RAISE Summit 2025 | Prashanth Chandrasekar, Stack Overflow

Clips
More from RAISE Summit 2025

Prashanth Chandrasekar

CEO

Stack Overflow

play_circle_outline Stack Overflow remains a vital community for builders, especially developers seeking coding assistance.

play_circle_outline Emphasis on community engagement and human-curated knowledge amidst AI advancements.

play_circle_outline Leveraging Stack Overflow's Expertise to Enhance Enterprise Productivity with Customized Knowledge Intelligence Solutions

Info
Transcript

Prashanth Chandrasekar, Stack Overflow

Prashanth Chandrasekar

CEO Stack Overflow

In this interview from the RAISE Summit in Paris, Stack Overflow CEO, Prashanth Chandrasekar, joins theCUBE’s John Furrier to break down how the developer community’s “knowledge graph” is powering the next wave of AI. Chandrasekar shares fresh survey data that shows 80 percent of Stack Overflow users now tap AI tools while trust in automated outputs has sunk to 29 percent. He explains why high-quality Q&A pairs lift LLM accuracy by up to 40 percent, how Stack Overflow’s private SaaS platform becomes an internal “knowledge intelligence layer” and why data lice... Read more

explore Keep Exploring

What is the role of Stack Overflow in relation to coding assistance and the impact of AI on the startup ecosystem? add

What is the direction being discussed regarding the use of AI and the role of human expertise in that context? add

What role does data quality play in the effectiveness of AI assistants and agents within enterprise applications? add

bolt Powered by CUBE AI

Prashanth Chandrasekar, Stack Overflow

search

>> Welcome back, everyone. Day two of theCube live at the RAISE Summit here in Paris, France. I'm John Furrier, host of theCube. It's been a great two days so far. We're kicking off the day, got a full lineup of interviews with all the top leaders coming into our podcast studio here on the show floor. Really is a show about AI leaders, existing leaders of existing companies transitioning to AI and catching that wave, as well as, and a nice surprise, a lot of developers, smaller teams, some getting big funding, some not a lot of funding at all, but at the end of the day, we are in a builder culture. Okay, we've got Prashanth , he's the CEO of Stack Overflow, friend of theCube. Walking by, wanted to grab him in. Great to see you.

Prashanth Chandrasekar

>> Great to see you, John. - We're on the boat cruise with

>> the VIP passes.

Prashanth Chandrasekar

>> Yes. - Thanks for coming in.

Prashanth Chandrasekar

>> That was awesome. Yeah. - So Stack Overflow, well-

>> known site and community around helping people build,

>> but AI with coding, coding assistance, the cost to do a startup is lower than ever, time to value is super fast, obviously an impact. You guys, one, are a site where builders go.

Prashanth Chandrasekar

>> Yeah. - AI is now providing those answers,

>> they're providing a lot of content, they're crawling sites. We saw Cloudflare put up a little metering, kind of a toll booth, if you will. Very innovative, I think, but I don't know how I feel yet. I'm not negative against it, but also, what does it mean? But I think it's been a free ride for a while, a lot of these communities are investing in content. It's a double-edged sword, on one hand, it's great. On the other hand, it's not. Give us a take on your view of what's going on at Stack Overflow. Again, well-known in the industry. So what's new with the AI wave for you?

Prashanth Chandrasekar

>> Yeah, no, thank you again, John. It's always a pleasure. Third time participant on theCUBE, so always a pleasure to talk to you. So it's been a fascinating time for the past couple of years for us. We've been around since 2008, been the preeminent place where technologists go to get information about any programming, obviously topic. And we have this amazing corpus of 50 million questions and answers, which has been now reused over 50 billion times. And experientially now with the AI, advent of AI. That's the B2C side, and then obviously, we have a very large enterprise business, which is a private version of Stack Overflow that's used in the world's largest companies as their private internal knowledge base, and more now as their knowledge intelligence layer, which I'll explain. But what we noticed with the AI world is that... Probably the best way for me to explain it is through a data point from our developer survey that we conduct every year. So this year, we noticed that the adoption of AI in the community has gone up to now 80% of our users, but the trust level has dropped from what used to be actually a pretty low number, 40%, down even a further low number, which is growing 29%.

>> Trust on the AI? - Trust on AI outputs.

Prashanth Chandrasekar

>> Yeah. Yeah. - And so in a world which is going

Prashanth Chandrasekar

>> to be clearly the direction is, "Let's go automate using AI.

>> " And that's the wave we're all catching.

>> There is a need for human loop, human- curated experts who provide trustworthy validation and answers to whatever the AI is generating. And so we believe we can play, really, a key role there by being the world's most vital source for technologists. And the way in which we want to do that is to really cultivate community, which I think is going to be even more important in the age of AI. And through powering learning, because you think about the-

>> Yeah. Of

>> course. - ...

Prashanth Chandrasekar

>> disruption to jobs for junior developers especially. And then finally, through unlocking growth, both in their careers, as well as inside companies through productivity tools like our internal product.

>> You and I both have had conversations off camera around AI's impact to our businesses. We have a lot of similarities, not a lot of overlap, but I think similarities in the sense that we have invested in high quality content, developer content and media content, where there's a mixture of experts and resident in our data. And then now AI makes that easier, so we're seeing... And I was just on stage making a comment about this here at the event, AI's going to create a wave of tsunami of vanilla content. On our side, we're seeing already it's easy to write a blog post, it's easy to write a marketing copy, it's easy to write bad code than ever before, hence the trust numbers makes a lot of sense. So we're predicting, and I'd love to get your reaction to this, if you see the same thing, a flight to quality. And so original, authentic, or in your case, quality code matters. And so I think we're going to see a transition of, "Wow, it's so easy, but it sucks. " Kind of vibe. So vibe coding is a trend. But getting into production operationally on the consumer side, so one, do you see that? And then two, the enterprise is going to have more programming skills, they're going to need to have higher quality, because now the brands are on the line.

Prashanth Chandrasekar

>> Yes. 100%. Everything you said is spot on, because we noticed is both on the public side as well as in the enterprise business. So on the public side, what we have noticed is that our knowledge is so valuable for LLM pre-training and things like RAG and indexing use cases, et cetera. That's why we work with every single AI lab as well as cloud provider. So OpenAI, and Google, and everybody else that you can imagine in the space now officially license our data, because the data produces something like a 30 to 40% improvement in the accuracy of their answers. So it reduces hallucinations dramatically. If you look at the human eval score, it goes from approximately 5 to 9.

>> . - Any index, we have even have our own index

Prashanth Chandrasekar

>> called ProLLM, you can go to prollm. ai, which we did with our process partners, and they rank every single foundational model in the context of coding related tasks with our data being used, not used, and it shows the rankings relative to the accuracy of coding outputs of those LLMs.

>> Sure. - So we have proven

Prashanth Chandrasekar

>> that the high quality data on our site is extremely high quality data, because the standards have been so high for so long on what qualifies as a canonical answer to a question. While it's somewhat difficult to actually engage on the site sometimes, because if you're a junior developer, it's somewhat a little bit harsh, but the reality is that the-

>> Yeah, harsh in a good way.

Prashanth Chandrasekar

>> Harsh in a good way, it's like the school of hard knocks, so to speak. On the enterprise side, what we also notice is that data quality is the long pole in the tent, because people are realizing, "Okay, if you want to use AI assistants or agents, then you want to make sure it's working off of accurate information. " So oddly, and kind of surprisingly, our Stack Overflow internal product that's used, again, by the world's biggest financial services institutions, tech companies and so on, the APIs of that product were red- hot over the past year. And when we dug into it, they were all being used in AI search use cases, AI assistance, et cetera. Like Uber as an example, built something called Uber Genie, they're a Stack Overflow internal customer with thousands of questions and answers. They tap into that, and then answer a bunch of Slack questions in their Slack channel, and so they drive tremendous amounts of productivity using that. So quality absolutely matters.

>> So it really is a- - It's garbage in,

Prashanth Chandrasekar

>> garbage out otherwise.

>> It really is a double-edged sword,

>> because one hand, you have done the work, I won't say labeling, but basically labeled, so great training data for the LLMs, great check there. So I guess is the strategy you have to embrace, you're leaning into the LLMs?

Prashanth Chandrasekar

>> Yeah. - So that makes sense, because they win.

>> You have more lift on the developer side. And then are you guys on the enterprise side strategy-wise creating custom-approved sets of data? So what's the strategy? Because you don't want to be eating alive, so to speak, because... And it's not their intention, their intention on the consumer side is provide great answers. You already have them?

Prashanth Chandrasekar

>> Yes. - Okay. So now what's the

>> strategy for you? What's the plan?

Prashanth Chandrasekar

>> Yeah. So on the public platform, it's to create all the additional dimensions where we can be useful to our community members, like I mentioned with the mission and the vision that I explained around community, and learning, and unlocking growth, and that will do that beyond Q&A. So we have unlocked, for example, human-to-human chat, where you can actually go and engage with the expert on Python on the site, or an AI topic. We've enabled even AI search and discovery through, not only everything you get on the public AI ecosystem, but also through all the corpus of information. And if you don't get your answers to your question there, ask you a question to the community either into the chat room or the Q&A place, et cetera. Even enabling things like code challenges, so junior developers can learn about the new stuff, so they can get certified as an expert on certain topics. Anyway, that's all on the public side. And then on the enterprise, it's about creating this real knowledge intelligence layer, so that we can ingest information from other parts of the company. And knowledge is always out of inside companies, so that's what this product was solving, but now it's on steroids a little bit by bring it all in, adding things like an MCP server on top, and then plugging it into your AI agents.

>> Again, back to what I love about your company is that there's so many similarities with theCube and SiliconANGLE, words like mixture of experts is a buzzword that's used in the reasoning side, domain expertise, these are buzzwords in AI-

Prashanth Chandrasekar

>> Right. - ...

>> they're built in. And even better now reinforced learning with human feedback. So I think that it's a great example of not being afraid of AI-

Prashanth Chandrasekar

>> Right. - ...

>> you lean into it. Because actually, you're the flywheel for human feedback.

>> Correct, correct. - That's a superpower.

Prashanth Chandrasekar

>> 100%. - So that's a value that's going to open up

>> that wasn't really there before.

>> So one door closes potentially, which actually,

Prashanth Chandrasekar

>> there's not a door close, it's more lift actually, and you swing back to value.

Prashanth Chandrasekar

>> That's correct. - Am I getting that right?

Prashanth Chandrasekar

>> 100%.

>> All content sites have to rethink their business strategy and what actually makes them special in this new world. For us, we've leaned into the fact that our trusted experts, and ability to really have fantastic human curated knowledge, while leveraging the power of AI in service of the human or the technology-

>> So you're building a trust network?

Prashanth Chandrasekar

>> Exactly right. It's a trust layer. Well said.

>> So that's exactly what we're doing.

>> Got it. Okay. So I have to ask you, on the developer side, what are some stats you can share? I mean, obviously some are saying that the junior developers are going to be the senior developers, because they get the superpower of the augmentation of expertise. So how are you guys looking at that sharing? I see chat being a first phase, what's beyond chat? Agents? More coding? Autonomous coding?

Prashanth Chandrasekar

>> Yes. Yeah, so the experience that we are enabling on the public platform, using the human-to-human expertise, but also incorporating AI. Then you have AI and humans engaging directly on the site in a way that's really productive for the junior developer and the expert on the site. And then to be able to ultimately bring them along in their journey, I think is really, really important. But overall, I think for us, we want to make sure that we leverage, again, the power of AI in service of the human.

>> Got it. - And that's really what we're focused on.

>> I'm just curious, because we were...

Prashanth Chandrasekar

>> Just something we were talking about on the boat,

>> licensing content, is that a new revenue stream for you?

Prashanth Chandrasekar

>> Yep. - Is it something that's powerful?

>> Can you share your thoughts on licensing? Because we saw the Cloudflare... We're living in an open web, but yet now we have starting to have metering, but there's never a free lunch. There's now new ways of experiment that was once taboo.

Prashanth Chandrasekar

>> Right. - I mean, it was once taboo to

>> say, "Whoa, whoa, whoa. " But now with all that sucking out and distilling, it's a feature not a bug, is it a bug or a... This is the conversations we're having in the tech world. What's a feature? What's a bug? What's your vision on the whole licensing? And then that notion of what you want to enable, the risk- reward ratio, if you will?

Prashanth Chandrasekar

>> Yeah. So we are part of the Cloudflare announcement. So yes, to answer your earlier question, data licensing is one of our three lines of business that we have. So we have the public platform, which is monetized using data licensing, as well as advertising. But advertising is now only about 25% of our company's revenues. So we're not like another content site that may be heavily dependent on advertising.

>> Yeah. You don't want to get too addicted to that, because that's a downward trend.

Prashanth Chandrasekar

>> That's correct. Just generally web traffic I think is a very different ball game these days. So I think anybody dependent on web traffic solely to monetize, I think is a very difficult place to be. And then the enterprise business is our SaaS and API usage, as I talked about. So data licensing ads in our enterprise business, those are the three lines of business. Data licensing, we, as I said, struck recurring revenue licensing agreements with all the big AI players. We're also seeing a lot of demand from those companies, as well as other early-stage startups that are also here to say, "Can they provide that reinforcement learning? " So for example, they're willing to even ask... Pay us for feedback on the outputs of their AI, much like some other companies, their whole business is based on that, because again, we have an expert network, if you will. And then the growth of that will be in a whole range of use cases, whether that is AI pre-training, whether that is RAG and indexing with that search use cases, and even custom slices and dices of data, which people could find. Because again, we have a really, really deep dataset that goes back 16 years, and the amount of metadata that we have is just tremendous. And it's useful, by the way, for not only coding- related tasks in AI, but non-coding-related tasks.

>> Yeah, I've always loved Stack Overflow, it's been one of the jewels of the internet, and it's almost like the world spins in your direction, and you're the treasure trove of data, and it's phenomenal. And I think too, the enterprise business, I think will go more micro-modeling, where they'll start looking for single-tenant custom solutions, really around work stream. You look at what Augment's doing in some of these great, more production-oriented where they're doing automated pull requests. I think that's going to be a direction, I think it's going to be compelling for the enterprise once they get their crown jewels locked down with the supercomputing capability, they're going to probably want a need for specialized models where... Is that on your radar?

Prashanth Chandrasekar

>> Yes, it is. More and more as we think about what you earlier mentioned, when people are now pushed... This is that third attempt probably trying out an AI experiment after three years, and they realize all the long polls in the tent. What does it actually take to actually roll, let's say into a bank, or push something into production in a bank? "Hey, you need this and this and this. " One of them is definitely high quality data. So in that context, "Okay, are you attributing correctly? Is it fresh and is it relevant? " And in that context, creating that knowledge intelligence layer that I mentioned, while bringing in data sources, like our own and others into that, I think is going to be a real, real unlock. And that's just kind of where people are headed.

>> Prashanth, it's great to have you on. Thanks for swinging by theCUBE.

>> Thanks. - Great to see you again.

>> Wonderful to see you, John.

Prashanth Chandrasekar

>> Love your business. Stack Overflow, again,

>> the crown jewels on the work that's been done.

Prashanth Chandrasekar

>> All the answers are in there, and more are coming. Of course, the LLMs are integrating in, it opens up opportunities. I think the lesson here is don't be afraid of the AI engines out there, they could help you, and I think that's a great example. Of course, more coverage from theCUBE here at the RAISE Summit in Paris. I'm John Furrier, your host of theCUBE. Thanks for watching.