Stack Overflow users don’t trust AI. They’re using it anyway

Right now, I’m speaking with Prashanth Chandrasekar, who’s the CEO of Stack Overflow. I final had Prashanth on the present in 2022, one month earlier than ChatGPT launched. Whereas the generative AI growth had tons of impression on all types of firms, it instantly upended every thing about Stack Overflow in an existential means.

Stack Overflow, when you’re not acquainted with it, is the query and reply discussion board for builders writing code. Earlier than the AI explosion, it was a thriving neighborhood the place builders requested for and acquired assist with difficult programming issues. But when there’s one factor AI is sweet at, it’s serving to builders write code — and never simply write code, however develop complete working apps. On high of that, Stack Overflow’s boards themselves grew to become flooded with AI-generated solutions, bringing down the standard of the neighborhood as a complete.

You’ll hear Prashanth clarify that it was clear kind of from the leap how huge a deal ChatGPT was going to be, and his response was pure Decoder bait. He referred to as an organization emergency, reallocated about 10 % of the employees to determine options to the ChatGPT downside, and made some fairly large choices about construction and group to navigate that change.

Three years later, Prashanth says Stack Overflow is now very comfy primarily as an enterprise SaaS enterprise, which offers AI-based options which can be tailor-made to totally different firms’ inside techniques. Stack Overflow additionally operates an enormous information licensing enterprise, promoting information from its neighborhood again to all these AI firms, giant and small.

That’s a reasonably large pivot from being seen because the place the place everybody can go to get assist with their code. So I needed to ask him: does Stack Overflow even entice new customers anymore, in 2025, when ChatGPT can do all of it for you? Prashanth stated sure, in fact. You’ll hear him clarify that whereas AI can deal with easy issues, for thorny, complicated ones, you actually wish to discuss to an actual individual. That’s the place Stack Overflow nonetheless brings individuals collectively.

You’ll hear us come again to a single stat particularly: Prashanth says greater than 80 % of Stack Overflow customers wish to use AI or are already utilizing AI for code-related matters, however solely 29 % of that inhabitants really trusts AI to do helpful work.

That’s an enormous break up, and it’s one I see throughout in AI proper now. AI is in every single place, in every thing, and but large numbers of individuals say they hate it. We see it, within the Decoder inbox, within the feedback on The Verge, and on our movies on YouTube. Everybody says they hate it — and but numbers don’t lie about what number of hundreds of thousands of individuals are utilizing it and apparently deriving some profit.

It’s an enormous contradiction and exhausting to unpack. However Prashanth was fairly keen to get into it with me. I believe you’ll discover his solutions and his perception very attention-grabbing.

Okay: Prashanth Chandrasekar, CEO of Stack Overflow. Right here we go.

This interview has been evenly edited for size and readability.

Prashanth Chandrasekar, you’re the CEO of Stack Overflow. Welcome to Decoder.

Great to see you once more. It’s been a sizzling minute. It has been three years because the final time we spoke, so it’s nice to see you once more.

I ought to have stated welcome again to Decoder. You have been final on the present in October 2022. One month later, ChatGPT launched.

[Laughs] That was an apparently timed interview, proper earlier than the world modified.

Proper earlier than the world modified. Software program improvement is actually the factor that has modified probably the most since AI fashions have hit. There are numerous new merchandise in your universe to speak about, and there’s what Stack Overflow itself is doing on this planet of AI. So I wish to discuss all of that.

However first, take me again to that second. We had spent a whole dialog in 2022 speaking about neighborhood and moderation, the way you have been going to construct a funnel of individuals studying to code, and studying to make use of Stack Overflow. That was an enormous a part of our dialog. The engineer pipeline, each them studying to jot down software program and being part of the software program improvement neighborhood, was very a lot in your thoughts. After which, all of software program improvement modified due to the AI device. Describe that second for me as a result of I believe it contextualizes every thing that occurred afterwards.

It was undoubtedly a really, very stunning second. It wasn’t an sudden second in some ways as a result of right here got here this know-how that clearly some individuals knew about however not in a means that captured all people’s creativeness utilizing this lovely interface. We have been in the course of wrapping up our calendar yr, and at that time, we have been serious about our priorities for the subsequent yr.

It grew to become very clear what we wanted to give attention to as a result of this was going to be this very, very large change to how individuals eat know-how. Welcome to know-how. It’s continuously altering, and I believe this wave particularly is totally unprecedented. I don’t assume there’s any kind of analogy or prior wave that I may look to, together with the cloud and possibly the web. I don’t assume we’re nonetheless totally consuming what that is in the intervening time.

So, we went into what’s the equal of a code pink state of affairs inside the corporate. It was an existential second, particularly for our public platform as a result of the first job, if you’ll, is all about ensuring individuals get solutions to their questions. Now, you’ve gotten this actually, actually slick pure language interface that permits you to do this at a second’s discover. We needed to kind of arrange our ideas, and what I ended up doing was carving out 10 % of the corporate’s assets to particularly give attention to a response to this.

We set a selected date to reply in a significant trend, so we stated the summer season of 2023. I used to be going to go communicate on the WeAreDevelopers Convention in Berlin, and I successfully advised the corporate, “We’ve received six months to go and produce our response.” At the very least, it might be our preliminary response as a result of that is going to maintain iterating.

That’s how we mobilized the corporate. We acknowledged it was a code pink second, we carved out a group of 10 %, so about 40 individuals or so since we have been a considerably medium-sized firm. Then, we started working. That was the second.

Take me inside that room. Only a few individuals ever get to ship the code pink memo, proper? This isn’t a factor most individuals ever get to do. Perhaps you consider doing it, however nobody’s going to learn your memo. Everybody has to learn your memo. You’re the CEO.

Take me inside that room the place you stated, “Okay, I’ve recognized an existential menace to our firm. Folks have come to us for solutions to software program improvement questions.” Once more, the final time you have been on the present, you have been speaking concerning the thought that there have been objectively proper solutions to software program improvement questions and that the neighborhood may present them and vote on them. Effectively, now you’ve received a robotic that may do it and might do it as a lot as you need, so long as you need. There are instruments like Cursor AI and Cloud Code that may run off and do it for you.

So that you’ve received all that, and also you say, “I have to take 10 % of the corporate.” I’m curious how huge the corporate is. I do know there’s been some adjustments, however 10 % of the corporate is 40, 50 individuals. How did you establish and say, “That is the second I want to tug these individuals within the room. I’m making this determination, and the proper reply is 40, 50 individuals are going to put aside their time to ship me a plan by this time subsequent yr?”.

The intuition has come from a pair totally different experiences. My expertise proper earlier than this was at Rackspace within the cloud companies house. The enterprise I used to be operating at Rackspace was really round responding to Amazon Internet Providers as a cloud know-how menace. I used to be on the group that in the end constructed that enterprise from the bottom up, and it was successfully 10 % of Rackspace’s inhabitants that went and created that. So, I had some apply seeing and responding to a disruptive menace that you simply encounter. It was my flip now to place that into movement at Stack by appointing anyone like myself after I was at Rackspace to do precisely the identical factor.

The opposite information level goes all the way in which again a pair a long time or extra after I was in enterprise faculty. My professor was Clayton Christensen, and he wrote the guide The Innovator’s Dilemma. I’ve all the time thought of that within the context of know-how. In know-how, it’s a very constant theme that now and again, you should have disruptive threats, and there’s a really particular means through which it’s essential reply to that. It’s very a lot about how historical past suggests it is best to carve out an autonomous group that has very totally different incentives and might pursue issues in a really totally different means relative to the remainder of your corporation.

And keep in mind, Stack Overflow is absolutely two components. Now we have our public platform, which is that this huge net disruptor and which we must always discuss extra broadly regarding the web). The opposite aspect is the enterprise enterprise, the place we’re serving giant firms with a personal model of Stack Overflow. Fortunately, individuals proceed to see worth in having a data base that’s very correct. More and more over the previous few years, it’s really even grow to be extra useful since you want actually nice context for AI brokers and assistants to work. I’ve received loads of examples, so we will discuss that.

So, that’s the place that response got here from. I had been via it in a pair totally different dimensions previous to that. And simply by way of how I communicated to the group, the memo was really like a sequence of memos. Each Friday, I ship an organization e-mail. I simply despatched one proper earlier than I received on right here. I’m fairly clear in these: “Right here’s what’s on my thoughts, right here’s what we needs to be doing, listed here are some nice issues that occurred, listed here are some individuals who demonstrated core values.”

I’ve executed that religiously for… I’ve been on the firm for six years, and I do that each Friday. The group principally is aware of what’s on my thoughts, and so it wasn’t one huge memo to activate it. It was a sequence of emails main as much as this second saying, “right here’s what we’re going to be, we’ve received to answer this, right here’s what we’re serious about now,” and so forth and so forth. This went on till I may put the flag ballot down and say, “Hey, by the WeAreDevelopers Convention, now we have to supply a significant response on the general public platform in addition to on the enterprise entrance as a result of it’s an awesome alternative to combine AI into our SaaS software, which is a unique vector.” Hopefully that helps.

Did you really kind the phrases “code pink?”

I believe I undoubtedly used “disruptive.” I used “existential second.” I used all these issues, however I don’t know if I used the precise phrases, “code pink.”

[Laughs] I simply take into consideration that second the place you’re like, “All proper, I’m going to hit the C and the O… I’m saying these phrases, it’s taking place.”

It was very clear. Now we have a really particular communication cadence with the corporate, like many others. The tone and seriousness of what we have been engaged on was very apparent to individuals, particularly once you carve out assets and take individuals away from sure groups. Individuals are going to ask, “Wow, what about my employees?” Right here you go, there’s the explanation. So, it turns into very apparent.

How did you make these choices to tug individuals away? How did you determine which individuals, how did you determine which groups? These are all trade-offs, proper?

Yeah, little doubt. I believe this can be a exhausting downside to unravel. You actually need very gifted individuals, however I believe you need the kinds of people who find themselves keen to interrupt glass or go in opposition to the grain and never be encumbered by historic norms. I very particularly picked a mixture of individuals. The people who find themselves main it have been newer and got here from the skin of the corporate as a result of keep in mind, we have been going via a metamorphosis.

I joined an organization that was engineering-led in 2019 and all about this public platform, and we have been reworking into this product-led group. We particularly appointed a more moderen one that had come from the skin, was considering constructing extremely progressive, quick iterating merchandise, and had the DNA and the drive to do it.

I additionally personally stayed a lot nearer to it. I, in truth, ran product for an interim time period with that individual reporting on to me. That was one other strategy to keep very, very near what was taking place on the bottom till the precise launch. The remainder of the group was a mixture of very gifted engineers, designers, and individuals who had context of how the positioning labored previously and who may present us with all of the unlocks that we wanted.

I take into consideration Stack Overflow in what are most likely two reductive phrases on this context. You’ve got inputs, you’ve gotten outputs. The inputs are customers answering questions. The outputs are the solutions to these questions that individuals come and seek for. There’s a complete neighborhood that makes that system run. The software program platform manages that neighborhood with the occasion moderators, however it’s actually inputs and outputs. There are people who find themselves asking questions and people who find themselves answering questions.

Each side of which can be deeply affected by AI. I believe now we have come to the open net a part of the dialog, the place the enter aspect is being flooded by AI generated slop. In 2022, you needed to ban AI-generated solutions in Stack Overflow. Then, on the output aspect, the flexibility for AI instruments to provide the solutions is overwhelming.

So, let’s simply break it into two components. How did you consider the enter aspect, the place there’s a flood of individuals saying, “Oh, I can reply these questions sooner than ever by simply asking ChatGPT and pasting the reply in. Perhaps that’s not ok, however I can simply do it.” Then, how did you consider the output aspect?

We observed two issues proper out of the gate. One was the variety of questions that have been being requested and answered on Stack went via the roof as a result of individuals began utilizing, to your level, ChatGPT to reply these questions. That fueled this spike, which is sort of counterintuitive, however I believe individuals simply felt like, “Wow, I can sport the system, so let me go do it.” In a short time, we needed to be extraordinarily shrewd, and our neighborhood members are superb at determining what’s actual and what’s not. They have been capable of name out in a short time that these posts have been really ChatGPT generated. That’s sort of what initiated the ban, which we fully supported and nonetheless assist, by the way in which. You continue to can not reply any of the questions on Stack Overflow with AI-generated content material.

The explanation for that, Nilay, is as a result of our proposition is to be the trusted important supply for applied sciences. That’s our imaginative and prescient for the corporate. So for us, it’s all about ensuring that there are just a few locations the place you’ll be able to go and never take care of AI slop, the place a neighborhood of specialists have really voted and curated it so you’ll be able to belief it for varied functions. On the enter aspect, it made sense to try this, and we proceed to try this.

Quick-forward just a little bit to now, and now we have created all types of latest entry factors onto the positioning, regardless that we’ve had excessive requirements to ask a query on Stack Overflow. We simply launched our AI Help function into common availability earlier this week, and it’s been tremendous thrilling to observe how customers are utilizing that. It’s successfully an AI conversational interface grounded on our 90 million questions and solutions.

Then, there’s the flexibility for individuals to ask subjective questions, going again to our final dialog three years in the past. Now individuals are capable of ask open-ended questions as a result of there’s a spot for Q&A, which is the canonical reply to a query. There’s additionally a spot for dialogue and dialog as a result of there’s a lot altering. It’s not like all of the solutions have been found out, so let’s simply be sure that individuals have a capability to try this. That’s aligned with our mission of “cultivating neighborhood,” which is without doubt one of the three components of our mission. The opposite ones are “energy studying” and “unlocking development.” So, now we have executed all these items to be sure that we’re not restrictive on the entry level and the question-asking expertise.

The opposite factor on the reply aspect is that we realized it’s crucial to go wherever the person is spending time. Now that the world has modified and individuals are in truth utilizing Cursor AI and GitHub Copilot to jot down their code, our aim is to be the important supply for know-how. So let’s present up wherever our customers are. We’veactually grow to be much more headless.

For instance, we lately launched MCP servers for each our public platform and our enterprise product. What individuals are utilizing our platform to do now’s to not solely invoke these MCP servers — let’s say in the event that they’re writing code in Cursor and wish to know the distinction between model one and model two — however to additionally to have the ability to write again to our platform straight from Cursor in the event that they wish to interact and get a deeper reply, which could be very distinctive within the trade.

So, that’s been our product precept: simply go wherever the person is. However in the end, we simply wish to be that reliable important supply for technologists, whether or not it’s inside firms or exterior firms.

How do you monetize in a world the place you’re headless, the place you’re simply one other database that somebody’s querying from Cursor? How does that make you cash?

We earn a living primarily in two methods. Now we have a 3rd means, fortunately, however the third means is the smallest half, so I’ll begin with the most important. Our enterprise enterprise, what we name Stack Inner, is now utilized by 25,000 firms around the globe. Among the world’s largest organizations, banks, tech firms, and retail firms use this product to share data internally. More and more, they’re now ready to make use of that reliable data to energy their AI assistants and AI brokers to go do varied issues.

A very good instance of that is Uber, which is a buyer of Stack Overflow Inner and it has Uber Genie. It has hundreds of questions and solutions on our platform. Uber Genie plugs into that content material via our APIs, after which it’s ready to enter issues like Slack channels to robotically reply questions and drive productiveness so that you simply’re not bothering individuals. It’s rooted within the group’s data on our platform.

So, the enterprise enterprise is our major enterprise. The second enterprise is our information licensing enterprise, which we really constructed solely over the previous couple of years. One of many issues we additionally observed was that numerous the AI labs have been leveraging our information for LLM pre-training and post-training wants, together with retrieval-augmented technology (RAG) indexing, so we put up a complete bunch of anti-scrapers. We labored with third-party firms, and really shortly we received calls from numerous them saying, “We’d like entry to your information. Let’s work collectively to formally get entry.” We had to try this, and now we’ve struck partnership agreements with each single AI lab that you can imagine, each cloud hyperscaler that you can imagine — firms like Google, OpenAI — and even partnerships with Databricks and Snowflake, regardless that they’re not doing LLM pre-training. That’s been our second enterprise extra lately.

And the third one, which is the smallest a part of our firm, is promoting. I believe most individuals assume that Stack Overflow is supported completely by promoting, however it’s solely about 20 % of our firm income. Now we have a really captive, crucial viewers of builders who spend time on the positioning, so now we have giant advertisers that wish to get their consideration on varied merchandise. Actually, there’s numerous competitors now, in order that they more and more wish to do this. That’s how we earn a living.

So, within the context of changing into headless, for us it’s about our enterprise product. It really works on a subscription and with hybrid pricing. That’s how we earn a living there. The info licensing is analogous in that if individuals need entry, they’ve received to pay for that. Then, sure, promoting is restricted to among the largest firms, and so they pay us for that. However there’s all the time going to be… I might say it’s an “and” versus an “or.” We’re not going to be fully headless. I believe we simply wish to give the person the choice to be headless. Loads of individuals nonetheless come to the positioning, and in that case, we’re capable of stability that out with these mechanisms.

Do you assume new customers are going to come back to Stack Overflow? Stack Overflow is a product of the cell period. There’s an explosion of software program improvement. There’s an explosion of neighborhood. There’s a tradition within the worth of constructing apps and companies, and there’s new instruments. Stack Overflow is without doubt one of the central gathering factors for that neighborhood in that period.

New builders at the moment may simply open Cloud Code, Cursor, GitHub, or no matter, and simply discuss to that. They may by no means really enterprise out right into a neighborhood in an analogous means. Do you assume you will get individuals to come back to Stack Overflow immediately and hunt down solutions from different individuals, or are they simply going to speak to the AIs?

I believe that for easy questions… By the way in which, after we noticed the questions decline in early 2023, what we realized is that just about all these declines have been with quite simple questions. The complicated questions nonetheless get requested on Stack as a result of there’s no different place. If the LLMs are solely nearly as good as the information, which is usually human curated, we’re among the finest locations for that, if not the most effective for know-how. It’s nonetheless a really lively website with numerous engagement and numerous month-to-month lively utilization.

The questions being requested are fairly superior, I might say. What we’re additionally more and more seeing via the brand new mechanisms that we’ve opened up… as a result of to reply your query, we wish to give individuals different causes to come back to the positioning apart from simply getting their solutions. So, now we have needed to broaden our website’s goal, therefore the mission of “domesticate neighborhood, energy studying, and unlock development.”

What we’ve executed is open up new entry factors, new methods for individuals to interact. We, for instance, unlocked the flexibility for people to speak with one another to get directional steerage. That’s been a very talked-about function on the positioning the place individuals are participating with different specialists. For instance, now we have individuals asking OpenAI API questions, and so they can go into the OpenAI chatroom and have interaction with different individuals who have related questions, or Python specialists.

We additionally opened up the flexibility for individuals to reveal their data with challenges, successfully like hackathons. We’ve opened up a complete sequence of challenges, that are a very talked-about function now. Folks spend time to go and resolve these challenges that we put up, and that means, they’ll showcase their understanding of the basics, which I believe is essential by way of the place the world goes.

If individuals are simply utilizing vibe coding instruments and code gen instruments, firms bringing in younger expertise have to know that they’re counting on individuals who not solely took the shortcut, but in addition perceive the basics. We’re one of many few locations the place you’ll be able to really show that you simply’ve discovered the basics. So, that’s the opposite cause why we’ve opened up these new mechanisms.

Then, there’s the third a part of the mission, which is unlocking development. There’s going to be numerous job disruption due to all this. If individuals’s jobs are going to vary fairly dramatically, junior builders are going to wish a house, regardless that I believe it’s a shortsighted transfer by many firms to cease hiring them contemplating you want a pipeline. They’re going to wish to attach with different individuals, to have the ability to progress, be taught, and get jobs. Jobs are an important half. We struck a partnership with Certainly this previous yr to accomplice on tech jobs. It’s simply to broaden the scope of our website in order that there are lots of different causes aside from asking the questions. They nonetheless do, however we additionally wish to give them extra causes to come back to the positioning.

This involves the large rigidity in all of this. I see it enjoying out in every kind of various communities. I see it enjoying out in our personal feedback in numerous methods. You wish to construct a neighborhood of people who find themselves serving to different individuals get higher, and that’s being disrupted on each aspect by AI. Communities which can be constructed round individuals are fairly immune to the incursion of AI.

This has undoubtedly occurred on Stack Overflow. Your moderators have primarily revolted over the flexibility to take away AI-generated solutions as quick as they wish to. If you partnered with OpenAI, a bunch of customers began deleting content material so it wouldn’t be fed into OpenAI for coaching, and also you needed to ban a bunch of them. How are you managing that stability? As a result of when you construct communities round individuals, I might say the tradition — proper now anyway — is that these communities will push again in opposition to AI very exhausting.

I might say one of the essential issues that we’ve centered on (and that I’ve hung out on over the previous few years), is that this complete push and pull of how we take into consideration AI within the context of our website. As a result of it’s fairly clear to us that if we don’t modernize the positioning within the context of us leveraging AI as an entry level that it’s going to be much less related over time. That’s not good. So, we’ve taken a really aggressive stance by incorporating AI into the general public platform with AI Help, which has been improbable to see. I’ll stroll you thru the choice on why we did that. Then, we did the identical factor on the enterprise aspect.

If I take into consideration the person base at Stack Overflow, it’s sort of like an enormous nation, proper? We’ve received 100 million individuals, and there’s undoubtedly individuals on either side of the spectrum. Now we have one thing referred to as the 1-9-90 rule. One % are the hardcore customers who’ve spent numerous time with their blood, sweat and tears curating data, spending their time on the positioning, and contributing. 9 % are doing it in a medium means, and 90 % are consuming and largely lurking.

We ask individuals on the positioning whether or not or not they’re utilizing AI. Our personal surveys principally say, when you took a have a look at the Stack Overflow 2025 Developer Survey, over 80 % of our neighborhood members are utilizing AI or intend to make use of AI. Eighty %. However the belief degree once they’re utilizing AI is barely about 29 %. Solely 29 % of our person base really trusts what’s popping out of AI, which is definitely fairly applicable contemplating the place we’re as a result of there needs to be skepticism of this new know-how.

So, there’s enthusiasm to strive it however to not totally belief it. And with this 1-9-90 rule, I believe what now we have is a core group of customers which can be all the time going to be the protectors of the corporate’s unique mission, which was to create this fully correct data base and do nothing extra. Then, now we have a really giant variety of people who find themselves, let’s say, the subsequent technology of builders, who need to leverage the newest and best instruments. It’s very clear to us primarily based on surveys and extra analysis that they wish to use pure language because the interface to have the ability to do that.

It’s the most significant change by way of pc science improvement. In the event you look all the way in which again to object-oriented programming many a long time in the past, that wasn’t really such an enormous growth. It didn’t really create this kind of change. However now, we’re on this second the place every thing’s been unlocked. It’s an enormous change effort, and we’ve needed to determine to respect the unique mission and preserve accuracy on the coronary heart of it. We’re not comfy utilizing AI for solutions, for instance, as a result of it is going to generate slop. It hallucinates, therefore why the belief rating is low. However why don’t we incorporate pure language interfaces in order that’s the popular strategy to interact? So, we ended up doing that, each on the general public aspect in addition to on the enterprise aspect.

That’s been effectively acquired by the overwhelming majority of customers, however there’ll all the time be a vocal minority who will push again in opposition to incorporation. Past the positioning, there’s only a degree of broader concern about what all this does to jobs, and what’s going to occur if we let the cat out of the bag. So,there’s that concern additionally, which is comprehensible.

Let me put a tremendous level on that. I believe I perceive that in a sharper means. If I’m anyone in your 1 % who spends numerous time on Stack Overflow serving to different individuals. The explanation I reply questions free of charge in your platform, which you monetize in a number of methods, is as a result of I can immediately see that my effort helps different individuals develop and that I’m serving to different individuals resolve issues. That’s one very self-contained dynamic. The final time you have been on the present, our complete dialog was about that dynamic and the way you bought individuals to take part in that dynamic and the worth of it.

Then all of the sudden, there’s a really clear financial profit to the corporate that owns the database as a result of it’s promoting my effort to OpenAI, which is going on throughout the board. It’s going to do these information licensing offers with all these AI suppliers, they’re going to coach on the solutions that I’ve painstakingly entered into this database to assist different individuals, and now the subsequent technology of software program engineers goes to get auto-complete that’s primarily based on my work and I’ve gotten nothing. I’ve heard that from heaps and many individuals. I’ve heard that in our personal neighborhood, and I believe I’ve felt that as varied media firms have made these offers.

How do you reply to that? As a result of it feels such as you have been offering a database that you simply needed to monetize in some methods, however the interplay individuals had was the worth, and now there’s one other sort of financial worth that’s possibly overshadowing, recasting, or re-characterizing the interplay that individuals have.

There are a few factors there. One is about this firm’s unique DNA and why individuals got here collectively to do that factor. Once I joined the corporate, I requested a query like, “What’s individuals’s incentive to spend time doing this?” I requested the founders, particularly [co-founder] Joel Spolsky, about this. His level was that the software program improvement neighborhood could be very altruistic. Folks simply wish to assist one another out as a result of individuals perceive how irritating… I used to jot down code a few years in the past. I lately picked it again up with among the code-generation instruments, which is attention-grabbing to check and distinction. I simply keep in mind how irritating it was when you received caught on one thing. Stack was an enormous boon when it was created to unlock this. It was really out of that. That was the explanation.

Even earlier than ChatGPT, we additionally requested the query, “Ought to we incentivize customers by paying them? Ought to we give them a financial profit?” That wasn’t a excessive ask by a person base. We went and researched individuals. Folks weren’t in for the cash. Plus, it complicates issues as a result of how do you choose the cost for a selected JavaScript query relative to a selected Python query? It goes down a rabbit gap, which is untenable. In order that’s one. What was the unique cause individuals received collectively; it was concerning the mission.

Secondly, by way of why now we have to do that and if it’s unfair. The first cause now we have to go down the licensing route is as a result of the mannequin of the web has actually been turned the other way up. I do know you discuss this, Nilay, with the “DoorDash downside.” Folks relied on the mannequin of the web the place individuals go to search engines like google and yahoo and web sites and also you monetize off of advertisements. I actually empathize with content material websites which can be closely depending on promoting as a result of I believe most content material websites’ visitors is down 30 or 40 %, one thing like that. There’s this large seed shift the place firms that assist these platforms need to… we’re a enterprise in the end. So, what do now we have to do? Now we have to do what is important and undertake a brand new enterprise mannequin to outlive, thrive, and do all of the issues.

Fortunately for us, we had an enterprise enterprise, which is unbiased of all of this. Fortunately for us, we nonetheless had the promoting enterprise, and huge advertisers nonetheless cared about our neighborhood. So, information licensing solely felt proper by way of ensuring that we will successfully capitalize on the second and have the ability to make investments again into our neighborhood in order that people who find themselves there for the proper causes noticed the advantages. We’ve invested with all these new options I simply talked about. Whether or not it’s these new content material varieties, challenges, chat, AI Help, any of these items, all of them take assets to go and construct. So, we needed to go and leverage the funds we acquired to have the ability to go do this.

Now sooner or later, we might think about different methods. For instance, ought to we pay our customers, give them a chunk of the information licensing income? Maybe. We all the time ask that query. There are all the time methods for us to proceed, however that is the present setup that now we have proper now. It’s balancing numerous issues.

You talked about that to get to the information licensing offers, you needed to put up a bunch of anti-scraper instruments. You had to enter secondary and tertiary layers of the stack to get offers from Databricks and different kinds of suppliers. The AI firms have been simply scraping your website earlier than. They most likely nonetheless are. Whether or not or not they’re paying you, they’re most likely nonetheless simply going via the entrance door as a result of all of them seem like doing that. Did it’s important to say, “We’re stopping you,” after which go get the deal? Or did you say, “Hey, we all know you’re doing this, however it’s important to pay us or we’re going to start out litigating?”

It’s someplace in between. We put up the anti-scrapers in a short time. We even modified the way in which through which individuals acquired our information dumps. Once more, there’s a stability as a result of we by no means needed to stop our neighborhood customers from grabbing our information for his or her reliable wants, like their faculty tasks or PhD theses. So, we’ve continued to be open about our information for our neighborhood members, however they need to be neighborhood members, and there could be firms trying to commercialize off the information.

We have been very particular concerning the coverage phrases. We put up know-how that prevented individuals from grabbing it, so we knew precisely who’s scraping and who’s not scraping. We reached out to a few of these of us and stated,, “Look, stand down since you’re placing numerous strain on our servers by doing what you’re doing, so take it straightforward in right here.”

However I believe my characterization of these firms is that they don’t care. A few of them care and wish to be good residents, and a few of them completely don’t care and they’d choose the smoke. You may simply categorize them. There’s a cause Amazon is suing Perplexity. They advised Perplexity to cease it and Perplexity gained’t. The New York Instances, as we’re talking at the moment, is suing Perplexity. Then, there are different gamers appearing in numerous methods and putting totally different sorts of offers.

Stroll me via a kind of offers. If you went and struck your take care of OpenAI, was it, “We’re going to cease you, and if you would like the door to be open once more, it’s important to pay us?” Or was it, “You realize that is incorrect. We will take all of the technical and authorized measures, however we must always really simply get to the deal appropriately?” Stroll me via that dialog.

We have been incorporating one thing like OpenAI into our product. Keep in mind the code pink state of affairs the place we have been about to announce our AI response. So, we have been really utilizing that know-how to do what we needed to do to include AI into the general public platform and our enterprise product. We had a relationship with them, and we additionally stated, “Look, this isn’t going to work. It’s not tenable, and that is the brand new means of working. Perhaps we’d like a brand new enterprise association so that you can use the information. Let’s even have a dialog.” And credit score to them, they have been very partner-centric round that. I used to be very impressed by OpenAI and corporations like Google which can be all very open to participating on this matter and needed to be accountable AI companions.

They received it instantly, even earlier than we requested them. It wasn’t this huge, “Let’s go have this dialog from the bottom up and justify why it needed to be executed.” We simply stated, “Look, that is what must occur as a result of this can be a new enterprise mannequin.” We received into the dialog fairly shortly. “What precisely are you searching for? Which format of knowledge do you wish to scrape the content material? Would you like bulk uploads? Would you like API calls? What would you like?”

So, we received into that complete combine. And remind you, Nilay, these are recurring revenue-type offers. These usually are not one time funds. If you’d like entry and also you need continued entry sooner or later, you’ve received to maintain paying, even with the historic information. In order that’s how these are arrange.

So sure, they have been very collaborative companions. However you’re proper. There are contradictory gamers. They are saying one thing and their actions show different issues by way of how they’ve engaged. There are holdouts for positive and people who find themselves not precisely in step with their phrase, and that’s unlucky. I believe each firm like us has to determine what to do about that. We’re in varied levels of those conversations with individuals on how to ensure we sensibly get them to do the proper factor.

Now it’s important to identify a kind of firms. Who do you assume is holding out in another way than their public posture?

I’d relatively not be particular, however all the same old suspects that you simply’re overlaying are the same old suspects that we’re encountering. That’s how I might put that.

Let me ask you concerning the recurring income piece after which I wish to get into the Decoder questions as a result of I believe they’ll be illuminating after this dialog. There’s a way that we’ve executed all of the pre-training that we’re going to do, proper? Scraping the web shouldn’t be the way forward for these fashions, and there must be another leap.

Stack Overflow’s current corpus of data is the precious factor. There’s numerous info. There’s 20 years of stuff in that database. What’s the worth of “it’s important to pay us once more to coach the subsequent model of Gemini or GPT” and the worth of “there’s incremental info being added to the prevailing database?” As a result of that looks like a transparent break up to me.

The way in which we’ve thought of that is that each mannequin that’s being educated is educated on some corpus of data. You’re going from GPT X to Y. In the event you’re leveraging our unique information or some by-product of that from a previous mannequin within the new mannequin that you simply’re coaching, then it’s important to pay us for it. That’s successfully the authorized requirement for doing that. So, it’s a cumulative side. Let’s not overlook that. Folks need to pay for the cumulative information. It’s not simply that it was used again within the day. And sure, relative to twenty years, one yr’s value of data goes to be much less, however that’s why you’re getting 20 plus one. That’s the thought. So, that’s the way in which the authorized settlement has been arrange.

Is it per yr? Is it that yearly’s value of knowledge is a piece of cash? How does that work?

No, it’s cumulative, like the entire corpus: historic information in addition to something going ahead for the next yr. All that’s one collected information set, and that’s successfully charged as one.

So, this yr’s information doesn’t get pulled into Gemini 3’s information set, which simply got here out, proper? Each new query and reply in Stack Overflow since Gemini 3 got here out shouldn’t be integrated in Gemini 3’s coaching.

So that you’re sort of betting that they’re simply going to coach ever-bigger fashions. Is that the way it’s structured in your thoughts?

Yeah. And a few firms have requested for extensive use circumstances. There are pre-training use circumstances. Even past that, you’ll be able to leverage the information in many various methods for AI and non-AI use circumstances, like search use circumstances. However right. There could also be eventualities the place extra bigger fashions are constructed, and our information goes to be helpful for these eventualities. However there’s going to be RAG indexing, post-training wants, all types of eventualities. It’s fairly attention-grabbing to see among the frontier labs ask for very particular slices of knowledge that they discover helpful.

Keep in mind, there’s not solely questions and solutions. We’ve received the remark historical past, the metadata historical past, the voting historical past, the historical past of Person A taking place this path. So, it’s numerous wonderful context for issues like reasoning and with the ability to mimic the human mind. It’s nearly like one human mind that’s been documented.

This, I believe, brings me to the Decoder questions. You’ve restructured the corporate. There’s been some rounds of layoffs. You’ve refocused on the SaaS enterprise in an actual means. I believe we must always discuss that. However there’s the concept that we’re going to coach ever-bigger fashions and that would be the rising a part of the enterprise, versus wanting some slices, versus RAG really being the longer term for lots of those different companies. You’d make totally different choices primarily based on which a kind of goes to develop sooner, and I don’t assume anyone is aware of. Perhaps you recognize. You may inform me if you recognize or you recognize somebody who is aware of,

However we’re in a really nascent interval for all of this improvement. How have you ever structured the corporate to even out all of that danger and be ready for a way individuals will really want the information sooner or later? Perhaps you recognize, I don’t assume I do know.

It’s exhausting to foretell clearly. We’ve received some sensible minds, like Demis Hassabis at Google and others, who’re developing with the subsequent generational leap of regardless of the equal of transformer know-how is to maintain going in the direction of this final aim, which is AGI.

So sure, you’re proper. It’s exhausting to know precisely what exhibits up and when. Nevertheless, the way in which we’re structured as an organization is in successfully two components. One half is the enterprise enterprise. Now we have a product group, engineering group, and a go-to market group centered on that. The enterprise merchandise enterprise could be very clear. Then, the opposite aspect of the home is what we name “neighborhood merchandise”. That group focuses on the general public platform, all of the options that we’ve talked about to this point, AI Help and all of the subjective new questions, and chat. That is the neighborhood aspect. The info licensing enterprise sits in that group, and so they’re tied to the engagement of the positioning. So, there’s a virtuous cycle there.

So, that’s how we’re break up. Once more, that features product assets, a small go-to market group, engineering of us, and many others. Additionally, there’s our neighborhood administration group, which spends numerous time with the moderators to interact there. So, it’s break up down the center, and our different features assist each.

How huge is Stack Overflow at the moment? I do know you laid off nearly 1 / 4 of the corporate in 2023 due to visitors declines. You’ve constructed different companies. How huge are you at the moment?

We’re about 300 individuals or so. Yeah, that’s our measurement.

Do you assume the income you’re going to see from information licensing or your SaaS enterprise goes to can help you develop once more?

We consider so. We’re a rising firm. We’re worthwhile. So financially, we’re fortunately in an excellent spot. Now, it’s all about inserting bets on the very best development alternatives. We consider creating this data intelligence layer inside enterprise via our Stack Inner product is an outstanding development alternative as a result of clients are pulling us in that path, which is improbable to see.

That’s the place you’re headed. These are a few of your bulletins. I wish to discuss that in a second, however I’m simply centered on the longer term is the SaaS enterprise for enterprise. The longer term is information licensing. Are you continue to seeing declines on the general public web site?

I might say it’s stabilized for a number of months. I believe the engagement and the exercise on the positioning are literally fairly steady. The drop in questions that I discussed beforehand have been all the straightforward questions, and it appears to have gotten to a spot the place complicated questions are being requested. Now we have a constant variety of individuals on the positioning each day. Now we have one thing referred to as a heartbeat. Actually, anyone can go test it out. In the event you go to StackOverflow.com and scroll to the underside, you’ll see what number of customers are on-line in the intervening time. And so that you’ll all the time see a really constant quantity there. I believe it’s exhausting to foretell the longer term, however actually I believe the worst of it was again in 2023, 2024, for positive..

The query I ask all people on Decoder, as you effectively know, is how do you make choices. The final time you have been on the present, you stated that you simply needed to be on the entrance traces as a lot as doable, and also you needed to learn by people who find themselves on the bottom. Has that modified previously three years? What’s your decision-making course of?

Not likely. I believe it’s crucial for leaders and other people like CEOs to have the complete context as a result of you’ll be able to’t have filtered info. I spent numerous time with customers and clients actually understanding what they care about. That’s how we even selected one thing as controversial because the AI Help function. It wasn’t apparent to say, “let’s go and construct that” if you weren’t listening on the bottom. As a result of when you simply take heed to the headline statements, it appeared like individuals didn’t wish to combine AI into Stack Overflow. However the actuality is that many, many customers — the 90 % I discussed — needed a pure language interface. That’s what they’re comfy utilizing nowadays, and that’s what they needed. In order that’s why we determined to try this.

One of many issues that I see in every single place is that break up. You talked about 1-9-90 earlier than. There’s a really vocal minority. We see it in our personal visitors on TheVerge. We cowl AI deeply, and we’re advised that everybody hates it. I perceive why. I perceive the feedback. That’s what I’ll say. I get it.

Then, I see the utilization numbers. I see the visitors on our protection of AI instruments. I see firms like yours saying that everybody’s utilizing it. There’s some gigantic break up there that’s in contrast to some other I believe I’ve ever encountered overlaying know-how over the previous 15 years. Everybody says they don’t prefer it, after which they’re utilizing the hell out of it.

The one different one I can consider that’s barely comparable is how individuals really feel about Adobe. Everybody makes use of the instruments and everybody’s mad on the Inventive Cloud subscription charge. It’s principally the one comparability I’ve. It’s not a very good one, it doesn’t map one-to-one, however it’s as shut as I’ve come to that break up.

What in your thoughts accounts for that break up with AI the place individuals don’t prefer it, are very vocal about not liking it, after which we see the numbers and everybody’s sort of utilizing it anyway?

I believe it comes right down to that information level I shared earlier: that 80-plus % of our person base needs to make use of AI or is already utilizing AI for code-related matters, however solely 29 % of that inhabitants trusts AI. Belief is a really deep phrase. Why don’t you belief one thing? You don’t belief one thing since you don’t assume it’s producing excessive integrity, correct solutions. You could not belief it as a result of it might change you in the future, and also you don’t like that both.

However on the similar time, you’re going to be interested in what’s going to be such an iconic power. So, you wish to preserve attempting and utilizing it and maybe getting higher to in the end, hopefully, leverage it to your profit so that you could be related as a person developer or to go rather a lot sooner sooner or later.

I believe that’s most likely the explanation, particularly with the developer viewers. I believe they’re a really discerning, let’s say analytical viewers, and they are often prickly if issues usually are not deterministic with the way in which it has been for a really very long time. This can be a very probabilistic kind of know-how. It’s nearly like going to a on line casino and utilizing a roulette wheel. You’re going to get a unique reply each time. It’s not essentially comforting for anyone who’s writing very particular code and searching for very particular outcomes.

I believe individuals will get used to that over time. It’s a thoughts shift change for individuals writing software program. Which may be the explanation why individuals are intrigued, as a result of it’s so highly effective as a know-how. Don’t get me incorrect. We use vibe coding in all places at Stack. All of the options I discussed to you, our designers and product managers vibe coded it first to indicate it and get person suggestions earlier than we went and constructed it. We’ve embraced these instruments internally for that profit. So, there can be methods through which you’re feeling comfy utilizing it, however I believe that’s the core cause.

I really wish to discuss that dependency. You realize that it’s not reliable, however you’re constructing merchandise with it. You might be constructing merchandise to allow it. The large announcement this week is Stack Overflow AI Help. You’ve talked about it a number of occasions all through this dialog. You’re betting that that is what individuals need, proper?

You’re betting that an AI-powered device on Stack Overflow will assist extra individuals. Perhaps that factor goes to hallucinate like loopy and provides individuals the incorrect reply. Perhaps it’s going to, I don’t know, inform Kevin Roose that it loves him on the entrance web page of The New York Instances. I identical to teasing Kevin. Hello, Kevin.

How do you make that guess when you recognize that the customers don’t belief it, however you continue to need to roll out the instruments as a result of that’s the place the trade goes?

We consider we’ve really unlocked an important side of that belief challenge and responded to it. Our AI Help function is a RAG plus LLM answer. Successfully, it offers a solution that first goes out into our corpus of tens of hundreds of thousands of questions and solutions. Now we have 80 to 90 million. These are first used to supply a response. In the event that they don’t, there’s a fallback possibility the place it goes and leverages our accomplice OpenAI, for instance, to have the ability to go and produce reliable data from different components of the online. So, it first searches via our trusted, attributed data base. It produces the hyperlinks so individuals can go down that path and be taught extra about it. Attribution is essential to us, proper?

That’s how we’re navigating the hallucination ingredient., We’re continuously testing it, and it’s not good. There’ll all the time be enhancements. However we’re additionally the place the world’s headed, and if these fashions proceed to get higher, then we must always profit from these enhancements. Finally, we must always have the most effective answer since you’re getting grounded human context plus the LLM strengths.

I believe the factor that I’m most considering is the religion that the fashions will proceed to get higher. I’m not 100% positive that’s true. I’m unsure that LLM know-how as a core know-how can really be clever. As you’re saying, individuals are very interested in the pure language element of those fashions and the interface shift that’s taking place. There’s the platform shift that we’re all going to develop software program with pure language or let the LLMs cause and principally self-prompt themselves into a solution.

There’s one thing there that appears dangerous. Are you perceiving that danger at the moment? Are you factoring that in or are you saying, “That is the place we’re at now and now we have to proceed till one thing adjustments?”

I believe the very first thing across the LLM enchancment aircraft is that I’m with you. It’s exhausting to understand how issues are going to enhance. If you simply take into consideration the previous six months and that it’s plateaued, growth, right here comes Gemini 3. Once more, we’re proud companions of Google. It’s a seed shift. It blows each different mannequin out of the water. Now, we’ve received a code pink state of affairs in different firms, different LLM rivals. So, it’s —

Sam Altman did kind the phrases “code pink,” by the way in which. I wish to say that. In order that’s superb.

Maybe that was the way in which for me to return within the day. The purpose is that it’s stunning that you simply’re capable of produce that kind of a leap when issues have seemingly plateaued. I don’t know. I can’t predict that as a result of these of us like Demis are deep, deep within the topic. So, that’s true, however there’s additionally going to be different improvements that we aren’t aware about. Like I used to be explaining beforehand, transformers have been clearly an enormous improvement on this house. There could also be one thing that these AI analysis labs give you that we’re not even conscious of that’ll push issues. Finally, we all know that the compounding results are very actual. We’ve received limitless compute, extraordinarily highly effective chips, and GPUs that at the moment are decreasing their prices.

I used to be at AWS re:Invent this week the place they talked concerning the Trainium chips, withTrainium3 and Trainium4 being constructed out. There’s going to be simply the proliferation, and you then’ve received entry to information, which we’ve already talked about. So, when these items mix and compound collectively, it’s going to supply very magical outcomes. I believe that’s the idea and why my very own assumptions are rooted in the truth that it’s going to enhance total.

The explanation I’m asking about this within the context of the instruments you’re constructing and everybody utilizing it however solely 29 % of individuals trusting it’s since you’ve received to carry that quantity means as much as attain the returns that each firm investing in AI is attempting to succeed in, together with yours. I don’t know if the core know-how can do this. I don’t know when you can stack a bunch of applied sciences to try this.

However I do know that one model of the way forward for software program improvement is every thing vibe coding on a regular basis. One other model of the way forward for software program improvement appears like writing intensely lengthy prompts for fashions which can be pages and pages themselves, which appears ridiculous to me however possibly that’s the future.

One other model appears like, “We return to people on the reducing fringe of software program improvement, and they’re co-developing with an AI mannequin after which possibly asking Stack.” That seems like a richer, extra attention-grabbing future, however it’s unclear to me the place we’re on that spectrum or how that even performs out.

I believe that now we have not solely a chook’s eye view within the context of our public neighborhood with this 29 % information level, however our enterprise clients give us a transparent view on the place they’re. The ROI query is being requested very closely inside firms, and I believe 2026 goes to be the yr of rationalization. If 2025 was the yr of brokers — the place each device’s being tried out inside these firms — there’s a really open panorama for CTOs to go and purchase and check varied instruments. So, I believe it’s been an incredible time for among the firms constructing these instruments.

However 2026, the CFOs will strain with, “Okay, productiveness enhancements have to come back from these. We’re going to rent much less individuals as annual planning occurs.” There’s going to be great strain within the system to show what the actual worth is. All people on the senior degree that I’ve talked to acknowledges that this can be a huge shift, and so they all are leaning into it fairly exhausting. They’re all ready for the enhancements. I believe most firms will say that they’ve seen enhancements within the small teams the place they’ve examined these instruments, however that’s a self-selecting group as a result of they’re fanatics. They’ll see nice productiveness good points, which might be true, however there’s an absolute drop-off in productiveness as you consider the adoption throughout the enterprise. In all probability for a similar causes, by the way in which. You’re telling workers to make use of instruments which can put them out of a job, so why would they wish to do it? Or extra essentially, if these instruments usually are not good and so they’re hallucinating and can be held accountable, that’s not good both.

Then in fact, there’s the method change, the mindset change, that it’s important to fully change your workflows of how you’re employed, all of the enterprise change administration work. That’s the reason our answer, Stack Inner, is constructing this human curation layer, or data intelligence layer. With the MCP server on high, the data base within the center, after which our capability to ingest data from different components of the corporate, we will create these atomic Q&A which can be extraordinarily useful to root your enterprise data in these AI brokers. That’s that answer. That’s why we’ve actually gone exhausting at producing that.

We’ve seen a very sturdy response from our clients. Now we have among the world’s largest firms leveraging, testing, and constructing this with us, like HP, Eli Lilly, Xerox, all these firms. It’s been superb to see them gravitate to us as a result of they wish to fulfill that ROI level that you simply’re making. What are the gaps? It’s belief once more. So, they need this belief layer via an organization like Stack that they’ll really insert in between their information and their AI instruments.

If you say the age of rationalization, what I hear is that you simply assume the bubble’s going to pop in 2026.

I believe the exuberance in simply attempting out varied instruments and these limitless AI budgets will in the end come to roost. I’m unsure concerning the bubble bursting. There’s undoubtedly going to be corrections alongside the way in which. There’s no query about it when you have a look at historical past. The variety of distributors which can be promoting into these firms. What I’m very shocked by is that there’s related performance. There’s 4 of them being examined inside these firms.

All these firms might have gotten to $100 million in annual recurring income (ARR), however sooner or later there’s going to be churn when the CTO decides, “You realize what, I’m solely going to possibly use one and possibly a second one as a backup.” No totally different from the cloud. If you consider the multi-cloud world again within the day, individuals didn’t have three clouds out of the gate. Now you’ve gotten possibly one major cloud and one secondary cloud. That is totally different, however on the similar time, you’re going to have like 4 totally different vibe coding instruments for my part.

When you consider Stack Overflow as being that belief layer, that’s the worth add, that’s what you’ll be able to possibly cost a premium for over time. You’re nonetheless depending on a device that solely 29 % of your customers belief. And I do know you’re speaking about RAG and that your different techniques are doing that. How do you assume you carry that quantity up with Stack Overflow? Is that doable so that you can do or does the ecosystem need to do it for you?

I believe it’s an ecosystem level typically talking as a result of that’s extra a mirrored image of what individuals have entry to past Stack, proper? They’ve received entry to all these different choices. What we will give attention to is being probably the most important supply for technologists. So for us, it’s about ensuring this content material is great and prime quality, however that it’s additionally an awesome place for individuals to domesticate neighborhood, join with one another, be taught and develop of their careers.

I believe the way in which we will do it’s via the truth that we’re working with all these huge AI labs and the truth that our reliable data that’s been painstakingly human curated goes to move into these LLMs, which in the end produce reliable solutions. We’re kind of one layer behind, however that’s the place we function. We function in that belief layer or the information layer, if you’ll, within the context of LLMs. So, that’s our oblique contribution to that 29 %.

Prashanth, this has been an awesome dialog. You’re going to have to come back again before three years subsequent time. What’s subsequent for Stack Overflow? What ought to individuals be searching for?

Our greatest focus can be ensuring that we construct this enterprise data intelligence layer for firms to really use AI brokers in a reliable means. We’re very, very enthusiastic about our Stack Inner product on the enterprise aspect, which we launched a few weeks in the past at Microsoft Ignite, in addition to serving to our neighborhood customers join with one another to essentially be taught in addition to develop their careers on the general public platform, as I’ve talked about all through.

There are going to be so many avenues and new entry factors, like our AI Help, our subjective content material, chat, and different issues that individuals hopefully discover helpful as issues change round them very quickly. They are often a part of this superb neighborhood and assist one another out. So these are the 2 focuses, enterprise in addition to our public neighborhood.

All proper. Effectively, when the bubble pops subsequent yr, we’re going to have you ever come again and we’re going to say you predicted it.

Thanks, Nilay. I recognize it.

_{Questions or feedback about this episode? Hit us up at decoder@theverge.com. We actually do learn each e-mail!}