Lightspeed search built for devs (Interview)

30:28The more nodes you go down, the slower it gets to do your lookup. So you have to find a balance between like, how many places can you go before it starts not only affecting performance but also like accuracy because then you start matching to a lot of things that like are not good matches which means your search is actually becoming less relevant when the intention is to make it kind of more relevant and still match when things right when the user makes mistakes but the data is there somewhere is there like a sweet spot that you found or uh yeah three okay just write that down all your competitors are like oh there's some work we don't have to do oh no actually the number is different based on how long the phrase is okay so the longer the phrase the more distance you actually want to give it because then you can account for multiple mistakes um and the shorter the word the tendency to there's a tendency to be less mistakes and also you want less false positives when you have shorter words so you tend to reduce that number when you go smaller so um there's a certain threshold where we we go to three but below a certain threshold we have we have a set of like two so i see so i think in the video you sent over we watched you were indexing wikipedia on a raspberry pi with search craft and you were giving examples of quick lookups with misspellings i think einstein might have been one you just spelled that singular word wrong uh and that would be a case where you would is it three characters that you're searching like what's the three applying to when it comes to misspellings yeah like up to like positions away from the original term essentially yeah gotcha where like you could have how many mistakes can you accommodate before you're making too many false matches right okay so yeah which is why like smaller words you'd want to do less because it's going to match to too many things yeah exactly and then for longer phrases you could actually match multiple misspellings in different words throughout the phrase and it would still that's correct okay and that is interesting and the way a lot of search engines work is like you have your word and each word is considered a token right which people are becoming more and more familiar with now because of LLMs, tokens, and utokines, right?

32:53So when you're doing the ingestion process, you pull in your data, you chunk it out, usually by like words, white space or whatever, into tokens. And then those tokens are used and stored in Colinar format, which gives you fast lookup times against that token. And as you said in the opening, I think it is a good time to be a search engine because everybody is now realizing that maybe our search isn't as good as it could be, you know, because it's kind of a thing that, and I've done this, I've done it twice. I rolled it out on Postgres and forgot about it. And then we switched over to TypeSense and we rolled it out on TypeSense and kind of just forgot about it.

33:31And it just is what it is. But then here comes ChatGPT and friends and everyone's starting to think like, well, couldn't you just talk to my website and it will just show you what you want to find? and it makes that interface that chat interface where they these models have you know seemingly endless knowledge of the entire world except for my website is so compelling that everyone's like ah maybe our search is not good maybe our search needs to be good now and so then you kind of poke your head up and start looking around and see what people are doing i was i'm on the type sense mailing list and they just ship some features with llms doing something i can't remember what they're doing i think they're actually like kind of writing your query for you like you put in plain text and they turn it into their query language with an llm which seems like a cool thing and i'm sure algolia is doing stuff i'm sure elastics like everyone's like okay we need to revamp all these search engines so what are you guys doing i'm sure i assume there's some kind of pivot involved with the current fast-pacing uh world of llms yeah i mean you're you're totally not alone in the fact that you're like you had set up search in a couple you know iterations and then kind of like forgot about it and left it right like every literally every you're supposed to every company did that right yeah like okay yeah i mean it checked the box we have the little magnifying glass in the corner all right let's we're done find stuff sometimes you can't find stuff like this would be better if you and i'm like yeah it would be but most of it's like most of the time it's like you can find stuff kind of right right like it's not great but it's there and we check the box.

35:12And a lot of companies were like that. They're like, we check the box, we have search kind of works. But we don't really care because people can just search Google and Google will give us the traffic to our article that we just published yesterday. Right. And that's what everyone did. Literally, they're like, Google will just give us traffic. So our search on our own site doesn't have to be great. But now because of like Google's doing like the whole AI summary thing they're like diverting traffic to sites less and less right so like this is not just opinion it's like companies are saying across the board where their traffic is just plummeting because google is not sending them there anymore they're sending it they're either keeping them on google or sending them to google-owned properties kind of thing which is kind of nefarious i i think um but their their traffic's dropping off so they're like okay well well, I guess we have to make our search better now.

36:03And what are people doing? They're like, well, there's this whole chat GPT kind of experience that's out there. Let's look at that, right? So companies like Algolia, Elastic, and us, we're like, okay, we have really powerful keyword search, but people are wanting to do this natural language thing. So let's do that. So we do have, as you said, like a pivot in that we have an AI offering that is coming out in our SDK very shortly. But the thing about these, like, I would call them first-generation kind of AI approaches is that a lot of them, while they've really focused on, like, the natural language semantic reasoning layer with the language models, they haven't put a lot of thought into the information discovery aspect of it, that middle layer that says, okay, like, yes, the LLM can translate the user's intention, but like turning that, like powering that with something that can actually serve you up accurate information, accurate and relevant information.

37:12And recent information is like a problem still. So like you'll use these chat interfaces and they'll give you like, you've seen it, like either non-existent information or wrong information. Right. And that's like an information discovery problem in that they aren't using the proper tooling underneath. Kind of like if you're going to do a keyword-based search, you typically don't want a wild card against a SQL database because it's not going to be what you need it to be. You need other relevancy tuning features in there. So I think companies will get there. I think the advantage that search-based companies have is they understand relevancy already, and they've solved relevancy now.

37:54And then, so we let the language models kind of handle like the semantic reasoning part and the language models convert that user intention into the proper kind of like machine API queries. Right. So like, let's turn those, that natural language set, like question into machine API queries and then query the right kind of system to get that information back. so I think you'll see more of that kind of like second generation approach where they're hooking it up to the right tools and then that experience is going to get better is it pretty much is fuzzy search and intention similar because it feels like it's similar to me but I don't think it is an implementation fuzzy search so there's a couple of different things there's fuzzy search which is like let me not have to put the complete thing I can do partial search and it completes it for you.

38:47There's also kind of related, but different is typo tolerance where like I misspell it and it knows what I'm doing. But then there's also things like synonyms, which is like I'm looking for shoes, but I should also match on like sandals or loafers or whatever. Yeah. Right, where the words are completely different, but they could have the same intention. When someone searches or types into a search box with search graph, Have you been Searchcraft? I want to be more clear on my, I almost said Searchgraph. That's kind of cool too. Searchcraft, if they're using that, are you doing, or is it, are you doing any intention analysis in addition to like fuzzy search and stuff like that?

39:31Or is that where the LLM comes in, what Jerry was saying? Yeah, so like our semantic layer is language model based. So it kind of converts that into like, okay, what, I don't know how much of the secret sauce I should give out, but it does convert it. into like what machines can understand. You already gave three. What else can you do? Oh no, just give it away, give it all away. No, it converts that into like what the machines expect, right? But there's also like, if you are using like a search engine underneath your semantic layer, you do things with relevancy tuning that improves, right? So like I mentioned before, like the synonyms, right?

40:09So knowing that shoes matches to sandals or other things like that. and that's just handled for you on that retrieval layer. So has this required you to go back and get more money or slap AI on your pitch deck and raise the second round or what are you doing there? We are doing with AI. Now with AI. So when we started out, AI like wasn't as, like it was just kind of nascent in starting out and like it wasn't really a thing. We started SearchCraft May of 2024. So we are kind of working on it on the side up until then. 2021. That's when we started working on it. Yeah. So we started building it in 2021, but like we had to build the core and then it got to the point where like, yeah, this is really good.

40:52Like maybe we shut down our consulting agency and build our own product company. And that's, that's what we did May 2024. And then at that point, my co-founder and I had to learn, okay, well, like how do we pitch and get investment? I had never done it before. Right. So we were fortunate enough that we were able to do our first raise with all angel investors, individual angels. And we were able to close that rather quickly, which allowed us to shut down our agency company and start full-time or going full steam on SearchCraft. So that meant turning essentially our proof of concept core and then making a company around that, building customer dashboards and billing and all of our like sdks that we made for the front end and all that is that company chalk is that what that is chalk yeah yeah yeah chalk was my previous company we uh so we consulted for startups that was our kind of our bread and butter you know how i found that i stopped you you kicked me out your company stock i uh i was just thinking i was thinking search craft is really cool in a no no offense to our friends over type sense but i think search craft says search because it literally has the word search in it and you couldn't really get a better mark than that like search craft is pretty strong and i was thinking geez i sure hope don has the trademark for that because that's a pretty cool mark and sure enough somebody has it you know who has it on chalk chalk does yeah yeah well we registered originally under chalk um but uh all good still at some point it's going to transfer it to my co-founder and i well it doesn't matter who owns it as long as somebody does that somebody owns it yeah yeah you're not gonna it's not well i guess it's ip itself you sold the company but for now don't worry about transferring it eventually it should get transferred though but yeah we um we made sure before we like were completely sold sold on the name that we picked something that someone didn't have yeah i would hate to like build up branding and register all the domains and then you're like oh Oh, wait, somebody else has it.

42:57Yeah, it's like no reason. It's like racing. And you're like, you're at the wrong track. Dude, I just won the race. Yeah, because no one's here. Okay. No one's here. You're wrong. Go over there. Not a one-to-one, but pretty similar. You know, wrong name, right business kind of thing. Well, this is the second craft that we've come across recently because we had a show with Andreas Moller of Nordcraft. previously Toddle renamed to Nordcraft you are Searchcraft of course there is Minecraft as we mentioned on that show we failed to mention there is Starcraft there is Warcraft there's all these different games that were crafts but now we have like startups that are crafts this could be like the new Web 2.0 naming where they just take the letter out take the vowel out of the end like Flickr and Fiverr yeah Fiverr did take it out or the if the if i's shopify friend of friendify that's right anyways craft could be like the new bulls or the dot ly oh yeah fastly bitly bitly i don't know cluely that's a new one actually shopifyly anyways a new trend here is there is there a craft tld i'm guessing no there's hey that could be your next business you know someone's gonna buy it crafty all these yeah no it's it's kind of funny because i like have like a little like hobby thing i was thinking about coding in a weekend and like i'm just like what should i name it and so i like it was gonna be like ending in craft because of course anything else yeah i get the trend going somehow just keep coming out with crafts but the name very much like the name the good thing about like once we settled on what the name would be like the branding was like real easy to come up with after that yeah i think you did a good job on that search craft is i think strong obviously because it's got the word search in it i mean you don't have to explain it there's no explanation necessary what we do is in the name there you go it's a good name right there we were we were talking about investment and yes to go back go back to your experience the first time fundraising yeah so all angels like what was your what were your steps what was your plan what did you actually do okay we'll start the the beginning of the the process so when you go to do a raise like you have this idea for a company but then you have to kind of like okay how do i translate that idea to business people right like and that's a pitch deck right you write a pitch deck i had never written pitch deck before so uh my co-founder and i um we kind of talked to a few different mentors in the space we had worked with like startups as our clients so we actually knew a couple people who were like written pitch decks for those startups so we're like can you like show us how to do this and they kind of told us like okay well you kind of have to tell this whole story about like this is the change in the world and then these are the winners and losers and this is how you can be a winner by taking advantage of this change that's happened in the world and that's kind of like how you sell it to investors right interesting there's a formula right and if you don't follow the formula they kind of instantly throw you out of the room kind of thing and we didn't know that either is that the exact formula you just shared or can you can you be more precise less vague i mean that's kind of like the general formula like there's multiple steps in along that way and then like as far as like they even get as far as like on slide three this is what you need to show on slide seven this is what you need to show like there there's a path like a formula to it and um the vcs know it but people first-time founders don't necessarily know this yeah i want to learn this this framework i want to learn it yeah so uh yeah so we had a guy who kind of showed us the ropes on that um he didn't know our product well enough to kind of come up with the messaging but he kind of gave us the basic structure of like how we should form the pitch and then we kind of took that and ran with it and refined.

47:07We've probably written like a dozen versions of the pitch deck by now, maybe even more. Like we, it's one of those things you kind of keep tweaking, especially as the industry changes and you're doing a different, like kind of, if you're a different round of racing, like it changes. They very much want to know more about like, you know, revenue projections, the further along you go and things like that. It's just giving me a really good idea for a series or something. it's like startup pitch decks like five years later or however like once they've arrived and you go with the founder and you go back to their original pitch deck and they walk you through it and then you can talk about all the stuff that's changed or what were they right where are they wrong what have they changed since then that would be very interesting we almost did that with beyond lou last time i talked to him because he was talking about it was either him or quinn i think it was quinn actually now i think about quinn slack his co-founder they were sharing how their vision has been similar, but it's obviously changed over like as progress has happened and technology has changed.

48:10But like their mission, their core mission with developers has been the same. He's like, if you go back to the original pitch deck, and I think he was supposed to actually link it up to me. So I should go back to those show notes or go back to them and get a link. But that was the closest we've gotten to my knowledge of like surfacing the original pitch deck and going through it. So you've tweaked yours a dozen times so far. Yeah, at least. I mean, I have a couple years into it, right? Like two years in. Yeah. I mean, I guess we're technically a year and like a couple months out. Yeah. Less than two.

48:38And you added AI. You sprinkled AI into all of every slide. Okay. So like the, from the individual angels, like they were excited by what we're building and sound the potential and they wrote the check and we got in there. But subsequent talks we've had with like more institutional offices, like venture funds and, and pre-seed funds and things like that they um are pretty like ai has sucked up all the oxygen in the room like that's all the dollars are going you have to have ai like unfortunately like it feels like it's like a like a cliche thing but it is the truth like they're only investing in ai right now and i don't know if that pressure is coming from like the lending partners wanting to be in that space or it's just kind of like the thesis supports investing in ai because ai is such a transformative kind of shift right now.

49:28Right. What it is, is the true overlords of this earth are forcing money to flow to AI to do this thing. Like, that's what's happening. It's the Illuminati. Okay. Oh, the Illuminati. Oh, that's right. You know what? Somebody is puppeting and orchestrating with the money. I was actually thinking about this the other day in that, like, so there's this huge fear as developers that AI is going to replace all of us. And I mean, not even just developers, like you're a writer, you're an illustrator, like all of these professions are getting supposedly replaced by AI, right? So is that just a transfer of wealth from all of these people to like, okay, a few handful of companies that control the models?

50:14Like Anthropic and OpenAI, they're going to kind of like all that funnel to them. Just through the roof, right? The valuations on these companies are astronomical, and they only actually make sense if that plays out. If that plays out, they make total sense. Invest right now. But if not, then obviously a lot of people are burning up cash making that bet. That's pretty much the bet they're making, isn't it? I think that is the bet, but to be existential about it, is that good for humanity? Instead of paying people, companies are paying the AI companies instead of people. But eventually they're probably going to pay the same amount.

50:52Like, I think a lot of that, like GPU usage is being subsidized by VC capital right now. And like the true cost is not there and you're hooking people in to make them like dependent on these systems. And then you're going to raise the price once the VC money goes away. I mean, that's always been in the play, right? It's like you just subsidize the price of the thing until you have critical scale and mass to the point where you don't have to do that anymore. and then you turn on the money-making machine and everybody who is subservient to that is less better off than they were prior you know we're all like wait a second this doesn't actually scale now it's more expensive you know that's what happened that's what uber did right that's what the uber was what i was thinking of when i was talking about that yeah yeah uber did that like they killed the cab industry now they control like transportation right um the Well, there's still cabbies out there.

51:47You know, they're still doing their thing. And they have had to change the way they work. And so I feel like it was an industry that was rife for disruption because it was entrenched and I think unionized as well. I'm not sure about that. It might depend on this. Definitely in the bigger cities. Yeah, the bigger cities. That was a big thing for New York. And just like not a good, I mean, it wasn't a good customer service. And now the cabbies, I think, like I've taken a cab because they've been like hustling for my money. You know, you're waiting for your Uber. And they pull up and they're like, what are you going to, I'll take you right now for five bucks less.

52:19Like, let's roll, you know? Oh, yeah. I got no problem with that. So, like, that's been good, but there's obviously been fallout as well. So, it is nuanced. It's not all good, not all bad. But certainly Uber did that, right? They just made it so cheap to take an Uber for so long. It's just not cheap anymore. Airbnb was the same way. It used to be cheaper than hotels. Now it's like, I go to a hotel. I prefer a hotel to an Airbnb at this point. I'm a big fan of like especially if you go to places like New York they have like really nice like boutique hotels that you can go and yeah totally like oh I don't have to worry about getting cleaning up after myself for the Airbnb host so I don't get you know cleaning fees there's always one thing that goes wrong with an Airbnb in my experience where it's like I had to go out of my way for whatever it was like the the key didn't work or the instructions weren't clear on the trash cans or this thing and you're like talking to the host and they're not what it was originally which is like just another friendly person in society like they're a small business trying to do a bunch of these things and whereas the hotels pretty much have that all figured out and like you can just go to the front desk and there's just like a professionalism to it and then and then it also costs less now because of the cleaning fees so it's been interesting to watch that because when Airbnb first came out, I was all about it.

53:38Like I was saving money. I was in the city, like not at a, what do you call it? A tourist trap or something. Yeah. Real people helping them out. Like living in someone's house was cool. And now I'm just like, nah, give me the hotel. So it's been a weird full circle for me. I mean, I think if you're still traveling with like a big group, it's pretty good. For sure. Like you got like five people or whatever. like it's going to be better than than separate hotel rooms probably yes if you have to get two hotel rooms one airbnb will almost always be cheaper but that i think is the exception to most travel which is yeah two people or whatever anyways where else has this happened though an industry where it's subsidized like i was thinking like honestly i was thinking about cars versus like horse-drawn buggies like was there any subsidization happening then because like Like it's a step change or a step function change in how humanity works to go from a horse-drawn buggy or cart to a car.

54:37And clearly we have not gone back, you know. So I'm wondering like where else a big step change happened in humanity where it was subsidized for a bit. Because that's what AI is. It's a big step change, right? Was gasoline cheaper than like feeding your horse or whatever or stabling your horse back then? Yeah. I'm not a historian. I don't know the answer to these questions. I thought automobiles when they came out I thought that was kind of like a rich wealthy person thing I don't know I'm sure it started there and worked its way down I mean that's how most new technologies work right it was probably like the you know what do you call that what's the Apple thing with the goggles what's that thing called again Vision Pro that has not trickled down yet yeah it's like that their sales are trickling out well the smartphone's a better example like when the iPhone first revolutionized phones which it really did yeah and internet communicators it was a rich person's device and now there are phones all the way down the pricing ladder to free with a plan for everybody you know pretty much for the populated civilized world that nobody could afford that original iphone if you didn't have all kinds of disposable income so i do think new technology generally comes out expensive works as worth the price down but the subsidies i think is where we were kind of keying in on don now we're it's the reverse though like the at least from the software and services side where it comes in cheap and then it's free yeah quote-unquote free they get you dependent on it and then yeah it's more of a drug dealer model which i've said before you know the first hit's always free and it is addictive so i mean it really does work to a certain extent where it's hard to go back to life before it isn't adam you said that a few times where it's like would you go back no i mean it's especially to like a car versus i'm talking about with your chat gpt relationship no don't take my chat gpt away from me please see yeah no it is a good product yeah i really couldn't imagine life you know this word calculator is just too good you know it really is i couldn't imagine life I couldn't imagine running businesses without it.

56:56You know, like you have to have this thing. Like it's, you have to think so much faster and it thinks bigger and faster for you. Now you're still evaluating the output. It's just helping you get to that result, the data faster, you know, or even running scenarios against a business model, for example, like destroy this business model for me. And it uncovers these new ways that you will die and either, either draws new fear in or it's like, makes you more invincible because you feel like, oh, I can sustain all that. You know? I mean, it's definitely a faster way to work for certain things. Like I don't go to Stack Overflow anymore.

57:32I go to my LLM and ask you the question. Well, Stack Overflow is dying. I mean, what do you guys stand on the whole thing about like trading data? Like was it fair that these models were trained on this information that was out there? Like what are your opinions on that? Let me think about it for a second because I don't want to be wrong about this. I think my gut reaction is it's public on the internet. so it's the internet's information but then that doesn't feel good because like sometimes you put personal ideas that belong to me adam on public does that become somebody else's idea or somebody else's thing it's really hard to answer that without specifics but i think stack overflow in particular it's like such user-generated content so it's kind of like it's owned by the world in a way and you know stack overflow enabled the ability to have this new place to go to this new website to go to and and built all the form and function around the usefulness of exchanging questions for answers essentially and hopefully this dialogue back and forth of which one's right and which one's the the one you should follow hard to tell you know really is hard to tell i think baseline judgment without more specifics will be like if it's public on the internet then sure because if I went and studied like a thousand machines could, I could learn the same information.

58:56It's just done at scale, you know, and in perpetuity, I suppose. Hard question to answer, really. It's a lot of nuance, a lot of facets to that one. But as you're learning, to take it back to like you could learn and learn all these things and read all this information and absorb it, you're giving an exchange for learning you're giving them an ad impression right so the people querying through like the language model interfaces they're they're bypassing that right like they're getting that from the language model versus giving a impression to the originator of the information yeah well there you go so that yeah but adam could read a book and he could learn from that book and then he can come tell me what he learned and i don't think there's anything immoral about that at all and so i think when it comes down to infringement on somebody else's creation i think it should be focused on the outputs more than the inputs and if you are directly copying and redistributing somebody's creative work as an output i think you're outside of bounds.

1:00:08That's what I was trying to say right there. That's it. That's what I was trying to say. You can't copy my sentence. I just said it. It was good though. I like, I like the way you frame that because it's all about outputs versus the inputs. You know, we can't, we can all learn things about what comes out, not what goes in. But the inputs do need to follow the laws, I think, and the copyright. And you can, you can have things that are consumable and things that are not consumable. So if you're consuming things that are not, you know, legally consumable, then you're also outside of bounds. I mean, Anthropic just won a big court case with what they did with these books, scanning books and, and bringing them into their model.

1:00:44Yeah. But I think that was more, they won because the plaintiffs didn't form their argument correctly. I think the judge was like, also said like, Hey, if you had actually argued it this way, you could have won, but you didn't. So I can't find for you. How would they have changed the argument to be a winner? Do you know? Oh, I don't remember. I, I just remember that was, that was the judge's opinion. And it's like he was kind of prodding them along. Like if you had came at this court case from this angle, like you would have had a case, but you didn't. So I can't find it. But like he kind of like showed them the way.

1:01:17Even Google's old book project, I can't remember what that was, called Google Books probably. And their entire purpose was like, you know, index the world's information and make it freely available or whatever their whole mission is. that project would just take books that were either public domain or purchased or whatever they would acquire through libraries and they would scan them into a digital form and they OCR them. And that was a huge court case, uh, years ago that I think they ended up winning. I can't remember the actual end result, but I think it's a similar construct, I guess it's similar in pattern to that where it's like you are an owner of a thing can't you not digitize that thing that you own like you purchased it for instance now if you're the other thing with anthropic was there was and meta is in a similar situation there was two kind of angles because one was like do you have the right to actually do this with purchased copyright or purchased ip versus they were downloading downloading pirated versions off the internet and scanning those in and i think that That was the clearly illegal stuff.

1:02:24And I could see why. But I mean, libraries have been formed around this entire thing. Like you get copyright in America at least, and then you agree to have it publicly available as well, then you can sell it. So it is. There are specific rules around like distribution though, right? Like you have the right to purchase and kind of like absorb that content, right? But like distribution is a different thing. That's why I said output should be what we should focus on because that is distribution, isn't it? Like if I'm redistributing 311's new album, do they make albums anymore? I don't know. That was the first autocomplete that came to my brain there.

1:03:01And I'm just like verbatim throwing it out. That's clearly illegal distribution of their copyright. However, if I'm using their album as inspiration for something new, quote unquote new, and that's where it gets really technically difficult to describe whether or not it's new or not. like some of these new video channels on youtube you guys been watching this where there's new youtube channels that are just taking over the charts and it's completely ai generated everything there's a specific like 80s synthwave band that doesn't exist that's like has 310 million subscribers or something now in the last six months probably a lot of bots but people are liking this music and it's like straight up using suno or whatever it is to create the music and it is i went and listened to their song and watched their video and it's like this is new music it is however clearly derivative of old music and like for me that's i'm kind of okay with that like this is new but what vanilla ice did to what's their names you know with the hi-hat that was clearly copyright infringement because he was just redistributing the same exact sound the music industry kind of like dealt with this right Like if you want to take the analogy from, from like the hip hop community and like sampling, right?

1:04:15Like early hip hop was all based around sampling. Beastie was all samples. So yeah, it was all, it was all, it was all sampling. So like they found like, okay, you can do that, but the caveat is you have to get like permission. They have to get a license to do it. Right. If you don't get the license, that's when you get in trouble. Yeah. And there's a lot of that stuff getting set up. I mean, Reddit is, has licensing agreements. I think Stack Overflow does well with a few companies. But then there's also people that just aren't going to get that license and they're going to just do what they do.

1:04:50And it's kind of a Pandora's box situation. Who's this ALSEP person in this argument? Who's ALSEP? Do you know? Was that the judge? Which argument are you back on the Anthropics? Anthropics still yet, yeah. Trying to get that figured out. out well i'm reading the cnbc and i want to pull a quote here but it says alsup wrote a lsup is the person's last name i'm trying to figure oh it's the judge sorry u.s district judge william alsup said this okay so what he said was uh this is the judge saying it back to whomever i'm not sure if this is cnbc the reporter the journalist whoever but this is the quote the purpose and character of using copyrighted works to train llms to make or sorry to generate new text was quintessentially transformative house approach like any reading like any reader aspiring to be a writer so the way he framed it was like they trained them on the works so they can learn how to write not to not to redistribute what it read essentially you know for lack of better terms like it read it consumed it oh this is how you write right it was like frameworks you know sentence formation story arc you know stuff like that versus let's literally redistribute there but i think right but you can tell them like give me a story in the voice and tone of mark twain and it will do that now he's long gone and dead but like that's and that's kind of what a lot of these music is like make it sound like the beastie boys you know only not like can't you do that with a friend though couldn't i say hey jared do me a favor i know you're really good at rapping right you're pretty good at rapping i mean would you mind rapping in the form of vanilla ice you know all right then you might you might yeah exactly you might use some of the hook phrases and that's what the llm is doing back to you all it is is giving you an output of similarity i know it's it's it's tough because it's like but then when you say give me this picture in studio ghibli style it's like exactly the style of studio ghibli right yeah in that case there that is that's the same concept i mean couldn't you do that to a to an artist though couldn't you take anybody who can draw and draw well who maybe train themselves on that style and ask them to do the same thing i suppose it's about output and potential usage because like that was a whole industry wasn't it wasn't duplicates um what is it called copy uh people that would actually clone artist work for yeah to first for resale at a cheaper price because they were good enough painters that they could paint it like Picasso without actually being Picasso.

1:07:30That's been a thing for centuries, right? Yeah, I just can't think of what it's called. Yeah, it's a console clone or whatever. Yeah, exactly. So, yeah, I mean, but was that moral and legal? I don't know. I mean, you're stealing his ability to sell. You still had to have an ability to do it? I don't know. Yeah, I mean, it took some skill. Just like me, my vanilla ice wrap is pretty good, you know. Yeah, all right, stop. You had it down. You were pretty good. like there's a difference between like okay i'm a musician i learn how to play instruments and i'm in a cover band covering vanilla ice and right right

1:08:14well friends it's time to build the future of multi-agent software and you can do so with agency. That's A-G-N-T-C-Y. The agency is an open source collective building the internet of agents. It's a collaboration layer where AI agents can discover, connect, and work across frameworks. For developers, this means standardized agent discovery tools, seamless protocols for inter-agent communication, and modular components to compose and scale multi-agent workflows. You can join Crew, LineChain, Lambda Index, BrowserBase, Cisco and dozens more. The agency is dropping code, specs, and services with no strings attached.

1:08:57Build with other engineers who care about high quality multi-agent software. Visit agency.org. That's A-G-N-T-C-Y dot org and add your support. Once again, agency.org. A-G-N-T-C-Y dot org.

1:09:17yeah there's something about the scale and the speed which just changes the calculus it does i agree with that it makes it more i don't know infringy just seem like you can just because infringy that's my like that professional word of the day i like it it's actually pretty good infringy well done you opened up a can you opened up a can of worms here my friend and uh i don't know how i get back on the right track i guess my request my my response be what do you think about that about uh the the whole like training on copyrighted material the infringy the infringiness of the infringiness i mean a lot of people try to argue it's like oh yeah i'm just like teaching myself to it's like teaching an entity how to read or whatever right um and i think there's a difference between like using the information to learn how to read and write versus i'm going to give you something that is clearly derived from this thing right so whether or not they're ever able going to be able to determine like okay this is like 30 mark twain so you owe Mark Twain's estate 30 % of your creation's value.

1:10:34I don't know. With the previous iterations of this remixing with music, they figured out a mechanism of licensing and royalties and things like that. Maybe that's something down the road. Maybe if you're training on data sets as a whole, you pay into a fund that compensates people who are creating. My fear is that like the people with these skills are, these skills are not going to be in demand. So like there's no impetus for people to become like an illustrator or a writer or a developer. Like, and so what happens 20 years from now when there's no one coming in into the ecosystem anymore doing these things, right?

1:11:19We just talked about that with Thorsten Ball in light of open source and our new agent at Coders where it becomes almost free to just generate some code. Why open source maintain and share libraries when a coding agent can just create me a new auth library very quickly? One-offs for everything. Why don't you just have a bespoke everything? And what does that look like in 20 years? because all of a sudden I'm not incentivized to go find open source, download open source, contribute to open source, open source my own code when it's just disposable anyways. All code is disposable. And then we lose public code eventually because this is just like a potential truth and not necessarily that's going to happen.

1:12:09But yeah, it's similar with illustrators and creatives of all kinds. When there's no longer incentives to do that. Although I think when it comes to music and art and even writing, people are going to continue to do it because they love it. And maybe that's the same answer with code and open source. You know, people didn't originally make music to make money. They make music and they want to make money so they can make more music. You know what I'm saying? And then some people found ways of getting real rich off that. And of course, the celebrity culture and there's all other reasons why people make music.

1:12:41But a lot of it's just because people love making music. and so I don't think humans are ever going to stop making music even if the robots are better at it and maybe that's true of illustration and photography and copywriting I don't know novel no novels but it's not necessarily it doesn't feel good because those people don't have a way of ever making money whereas they used to but I think they're going to keep creating I think fundamentally we do we are creative well it'll be one of those things though where it's like it's never at the scale that it was or it's never something for like creating commercial value like i'm never going to be an illustrator and make money off it i just do it for fun potentially i think that i think that's a real potential it's that's where i think your position jerry say input output but then i would also add usage because i feel like it's not actually deployed to the world in a refactored format from input to the LM output as request from the prompt, you know, reviewed by the human and used deployed into the world.

1:13:48I feel like that last mile is potentially where the, where the hook is because if I end up using copyrighted material, knowing it was or studio Ghibli or Ghibli or however you say it. Um, if I'm breaking the copyright law by trying to act like somebody else, force the LLM to help me be similar to somebody else's creative works, it's up to me to have the moral and legal boundaries of just following that. I would say like the LLM is just, is just trying to give me what I want. and it's my usage that takes me to the legal side of it. That's how I feel, I guess. If I wanted to recreate something similar to Mark Twain, if I'm the kind of person who would do that and would put it out there like it's Adam, but it's also Mark Twain, well, then that's kind of my fault, right?

1:14:45Not the LLMs. Well, I'm not blaming any computer programs. I'm just talking about how we use them. So at the end of the day, Anthropic, the business, is getting sued, not Claude. personally so i agree with you on that one for sure yeah but it's still complicated and we can't simply rely on the morality of other people like we have to actually institutionalize rules and consequences for rules and i think it's going to be super muddy over the next 10 years trying to figure this out and i think there's gonna be winners and losers and hopefully the losers aren't the people who are really just good at drawing and making music and you know writing poetry right on but they might be we might be to take it back to like software development um okay so the the selling point is like okay seniors can do more they can be 10x productive right like now i have cogen i can't i have agents and i have you know conversational you know uh ai and my ide and all this and so i can produce more code so like maybe i don't need to hire juniors anymore but if there's no juniors being hired then eventually who's going to replace all these senior level people yeah i think that's where it works i don't know yeah maybe so the advancements of the agents may transcend the need for a future true junior or what we call a junior human being developer i think the answer to that to adam's point is like where does the technology plateau if at all like if we're getting to the plateau i don't think any engineers really unless it's ever the really sucky ones are going to go anywhere but if it continues to advance at the rate of change it has and i'm just been very impressed with the most recent round of models where i was unimpressed previously same thing with the image gen stuff like it's way more compelling now than it was 18 months ago 18 months ago and if it stays where it is i think we're not going to have these existential problems that we're discussing, but if it continues to change at the rate that some of the capitalists who are highly incentivized to say it's changing, it is changing, such as Anthropics CEO who says, you know, 18 months from now, there'll be a reckoning.

1:17:01And I know because I can see the future because the future is inside of our walls, so to speak. And if he's right about that, then I think everything does change and we don't have answers and maybe we don't need juniors or seniors um but maybe it's going to plateau right kind of where it is i don't know the answer to that i mean i think there's some exaggeration on both sides the model companies they have to say that because they've received money to make that happen and they're currently raising more money yeah i mean billions and billions and trillions of dollars right like so they they have to say that but i don't think that's necessarily accurate we'll see um but on the other side will it get to the point where it won't need supervision i don't maybe it's plateaued maybe it hasn't right if it if it hasn't plateaued then maybe there is no need for developers right i mean it's getting pretty good and i could imagine 18 months from now me saying things like i don't even look at the code ever anymore i'm just telling the thing what to write at small portions and building up systems with my pros and it just works i'm not i'm saying that with like personal scripts and stuff right now things that don't leave my machine like i told adam recently like yeah i'm just letting the thing write it and then i use it and if it works it works but that's not that's not leaving the network you know i'm saying i mean you can still say that with code that like okay i hire like a code school bootcamp.

1:18:35Sure. Right. And they write, they know how to write react, but they don't necessarily know type strip very well. Sure. And they write code that works, but it's not written in a way that like is great. Right. From someone looking at it from like a senior developer perspective, you're like, Oh, this code is not good, but it does do what the like Jira ticket said. Right. Right. So they didn't write it in 24 seconds. They didn't. Yeah. That's the difference. And I can say nah not like that like this and it's 30 seconds later and i have my next iteration and so i think that person they're dead to me oh my gosh they're dead you know what i mean in the functional sense of i'm not going to go to that person already so yeah i mean and then i think that a lot of these boot camps not to like crap on them but like they sprung up when there was a really high demand for developers there There was not enough developers to fill seats, and so they churned people out who maybe shouldn't have been writing code to begin with.

1:19:36We're just looking for it from a paycheck perspective and not like the craft of writing code perspective. And the not-as-skilled people are always going to get filtered out no matter what. I don't think it's going to kill software, that's for sure. I don't think it's going to kill a software developer. I just wonder if really the end result of all this is, like you said, I don't get the code anymore. And maybe that's true for certain things, like you said, for scripts. It's already true. Yeah. I can't disagree that since that conversation, I've had this fear. I'm not sure I want to call it a fear.

1:20:13It's like this, I can't let it go. I know something that a lot of people know, and I have concern. Like, my friends have no idea about that conversation. My friends have no really idea about the deep conversation we had with Steve or others that scared the crap out of us, basically. Like that conversation with Steve Yege was probably the one that scared me the most. When I think that was that conversation, you said, what do we do about open source? And the moment you said that, I was like, yeah, what do we do about open source? And I kept thinking about the ability to generate quickly. Now, does it kill all open source?

1:20:43No. I think it will diminish the need for such widespread open source. Like, do we really need to widespread open source larger frameworks? Yeah, probably. But at some point we may care less about the code because we've got quality assurance checks that balance out and they are always true. And so we look at the code less. I've really been a little concerned about open source since then. Does that lead to the development of like new programming languages kind of going away? Like if what we have out there works well enough, is there no impetus to like make another language? I think it comes down to who's writing it and who cares, right?

1:21:25Like you just said good code. And I was thinking when you said good code, I was like, well, just because I think you said a React developer wrote some TypeScript, is that right? Was that the analogy? Or was it reversed? Sure. Yeah, they know React, but they don't know TypeScript very well. Right. They know React, but they don't know TypeScript very well. But it works. So is that good code? What do we call good code? something that a human can read and it's well formed and understandable or something that actually gets the job done and works well which one of those is good code well it has to it has to work correctly first for sure and then it has to be correct or maintainable i mean i think it's its ability to change easily right is one of the prime characteristics of good code and that's why i said i think the calculus changes if these things continue to get better at a rapid pace because maintenance might not be the same concern as it always has been.

1:22:24I've always written software to be maintainable because any software product that's useful is going to continue to exist and change. It's never finished. And if you can't change it, then you're atrophied and you're stuck. And so that's why I say slow down to go faster, get it right in the first place. You go faster over the longterm. However, that calculus changes if now we can generate quickly small amounts of code that are good enough to replace, not refactor, replace, rewrite. It used to be very expensive and tool hardy to do that, but maybe if it becomes very cheap and quick and you can get 10 versions, which one's the best?

1:23:01Plug it in. Oh, it's not the best anymore. Plug in a new one. Just keep plugging in, you know? Then all of a sudden, just the way that we think about building software does change. And a lot of my old tenants, like a lot of things I've been saying for 30 years, that's not true. 15 years they go away they're wrong now and that's the part that scares me the most maybe I go away should I keep using the word scare I'm not really fearful at night maybe I'm dead to them that's the part that really gives me concern I would think then because at that point it's just a whole different world when the calculus changes like when you thought truths for 30 or 20 years were true and suddenly a lot of these like foundational truths are now up for grabs for change, I think a lot of it changes really.

1:23:47You know, how we think, like you said, building software has been about sustainable software and that's been good software, sustainable software. But if you don't have to maintain it and you trust the thing that generated it and so long as it knows the language and, you know, what is good code at that point? What is the code we need to have to get the job done? This is why everybody wants AI on their pitch decks. That's why. The implications are massive. And anytime you have this much up in the air, you know, people are looking to catch some of that for themselves. And so they want the AI solution because it's changing stuff and I want to be involved.

1:24:24You know, I want to be on the upside of this potential future, not on the downside. Because the downside is super scary, right? Like a lot of people's skills become invaluable. Actually means valuable. That's a weird word, right? Not valuable. I always thought that was weird. Invaluable means valuable. but uh non-valuable perhaps yeah and invaluable is like what the the valuation can't be measured yeah exactly but it does sound like it just means very valuable but usually it means not anyways but what about unvaluable does that work yeah they become unvaluable no it's not but we're making up words today no i there's a lot of fear right and i think making decisions on fears like not always a great thing sure so companies are embracing ai regardless if they Just in case.

1:25:09Yeah. They're like, we have to have a story for AI. So let's do the thing. We have to have a mobile app. Yeah, totally. It's like the mobile trend. Like, is it the right decision for the long term? That's like, you have to ask yourself that question as a business owner. Like, does it make sense? Or are we just doing it to follow the trend? So what does search craft look like over the next 12, 18, 24 months? Like you're raising more money. You're switching a little bit of things. Yeah, we're raising some more money. and we're also, you know, we are rolling out some of our AI initiatives because we have to.

1:25:45Yes. Some more. If you want money, you got to have AI. That's right. But no, it's because people are asking for it because it's what people are excited about, right? Like they want a search experience that has some AI and it's at least that natural language and semantic reasoning. And so we're building for what people are requesting. So that's what we're doing. As far as, you know, the next 12 months, it's like we're looking to kind of work with more customers. And we're also looking to expand into more platforms. So like we rolled out with like our very first initial like handful of like front-end SDKs.

1:26:24So of course we did a JavaScript one, we did a React one, we did a Vue one, but like we want to get into other places where developers are at as well. Like, you know, we want to get into Laravel, get into Ruby on Rails and all of these other areas as well. A lot of easy wins there are like ways you can scale horizontally across different ecosystems and just provide the service to more people that way. Yeah, so there's like language platforms. There's also like, you know, direct integrations with other systems because at the end of the day, like search is really, it's a mechanism of information discovery and information discovery is a way of delivering the most relevant and accurate information to a system.

1:27:04So we're kind of essentially like a data pipe for any kind of platform. So learning more about what people are building and how we can give them the information they're looking for in a quicker and more accurate way, that's the problem we're solving for. And so we're probably in more integrations for these other platforms. Is there a common API integration that anybody can go out and code against? I assume you haven't attacked the Elixir ecosystem quite yet. We're pretty small, but can I code against a REST API or something and get information in and out of SearchCraft? Yeah, so we have our core system.

1:27:49It's like a REST API, right? So the integrations and the SDKs just make it more convenient. Yeah. But anyone can use a REST API. We also did put out an MCP server. So if you didn't want to manually configure your search index cluster, you could just pull down the MCP server project and tell it to do that for you, and then it does it. Nice. That's the better way to do it, right? That's right. Yeah, sure. Save yourself some time. Yeah, so we're API-based. I like to think we have pretty good documentation. If we don't, someone please tell me. I'm sure they will. For sure. One of our favorite things to complain about.

1:28:32But yeah, no, as a developer-focused company, documentation was very important for us. And my mentors have impressed that upon me. Like, hey, if you're going to do a developer tool company, your documentation has to be good or people are going to come for you. I mean, as a developer, I find it incredibly frustrating. I'm only going to complain about some people right now because like aws a lot of people use it but come on i i can't be the only one who thinks their documentation is not good i've always i've largely avoided aws from the get-go except for s3 mostly because of the documentation and then just the web ui was just a hot mess forever i probably still is they got lots of services so it's not an easy problem to solve but the docs were always lacking and so i i was always looking for other platforms except for s3 which was kind of became a de facto for object storage the other services of aws i've never really glommed onto i used uh you know cloud front a little bit and their domain service their dns service route whatever it is 53 yeah and that's about it because of that reason i mean they literally have like a service for everything they have to much so much so like you don't sometimes it's a decision paralysis situation where you're like i don't know which one to choose but i think i think their document their weakness in the documentation has kind of like led to opportunities for other companies like you know these other cloud providers that make it a simpler experience for developers like for cell or flyer these other ones you know awesome well search craft is it's is it well baked is it ready for use get out there and try it as that's a situation yeah so i mean obviously like I said at the beginning of the talk, we're not trying to do everything, but what we do, we try to do very well.

1:30:22And what we try to do very well is content-based search, not like necessarily like log searching and whatnot. But yeah, people can, they can either sign up for the cloud account and try it there if they don't want to host it anywhere, or they could download the Docker, the Docker image and run it locally and play with it. Very cool. Adam, any other thoughts or questions before you let them go? I was just going to make note of what I think is a bunch of AI generated images. Don, do you know about this? He's smiling. Are you holding me again? Okay, so. I mean, I'm just noticing your website and I'm just noticing that they all seem, especially this one that has you on it when you're a pilot, Don.

1:31:05Okay. Are you talking about the Dr. Don? That's right. Dr. Don, tell me more. Okay. I've browsed away from it and I can't find where I was at. so we kind of go after elastic a little hard on on a site that's right elastic versus searchcraft so there's it's like a feeling bloated page right and so like we're kind of making fun of these like um you know the whole nlp or what was it the like the we go the exempt drugs right oh yeah feeling bloated right and so like uh i'm a i'm a well dr don is not a real doctor but i prescribe search craft every day to be less bloated kind of thing that's funny at the very the very bottom there and that's that is you that is that is me that is my face on somebody else's body see that's cool so we do we do uh that's cool that's cool that's the other problem with these things it's so stinking cool it is cool but i just noticed that every image on your site that's like a person seems to be this new trendy i don't know if it's trendy or not but it seems to be like out there a lot so i think it's trendy yeah so we we have a very strong like we have a very it's like a science fiction space kind of theme is what we we have um but yeah my co-founder his background is in design um and so he he makes a lot of these he'll he'll like have an he has the idea for like what he wants to do for the page like this page is going to have kind of this theme and then he's tools like mid-journey to get part part of the way there and then he'll like take that kind of generated image and then like edit it and refine it in like Photoshop or whatever, you know?

1:32:49I didn't hear you say the word Vectron unless I missed that part. Did you say Vectron? Did you talk about the dashboard? Oh no, we can talk about that. If you want to talk about the space thing, it's really cool. It's like Vectron is our user dashboard for people who are on search ref cloud. It's where you like sign up for your account. You manage your plan that you can do. There's like a UI for managing all of your search cluster in the schema and things. So part of Searchcraft is like, as far as like our experience building out search for customers and that we would hand it off to these startups, right?

1:33:28And maybe they didn't have search experts on their team after we handed it off. And so they would come back and be like, hey, can you like tweak this setting or like add some synonyms or whatever? and we're like, okay, well, we can, but we'll have to charge you our rate, right? And we're like, you know, that's kind of a gap we feel like in the market in that, let's make it easy enough so non-developers can tune relevancy for their search index. And so you can do that in Vectron. You can go in, you can add synonyms, you can change language settings, you can do things like boosting and weighting on your fields and not have to do a developer, and it doesn't require a deployment, You just save it and it updates it right away without having to like bring your cluster down or do a deployment or anything like that.

1:34:14So you don't have to talk to your developer once it's set up initially. Now that you say that, you don't have to talk to your developer. It makes me think we've been trying to do that for years. And now we're finally there where we can do that, I guess, ad nauseum. It's like it's a big deal. I like the imagery. I like the space-ness of it. It's cool. It's got a good tone to it. Vectron's a cool name. I like very much so that old school computer that you've got there to show it off on that comparison page. We'll link up with the show notes. Oh, yeah. Search and search craft. If you if you go in and go through the like the account creation process, it's like we're very inspired.

1:34:50Like I don't like Ridley's got alien like terminals and like the inspiration for that. Yeah. So I like it. Yeah. Being inspired is cool. It's cool. That's all I got. I was just noticing as I was browsing these various styles of imagery and I wanted to, since we've been so deep in AI, I was like, well, these seem like they're AI generated. So they probably are. Yeah. The images, the images, well, there, there was human curation and intervention. That's right. You know what? I feel like we have to apologize. It's a jam session. Yeah. Somebody was like, Hey, was, is that, did AI do that? Yeah, absolutely.

1:35:26Because I'm concepting and I'm ideationing with my new buddy. That's how it works. Yeah. It's like to take it back to development. It's like a pair programmer, right? Like you're collaborating. So instead of collaborating with the person, you're collaborating with the machine. Same with design. Like you're collaborating with the like mid journey, right? To make images. But the idea still comes from the human. Yeah. Yeah. For now. For now. For now. Well, Don, this has been awesome. We look forward to seeing you in Denver at our live show, hopefully. And searchcraft.io, of course, if you haven't checked it out yet, go see Dr.

1:36:04Don at the bottom of the Elasticsearch comparison page, as well as some, there's some cool comparisons. Each one has their own vibes. I like the Meily search versus Searchcraft is like old school, pixel-y 8-bit Pong comparison. So each one has their own theme, which is rad. Any final words? Anything else to say before we let you go? Yeah, just, you know, encourage people to check it out and give feedback. And we're very much a collaborative bunch here. So, like, if you have ideas about things, like, we'd love to hear them. And we'd love to hear what you're building. Come on. Awesome. Nice to meet you, Don.

1:36:45Thank you. yeah thanks for coming on great being you guys

1:36:52well that's it this show's done thank you for tuning in fun conversation with a member of the community don mckinnon so awesome having him on the show as we discussed we are going to be live in denver shortly go to changelog.com live learn more 15 ticket for anyone free for changelog members plus plus members that is changelaw.com slash plus plus it's better you know what it's been a while it's been a while but it is better again changelaw.com slash plus plus and of course the beat freak in residence break master cylinder those beats are banging banging banging and hey break master will be there too so denver awaits okay that's it this show's done we'll see you on Friday.

1:38:20Thank you.