Are developers happy yet? Unpacking the 2025 Developer Survey | Stack Overflow’s Erin Yepis

27:54Now you can take all of those skills that you gain from years of being a builder, plus all of that strategic capability and knowledge. And with AI, you're able to do so much more than we've been able to do in the past. And it almost makes the role exciting again in many ways because I have certainly felt this. I started my career doing a lot of sysadmin automations type development work. And the things that I'm able to do with AI back when I started my career, I mean, it's incredible how much faster I'm able to do things that would have taken me hours or even days back then. So pretty fascinating to hear, you know, experience.

28:34Like it definitely tracks with what I'm personally feeling and what I'm seeing from other people in the industry. But, you know, with all that said, so, you know, it seems like roughly one and a quarter or one out of every four developers are happy. About one out of every four developers are unhappy with their job. That's a pretty equal split. And in the middle, you have about half of developers that are still feeling complacent about their work. So I'm wondering what the data tells you specifically about these developers. What seems to be leading to still having complacency being the most common reaction that developers have for their job?

29:11We also asked in a survey similar to last year, but a little bit differently. We asked developers, respondents to rank what attributes of their job contribute most to being satisfied at work. And so for everyone, those were like at the top of the list is autonomy at work, followed by compensation, followed by solving real world problems. I would say for those that are complacent at work, those are still very important things to them, but they're probably just ranked slightly differently, depending again on role. So role, I think, is probably a huge driver of why an experience level of why you are or are not satisfied with your current job where you're working.

29:55So it's there's individual shifts for, you know, what at this point in your career is more important to you. conversation is important to everyone. But I think that as you get more experience at work or more experience in your specific role, the other two things sort of start like shifting with each other in terms of importance. That autonomy at work is something that we see shift a lot with different roles and different experience levels. With the less experienced developers, autonomy at work is not as important. And I think that's because they're trying to learn and learning comes with collaboration and having like some direction and having some mentorship.

30:38I definitely think for the complacency factor, it's just like having sort of like the importance that you assign to those top three factors is sort of shifting a little bit. And you have a little bit of more of one than the other in your current position. So I want to pivot just a little bit to the subject that we always end up talking about on every episode of Dev interrupted nowadays, and that's AI, specifically sentiment and usage around it. And I think the big takeaway that we had when we looked over this data was this sort of almost conflicting trend that we see a little bit of. So in the survey, it says 84 % of respondents are either using or planning to use AI, which we already mentioned that.

31:19I think it tracks with a lot of other surveys that we've seen out there, like the report just came out a few weeks ago and said, I think about 90 % of developers are using AI on a frequent basis. And that's up from, you know, in the Stack Overflow survey is up from 76%. So it's a noticeable jump. But at the same time, positive sentiment for AI has actually decreased by about 10 % from 70 % to 60%. So like what's going on here is, were we in the honeymoon phase last year, and that's wearing off or is there something else happening? It is very interesting. And it's been probably the most intriguing finding from the survey, I will say.

32:02I think what we also see, I know, is that, so as I had mentioned at the top, we changed this question a little bit to be more granular. Like if you use AI tools, are you using them daily, weekly, or monthly, or less frequently? And so we see there is a correlation between developers that mentions they use AI tools daily or weekly and their favoritism towards AI tools in general. So they have higher favorability scores than those that said they use it less frequently or are planning to. I think that speaks to, again, you know, as you have gone through the learning curve, you have found something that works for specific purposes, you've seen it succeed.

32:45And in the areas where you have problems, you know how to troubleshoot those problems. One of my other favorite questions from this year's survey, I don't think a lot of the respondents liked it because I had to read their responses, was the one about vibe coding. When you look at the responses for that question, there's a lot of responses that are like, I tried to use this and, you know, three out of four times it was wrong. So the hallucination factor. and I think that is so if you're not using these tools as frequently that the effect of you know I came to this restaurant once I'm giving it a one-star Yelp review it's I just it was too frustrating to even like go back to or try to find an alternative because of the bad experience I had a handful of times infrequently and you know I I'm here to like tell everyone about that But again, we're still in the honeymoon phase, I think, like you said.

33:42So there's a lot of maybe turmoil in trying to find the right tool for the right job and also finding the time to learn how to use these tools. We asked another question specifically in the last year. Did you learn to code for AI? and about two-thirds of respondents mentioned that they had learned to code for AI in the past year but only half of them did it at work and the other half did it for personal projects or at home so that also I think speaks to you're not getting enough time at work to even like learn set up a foundational basis for how would I use this when is an appropriate time use this What are the skills that I need to get better at using this for this specific use case?

34:30Also, besides that, we have like a cool chart that pivots that sentiment score against what the specific uses that you're using AI tools for. So while most developers are using it for search and code completion, you see that favorability score drop for some of the other use cases. The less used use cases, deployments is a big one. So I think that there are some complex things that we're just in the beginning stages of getting the right tools for that job where and then now we have to get those roles, the people in those roles used to and time to learn how to use those tools for that specific job.

35:13Yeah, I really love your restaurant analogy because I even feel like there's a big element of, oh, well, I went to that restaurant six months or a year ago when they first opened and the service stunk. So I never gave them another shot. And I feel like AI has gone through a very similar transformation or impact because, you know, a year ago, context windows were tiny and hallucinations were a big problem because of it. But today, context windows have gotten much larger. The reasoning has gotten better. and some things that you know didn't work at all a year ago now may actually work just straight out of the box and and we've also seen you know there's sort of like an inverted bell curve with ai adoption like when when a developer first adopts ai they're like wow this is new and exciting there's so many cool things i can do with it and they get to sort of that like intermediate level and they start to see all the problems all the failures that it has and they they like their perception will often drop and think wow this is terrible like why am i wasting my time with this.

36:08But then that curve comes back as they get better with it, as they start to use it, they find use cases that they can use it daily or weekly. Suddenly their eyes get opened and they're kind of back at the top of the perception curve where they're like, wow, this is incredible. Like my productivity is way up. I'm doing things that I never thought I would be able to do before. So there's, I imagine there's probably some sort of element of that too, where a lot of people are just, are still sort of stuck in that middle area where they've seen the potential, But when they've tried to implement it in reality, they're still struggling.

36:38And it definitely resonates, you know, relates to what I'm hearing from just people in my own personal network. Like I know a lot of really brilliant engineers who are doing AI constantly in their free time, like in their nights and weekends. Just like, again, like that getting excited about building again. And they're doing all these really amazing personal projects. But then they get to work and they're like, well, I can't really figure out how to apply like what I'm doing at home to my work environment. because we have so many additional constraints that just make it unfeasible. You know, one of my favorite stats from this whole survey is, you know, and I resonate with this.

37:11As somebody who uses AI daily, I still resonate with this stat. But 66 % of developers are frustrated with AI solutions that are almost right, but not quite. What other frustrations are you seeing from developers who are reporting lower levels of satisfaction? I think, again, going back to when the experience level speaks a lot to this, we see based on experience level, these developers are using very different tools. They have experience with different AI tools. And it definitely is related to the restrictions that you have at work. We have to use Microsoft Copilot at work. And I don't like that.

37:53We have to use, you know, insert blank here. So I think it has a little bit to do with that. Whereas like for the early career developers, or especially those that are learning, they're playing around with some of the sandbox areas, the free trials, the non-commercial use tools that, you know, especially like we have, there's free versions of almost all of the IDEs that maybe not so much anymore. But at least back earlier in 2025, so they were not beholden to maybe those strict larger organization enterprise-wide, like locking this down, you can't try this new one that came out or the one that everyone's buzzing about.

38:38But I think it also has to do with the fact that when you see in that chart that sort of pits the favorability scores against the what are you using these tools for, because there are some tools that I think are marketed as they can do it all. The fact is that they can do technically can do it all, but they do some things better and other things not as well. so they are having frustrations with uh okay we were sold or you know our company was sold this we're locked in and now my role i have to do deployments this is not helping me so i think it's those frustrations of like we need a different tool for this i need a different ai tool or i need a different ai agent i need a different integration something that is i think more tied to work definitely work and experience level.

39:33Yeah, it's really interesting because just think about my own usage. You know, I'm absolutely not beholden to a single tool. You know, I found that, you know, there are certain things that Claude is amazing at and there are certain things that it's terrible at. So I try to use it for the stuff that I love it for and avoid it for the other things. And, you know, the same is for chat GPT, the same is for basically every model that I've used. And then you have all the tools that are built upon those models as well that are all sort of built for different functions. So definitely checks out with my personal experience.

40:03And, you know, so building on this, the survey found that a lot of developers actively distrust the accuracy of AI tools more so than they trust them. How are you seeing this trust deficit play out? Like, does it differ between like professional developers and those who are still learning to code? Or is it, you know, sort of just based on what you were describing, like the professionals are sort of stuck with a certain tool set. And because of that, they have lower levels of trust. Well, so we do see that trust score improve slightly with younger developers, less experienced or younger in age, but not by much.

40:40I think more so it's not that it's like it shouldn't be seen as a negative effect of AI tools, rather that with more usage comes the knowledge that AI tools should not be fully trusted. That is part of the deal. That is built into the foundational knowledge when you start using them, especially at work. You know, your InfoSec team sets you up with a lot of these trainings every year about phishing and then don't click on these links and do not answer the phone and give someone your boss's email address and phone number. There's lots more information now than there was back when AI tools first started coming out.

41:26That is like, no, this is trust but verify. Trust but verify. Do not start with trust. There is this is the you are the important. The human is the important part of this equation. The AI tool is here to help you expand on what you are already doing, but you are trust begins and ends with the human. So rather than seeing it as like this is maybe the quality is changing or anything about that, it's more that developers are smarter. The whole world is smarter. We know we're not supposed to trust AI tools completely. Yeah, it's funny because there's so few tools in our workflow that you would use the word trust with.

42:09You know, like, do you trust your IDE? It's like, well, I trust it as far as I can build something with it. Well, and there's like specific tools that have the word trust in them. Yeah, yeah, yeah. I mean, if you're thinking about like a security tool, for example, trust is a huge aspect of it. Like, do you trust that it's actually catching the things that are important to your organization? So I want to get a little bit more into like workflows and developer tooling and specifically talking about how developers are using these tools. So one of the distinctions that the Stack Overflow survey made this year was between, and you already touched on this a little bit, between AI tools and AI agents.

42:43And I found that the majority of you of developers are still not using agents with 38 % saying they actually have no plans to at all at this point. So can you, first of all, define what the survey calls or how the survey defines an agent, because there's a lot of different definitions for that term. And then let's talk a little bit about why you think that adoption may be lagging for that. Yes. And I just pulled up the survey website because for all of these stats, we have the actual question that we used in the survey printed alongside it, so I can keep myself honest. But so for AI agents, we specifically said AI agents refer to autonomous software entities that can operate with minimal to no direct human intervention using artificial intelligence techniques.

43:34I know that sounds vague, but it's definitely less vague than how we define do you use AI tools, which is the only thing we were specific about there is like, are you using this for development at work? Not just like asking, try to be T over the weekend, where should we go get dinner? Because we're all doing that. And that would just say 100%. But so yeah, for AI agents, we have like a lower usage score there. I think that makes sense. The fact is, and I try to be, you know, specific and generic, because I think the, what an agent means, you know, it's marketed in one way. I think, Like, you know, when Salesforce came out with AgentForce and a lot of the marketing around that, it made it seem like agents were just something that you, like extensions that are part of enterprise SaaS programs.

44:25Yeah. But as time has moved on, like agents are something that is definitely a part of open source, huge part of open source software right now. And that's probably a bigger place where you see like the sort of the growth of like what an agent can do. What exactly does that mean for a software developer's job? Even though I try to have that sort of all encompassing definition, I do believe that that marketing tactic is a little sticky. When they see that question, they're like, no, I don't have this enterprise software solution that says agent in the title. I think that might be part of it. But the other part of it is, you know, especially for the professional developers, which, you know, a huge segment of the software respondents are of the survey respondents, apologies, are professional developers.

45:13a lot of them are experienced professional developers and at their work that, you know, again, with the locking down, like you can use these tools. This is what is okay to use at work. Agents, they're like, no, we have to verify all of this. There's too many of them. And that sounds like it will be more of a security risk. So I believe that maybe a little bit of the lag there is just speaking towards like the, the workplace getting up to speed with what they can and cannot use? What is an agent doing for them? Yeah. And I think a lot of the challenge with agents is that it's, I don't really know of a single agent out there that you can just buy off the shelf and start using like on day one.

45:55You know, a lot of this stuff takes a lot of effort to build. Even if you are buying a product, you still have to get your organization ready to use that agent effectively, whether it's cleaning up the data that you feed into it or having processes built around it. And I think that's, you know, a lot of the challenges of adopting an agent is actually organizational. It's not even a tooling problem. It is a technical problem because often giving an agent everything it needs to be effective does require technical work. But it really feels like a lot of the challenges are organizational. And, you know, speaking of that, one of the things that that caught my eye in this survey is how agents are being incorporated into development workflows.

46:38So, all right, you know, the report specifically mentions that there's a lot of personal productivity gains that agent users are seeing, but there doesn't really seem to be a whole lot of team collaboration improvement that's coming out of them yet. It seems almost like these AI agents are currently like a solo productivity tool in particular. Like, do you think that's going to change sometime soon? Or what do you think is contributing to that? Yeah, I definitely do think that's going to change. And I, you know, even in my own limited experience of starting to work with AI agents and see what they can do for my job, I, so a specific example is we have a tool at work that is approved and they allow you to connect agents or create your own, I was partnering with another researcher and we're like, yeah, let's build this agent.

47:31Let's go. Let's learn how to do it. And I go to configure it and it's, I can't share it with her. I can't share my configuration in the UI with her. So there's that. It's like literally built sometimes to not be as I think less so on the open source side, but for the, these enterprise ones that have been approved by work, it's like not yet there, Like the collaboration options are limited for some of them, maybe. But also the other part of it is I think that you have, that's a cultural thing. Collaboration, that has to come from your team, your leadership, that has to be built into the way that you do work.

48:11And they have to be pushing, you know, your managers or whoever team leads are pushing, should be pushing like these projects, like let's get collaboration. The goal is collaboration. And now let's try to implement an AI tool for this specific project. or we have to complete this project. Let's see if there's any creative ways that we can do parallel work with an AI tool that we have or start playing around with a new agent that we are trying to see if it may work if we want to buy the licensing for. So that I think more is, it has to come from the workplace, the collaboration part. But then also there are those technical capabilities that they have not built that in to some of these tools yet.

48:57I will say, you know, as far as collaboration goes, though, that they're, I've seen they've been coming out with a lot more, especially for stuff that is related to CICD and GitHub, for instance, and trying to like get agents plugged in there. That's built for collaboration. So whenever there's a way to sort of integrate it into tools that you know are collaboration positive, there is an interesting tool that I've been playing around with. Still trying to get it set up like regularly is a spec story. So the whole point of that is that in your PRs, you're sharing what it is that you and the agent were doing.

49:40Anyone can read that. yeah and i mean just thinking back to like where i was a year ago just generally speaking about ai like not even agents but a year ago i feel like the technology was changing so rapidly the tools we were using was still relatively immature it was really challenging for me to share what i was building for myself with other people on the team because by the time i share it like some of what i share becomes irrelevant or it might be that i i built this thing specifically for myself and there's not a whole lot to like take and adapt to other parts of the organization. But sometime in like the last three to six months, I've started to feel that shifting where now the tooling is catching up to our team needs and we're able to more openly share things that we build, you know, AI services that we build with each other, give each other tips that actually have a lasting value rather than just being an immediate value to yourself.

50:31So yeah, it is a hopeful trend, I think, for agents as a whole. I'm not going to let you leave without talking about one of my favorite subjects right now, and that is vibe coding. And I know this always generates a bit of controversy when you bring this up. But in the survey, most respondents, so about 72%, are not vibe coding today, at least according to their own admission. And an additional 5 % are emphatic about it not being a part of their development workflows. And this kind of flies in the face of what you'll see from all the people out on social media, either saying vibe coding is amazing and it's the future or it's the worst thing ever and no one should ever do it.

51:09What's going on here? Like, you know, why aren't people vibe coding yet? Well, you know, and I, again, the wording of the question is definitely in the SurveyMicro site. I try to be very specific. I linked to the Wikipedia page for it. It's a, you know, it's a term that came up in the past year or so. And I only refer to it because, well, one, And we definitely had leadership at Stack Overflow. They're like, you should be asking about vibe coding. We want to know. But it has, I think it's a loaded term. And maybe there's some, I feel like there it's the loaded nature of it is because it infers maybe that you are not capable of doing your work.

51:56You need the help. And I think a lot of people that have worked with these tools more know that that's not the case, but they still are not like, I'm not going to call them fine coding. So that is why I added that emphatic part in there, whatever. It was an open response question. So I used an NLP algorithm to sort of group together a lot of the responses. And there were very colorful responses. And, you know, in general, we had a question, you know, I added a give us feedback on the survey question this year. And we got a lot of feedback in the survey in general, like, I don't like that you asked so many AI questions this year.

52:38So, you know, the Stack Overflow community is more aligned with the human part of the process for developing, getting answers, thinking through problems. I think that the reason for the response for Vibe Coding and the survey this year was what it was is because they are doing all of those things. They are thinking through, using their own skills. they are using ai tools they are starting to use ai agents but they would never call what they're doing vibe coding because they are still there's definitely the human in the loop like emphatically that yes emphatically no they're not just like walking away from the computer and refilling their coffee cup and coming back to a finished product yeah there's definitely a negative connotation with the term that i feel like it doesn't deserve at this point i wonder if maybe maybe if you next year just as a recommendation, I guess.

53:37If you called it agentic coding, maybe there would be a lot more positive or maybe even do both. Like, do you vibe code? Do you agentic code and sort of see the difference in results? Well, and also I think the loaded term, it's, it kind of is giving Gen Z vibes. And so we have, you know, like I said, we have older skewed respondent base and they're like no i'm not gen z i'm not doing skibbity whatever oh yeah yeah that's yeah i'm with you on that so all right so i'm going to close out with some some high level takeaways from this so you know one of the things we've been talking a lot about professional developers organizations that are dealing with you know poor job satisfaction complacency at work so for our audience of engineering leaders who are listening right now you know they're probably thinking about things like retention, team health.

54:30Like, what are the biggest takeaways from the survey that you have on how to re-engage those developers who might be feeling complacent or just down about their work? Yeah. So I would refer back to those top facets that respondents ranked, like the top three for what contributes to their job satisfaction. While maybe many leaders, they definitely have a say in compensation. they might be a little hamstrung on that but the autonomy at work and solving real world problems could definitely be something that becomes more of a focus and i think that you know something that's come up in uh i've heard from our leadership and i've heard from this podcast is the the idea that you know instead of it being top-down mandated here's a tool use it or consequences.

55:24It's trying to encourage like the, what do you think? Getting input from the development teams about these AI tools or if we should implement this rather than just having the finance team, you know, set up a workflow that is like, all right, get used to it. This is what we're doing. It's that part can incur one. It will bring into the collaboration factor up That we're having open conversations, trying things out together, trying to find solutions that work well for everyone that's going to be involved with the AI tooling that is going to impact their work. but also giving them that space to learn that they, you know, we all need to learn, uh, in order to use some of these tools.

56:10Some of them are easier than others and that they have time at work to like really dig in and, and give it, you know, the time that it needs so that it can become part of their workflow. It can be, and it's not less of that frustrating factor of like this failed a couple of times and I don't have time for this. I don't have time to learn how to fix it. Yeah. Yeah. All right. I got one more question before I let you go. So as you're looking forward to 2026, you know, what are the trends that you're watching, you know, and specifically like what would you want to evaluate when this survey comes out next year?

56:43I think the biggest one is I'm always constantly interested in how developers are learning. I know that they're there of every age range of every experience level. They are learning every year. They're learning for different reasons. They're using different ways to learn. So that's top of mind for me. You know, one of the new questions we asked this year was about which community platforms have you been using in the past year? And, you know, Stock Overflow is one of them. But it's just so interesting to see how YouTube continues to be a huge, it's a huge community platform for developers. And, you know, podcasts definitely, I think it plays into that too.

57:26getting to see like those overviews, the video format. So what I want to continue to explore, like what new things are coming out that are helping developers or anyone in the tech space learn and how is that changing? And, you know, how is digging more into this idea of like, how has work been affected by new tools rather than just like asking the trust question and the favorability question, but that could easily be, we dig into that a lot more. Like what, what are you doing with this? Uh, I feel like we already asked a bunch of questions, so I'll remove some of them. I'll replace them. So there's not just like a whole entire AI survey, but there's so many things to ask about.

58:10And you know what, going back to, uh, I just started looking at the JetBrains developer survey for this year. They mentioned they asked 500 questions. Wow. So yeah, I only asked like 70. So, you know, it could be worse. And with AI, it's more easy than ever to generate survey questions. Yeah, and you know what? You know, anyone can use AI to answer any of these open response questions. I read them. I'm like, there's no way AI, especially the Vibe coding way, AI did not respond to that question. All right. Well, Aaron, this has been really fantastic. Before we wrap, we're going to listen to all your work and learn about what's going on over at Stack Overflow.

58:53Yes, I definitely encourage anyone to go check out the Stack Overflow blog. I usually post a survey analysis there. We're going to have a new survey analysis coming out soon about what developers are frustrated by and spending time on at work. And check out Stack Overflow. We have tons of new stuff coming up soon that is going to be really exciting. Stackoverflow.com. Fantastic. We'll be sure to link to everything in our show notes. And that's all for this week. But remember, you're only getting half the story if you're only listening to this podcast. Go check out our Substack, devinterrupted.substack.com to find all of our in-depth news articles and stories.

59:33And Erin, we'd love to have you back sometime either on our Substack or on the podcast. We always enjoy having a conversation with you. And thanks again. It's been a ton of fun having you here. Thank you so much. I really enjoyed it.

59:56Outro Music