11 years. I also have a buddy Zubin, who is a lawyer turned Google software engineer.
Speaker 1:So between the two of us, I'm sure we can give you some insight. I'll only tell you things that have worked for me or people that I personally know. I'm never just going to try to fake an answer or anything like that. So if you trust me and you want to hear it and you want to get shouted out on the show, just leave your question optionally, leave your name and you're going to hear from me. Anyway, grab that little mini AI HTML JavaScript course. I hope you really enjoy it and I will see you soon. Have a great week. That'll do it for today's episode of the Develop Yourself podcast. If you're serious about switching careers and becoming a software developer and building complex software and want to work directly with me and my team, go to parsityio, and if you want more information, feel free to schedule a chat by just clicking the link in the show notes. See you next week.
