AI for React developers and beyond with Alexandra Spalato
===

Noel: [00:00:00] Hello and welcome to PodRocket, a web development podcast brought to you by LogRocket. LogRocket provides AI first session replay and analytics, which surface the UX and technical issues impacting user experiences.

Start understanding where your users are struggling by trying it for free at logrocket. com. I'm Noel, and today I'm joined by Alexandra Spolato. ~Uh, ~she's an AI integration specialist, front end visionary,~ uh,~ here to talk to us about AI for React developers and beyond. Welcome to the show. How's it going?

Alexandra: Hello. Thanks to have me.

Noel: Of course. Yeah, I'm excited. ~I, uh, ~I went through your talk and read the slides~ and brushed up as best I could here. Um, I, ~in my head, ~there was, ~there's kind of ~a, ~a couple of different pieces ~you were, ~you were covering. One is like the tooling side,~ uh, ~like the dev experience of using, ~you know, ~LMS and the benefits they're providing for development.

And then the other side's like the functionality, the product side and ~using, you know, ~Using those tools to ~like ~enrich customer experiences and add features to apps and stuff like that. ~Um, ~is there one of those you'd like to kind of jump into first? Is there a place you think it makes more sense to [00:01:00] start?

~Yeah. So I guess I'm, yeah, I'm, I'm curious. Like why, I guess if you think it makes more sense to kind of talk more about the, the tooling side that like devs specifically front end devs are using to help develop or like the tools that they can leverage to, you know, add functionality to their apps, we can start with whichever, whatever you prefer.~

Alexandra: yeah, both things ~are ~are interesting, but I would begin by the fact that I think that there is really a gap between AI and front end development in general, not only React, because what I noticed is people that come from data science, machine learning, etc. ~Uh, ~don't know front end. ~Uh, ~there is other people that do business and use no code and low code tools, because there is a lot of business to do, don't know front end neither.

So they use,~ uh,~ no code tools to build the front end. And front end developers, uh, think that, ~uh, ~AI, It's impossible for them because they think they have to learn data science and machine learning and all this stuff ~and, ~and if you search on Google or even on chat GPT ~and, ~and you ask, what do I need to learn if I want to become an AI engineer?

It will answer,~ uh,~ machine learning, et cetera. ~And, ~and this is overwhelming and the good news is that is not the case. But yes, let's begin with the tools because I think that's. [00:02:00] What any front end developers,~ uh,~ can and should use, ~should use~ because,~ uh,~ thanks to these tools, we can code faster. And so as,~ uh,~ this famous sentence, AI will not replace you, but somebody that uses AI will replace you.

That is what happening. So there is many tools, the most well known, it's GitHub Copilot. Personally, I use cursor. It's,~ uh,~ An editor that is exactly like Copilot, but with AI integrated, and you can choose the model. ~Uh, ~you can also,~ uh,~ give them documentation. ~Uh, ~for example, if you are working with Remix, which is ~my, ~my favorite framework, you can,~ uh,~ charge.

The remix documentation, it will be there,~ uh,~ for your next projects, et cetera. And their copilot completion,~ uh,~ doesn't break your code. So with GitHub, sometimes it breaks the code, et cetera. ~Uh, ~and so I really like, ~uh,~ this one. So there is, Other tools also to [00:03:00] program with,~ uh,~ with AI,~ um,~ and all the LLMs that,~ uh, that, ~that you can use.

But,~ uh,~ there is a lot of opportunity ~to build, ~to build things. Yes.

Noel: ~Yeah, for sure. Is there, do you, ~do you think in particular,~ um, ~front end devs are less likely to be using something like, you know, some editor assistant, like copilot and it's ilk, or do you think it's kind of across the board? There's hasn't been a lot of adoption yet.

Alexandra: ~Um, ~I don't know. I think many,~ uh,~ many devs use our systems now. Most of them use. I was in a conference in Madrid with 100 people ~and, ~and many people raise their hand. ~Uh, ~there is also things I have not tried for the moment, like builder. io. You can take a design from Figma and get,~ uh,~ The code in, in, in react with tailwind, et cetera, adapt, adapt it to your way of coding.

And that gives me also,~ um,~ what you make me to dive more also ~into, ~into design, because I have also design skills, but as well as you need to know how the code works to use [00:04:00] this tool, I think,~ uh,~ then you need to know how Figma works. Chef sensors designed to do good things,~ uh,~ with this. Tools together, but,~ uh,~ for creative,~ it's,~ it's great because you can go faster and create more things, et cetera.

Noel: ~Yeah. Yeah. I think, ~I think anecdotally for me at least, ~like I'm, ~I have found, ~you know, like~ the editor tools are probably the most impactful, ~like~ to my general productivity.

Alexandra: Yeah.

Noel: In comparison to any other tool, like, you know, ~whatever ~I'll open LLMs and I'll prompt things and ask questions and try to use it to help me debug and stuff.

~But I think like it's, it's autocomplete, but the really good autocomplete of the LL, like the, ~like, I think that saves me a measurable amount of time. So ~I, I, ~I'd agree with the thesis that I think most people are.~ Is there something, is there any, I guess, ~do you think ~it's, ~it's more difficult almost in the front end world?

Do you find that these autocomplete tools have more trouble with front end code than, you

Alexandra: I don't think so, but ~especially, ~especially cursor, I don't know if you have tried it compared to a copilot as you have a copilot plus plus, ~uh,~ for example, you change a variable and he immediately, he changes, uh, you, uh, pinching a tab and, and he changes the other places, or if [00:05:00] you are coding,~ uh, ~using the sidebar, ~you know, to, to, ~to ask things and the code appears ~and, and, ~and you push a button and Change it in your page without breaking it, which are breaking the parts that you don't have to change while copilot.

Um, and then you can just break this. I hope that will change, ~but, uh,~ and so that's, that's really good. Also, one thing I love with these tools, it's once you are in your editor and if you're stuck, or if you don't know something, you ask questions and it's like having somebody besides you. ~Uh, ~and you don't have to go on the web to stack overflow or Google or whatever, and then get lost and find other things.

And so you stay in the flow ~and, ~and you feel,~ uh, ~You have company, ~you know, and, ~and sometimes it's stuck and ~it's, and, and~ it's no way. So then you have to search a solution in another way. ~I,~

Noel: for a while there for, ~I don't know, ~some measurable amount of time,~ uh, at least~ I would speculate that a large number of front end devs were kind of operating ~in a pretty slow pace. Um, like~ with a pretty slow feedback loop on any changes they made and like you'd have to all type out like webpack would take a long time to build and dev [00:06:00] environments and stuff like that.

So~ I do, ~I'm with you like that, maintaining that flow state whenever possible ~is, is, you know,~ there's a lot of utility there. Are there any other tools that you ~like~ think are being underutilized by developers kind of to help them with their development process or~ is it,~ is it mainly just checking out ~what's,~ what's coming out?

Alexandra: nothing comes to mind. I use cursor all the time. And sometimes I will ask,~ uh, uh,~ yes, I don't choose any more chat GPT. I use, An app named Type in Mind, and you can use any model with it. You can put it in folders, etc. And generally I use,~ uh,~ now Cloud Summit 3. 5. It seems that is the best model now ~for, ~for coding, and it's not so expensive.

~Uh, ~so yeah, it's, What's what I use now, there is GPTs for coding, et cetera. ~Uh, ~you can set up things for learning. There is many things to do, but yeah, for coding,~ uh,~ I use mostly, mostly course. So I know you, yes, we flipped it. ~I, ~I try it today. ~Uh, ~and,~ uh,~ it can give you the, [00:07:00] you know, the workflow to make a project.

So that's another level, but I didn't try totally. I saw a guy on Twitter that. Has never called and in four hours, you made an app,~ uh,~ with,~ uh,~ to make,~ uh,~ notes. ~So, ~but I think the guy that does that, you know, has already a pro, uh, developer mind. You know, it's,

Noel: primed. They're in that mind space already.

Alexandra: yeah, it's not, Oh yes, AI can do it. No, the guy, even if it's beginning, perhaps you will learn code after, you know, it's yes,

Noel: to know how to prompt to get the code to be right. There's a little bias there for sure. Yeah, I guess, do you worry about this at all? I feel like there's a lot of, I don't know, fear is the right term. Maybe skepticism is a better term, but ~Um, ~and I think I tend to agree with this.

~There's a lot of like,~ I would not want to step into a large code base that had been initially spit out by an LLM and then modified by an LLM over time, it just kind of feels like there'd be. A lot of inconsistencies maybe in ~like~ [00:08:00] paradigms and models used ~and, um,~ and stuff like that. ~Do you,~ do you have any of that fear?

Or ~do you think, ~do you think like ~there's some,~ there's some force that's kind of pushing people to move a little too quickly in that direction or?

Alexandra: ~I'm not, uh, no, ~I'm not afraid, but ~I'm, ~I'm a creative. So it's perhaps my way of seeing things that these tools, finally, ~you know, ~the first people that begin coding or Ada Lovelace that just have pinch card, et cetera, and I'm not imaging. How hard this was, ~or, ~and now we have,~ uh, ~JavaScript ~and, ~and these languages that are almost like English and these beautiful code editors with nice colors and not a completion and everything.

So, ~uh,~ perhaps with people of these times, I'd begin with almost nothing. This, ~you know, ~the question would be the same,~ uh, ~but AI is going faster. ~Uh, ~so at the end, ~it's. ~I'm a creative. I become developer because I'm a creative. ~So it's a, ~I think there is different type of developers and I think code is something very creative.

~It's finally, ~I like it because ~you can, ~you can build things. You can create things. So if there is a tool that unleash my creativity and help [00:09:00] me to code faster and do more things, I say, yes, I want it. ~Um, and ~one thing I noticed is that. In AI in general, not only in coding, ~it's finally, if you are not, ~if you have not the knowledge, if you have not the creativity or the talent or the culture, you are not going to do good things.

Even if you master, The AI tool. ~Uh, ~so here you need coding, you need ideas, you need all that. And then for example, my partner,~ uh, which ~is an artist and now ~is, uh,~

Noel: ~don't know. Yeah. It's weird camera~

Alexandra: ~now he has ~He has really delved ~into, ~into mid journey and he's doing really great art with that. But if I look at the showcase of mid journey,~ uh, ~there is perhaps 1 percent of people that does really interesting things like him. And it's not because he's my partner. ~So, uh, ~This creativity in this talent is transferable.

It's,~ uh,~ yes, it's what it is. ~So, and, ~and I was speaking at this conference in Madrid,~ uh, uh, ~and there was a guy from Microsoft doing a talk about that creativity and AI. And the equation was that before AI, you [00:10:00] need,~ um, ~Talent plus,~ um,~ skills,~ uh,~ plus resources,~ uh,~ to do something. And here, if you have talent and AI, it's a limited creativity, but you need the talent, you need the creativity.

~Uh, that's, and ~that's what makes it human. So AI can demultiplicate it. And I think ~this, ~this is true for ~all the, um, ~all the areas.

Noel: ~Yeah. Yeah.~

Alexandra: Not only coding.

Noel: I think ~it, it, ~it tends to butt up a little bit against, ~I don't know if like,~ maybe ~how, ~how a lot of devs kind~ kind of ~look at code,~ uh,~ and look at ~like, um, you know, ~design principles and like how code should be written. Cause I think, ~I think in, in, ~there's definitely a camp of people. Or a line of thought that is like ~for,~ for most applications, ~like there is, ~there are, ~is a correct way or ~a few correct ways that are ~like~ the best, ~like~ the cleanest architecture for a given system.

~Um, ~and I think that there's something to that argument. Like ~there, ~it is ~like~ an engineering practice. I mean, ~like in one, like. ~One could even argue ~in, ~in really any engineering, like there is ~an art,~ an art to those things, but there's also ~like,~ I think, measurably wrong ways to do things a lot more of the time [00:11:00] than in traditional art,~ uh,~ you know, like our ~post ~postmodern view.

On art. ~But like, um, I think, yeah, like,~ I'm not sure if that means that ~they're like ~this fear of like, oh, ~any,~ any code that any AI is ever spitting out. It's always going to be problematic and buggy and not worth it. And I think there's ~like~ a great value just in the capacity to be able to iterate quickly, even if it's ~like~ maybe not a perfect version right away.

It's like, well, I can like get something off the ground and

Alexandra: Yeah, but you have to do it step by step and

Noel: Yeah.

Alexandra: it's fine, et cetera. ~And, and, and, ~or tell it to do it this way. Sometimes ~I. ~I tell AI to do things I know how to do, but oh, I have to do this form with that, that and that. Okay. I know how to do it, but AI will do it faster than me. ~So, ~or today I was searching,~ uh, ~with,~ um, uh, ~APIs for flight APIs, because I'm building a product that works ~for, ~for travel.

So instead of say, okay, there is this API now, okay, extract me,~ uh, ~this and this information about, about the flights,~ uh, and, ~and things like that. Things that. Are tedious to do, you know, you know how it works, ~but, uh, and, uh, ~and AI do it perfectly. Because it's not complicated. ~So~

Noel: I agree. ~I'm a person like I like,~ I like [00:12:00] examples ~like~ when I'm exploring new libraries and stuff a lot of the time. And sometimes the docs will give you those, but it's hard to find ~like~ for a given utility, like the one that's kind of a good projection of what you're trying to do. And I've found that it's like, I can be, I can feel like generate an example that does this and like, Oh, okay.

I see. Like, I'm not even going to use this code, but I understand now ~like~ how this library works ~or~

Alexandra: yeah,~ and,~ and AI explain new things too. ~So, ~and you can ask it to explain you like a five years old, if you need, if you don't understand. So it's, and in your language, if you want, okay, I'm French. So explain me in French, please. ~And~

Noel: Yeah. ~Yeah. I'm sure. I'm sure that's a whole. Yeah. Like,~ especially for languages where docs are probably not often written in their first pass. ~Like, ~I bet that that is super, super beneficial.

Alexandra: yeah, it was a way that ~it's, ~it's explained, ~you know, ~sometimes some docs are explained in a complicated way, or you want just to, ~you know, To~ find something in the dark to do something, ~you know, and, and, ~and you don't want to read everything, et cetera. So then ~you can, ~you can ask AR ~to, ~to explain you how to do that.

Noel: Yeah. That's cool. Okay. ~This is a, ~this is probably as good a segue as any to ~like, let's~ talk about some functionality now that AI is ~kind of, um, you know, ~Opening up. [00:13:00] And again, I'm going to start at a hyper specific point here, just cause ~you sparked, ~you sparked the thought. ~Um, ~but I know like, there's been quite a bit of talk about how we make data searchable ~or like ~for ~any, ~a given domain that a company's got.

It's like, okay, I want to make it so people can search this and. For a while, we could do this when, ~in, in devs, like ~we were doing manual vectorization to find good matches. So we could do search and now there's like ~kind of ~products being built on top of that, where they're still vectorizing under the hood, but they've like abstracted away where like, as the dev, ~you don't, ~you just like upload things or point this at your database and like, we'll vectorize it and you don't have to worry about anything.

Have you found that that. works pretty well ~in like, um, for, uh, like ~across language boundaries or is, ~is like ~vectorization of most data, totally broken.

Alexandra: You mean across language about,~ uh,~ language, Spanish, English, or ~like,~

Noel: ~yeah, ~yeah, exactly like if you're

Alexandra: ~I, I, ~I learned a very interesting thing at this conference ~in my, ~in Madrid,~ Scott, ~Scott Hanselman was doing the keynote ~and, ~and I learned two interesting things,~ uh, is that, um, ~if you are using another language in English, it's going to cost you [00:14:00] more on tokens because.

Yeah, I is training mostly in English. And the second thing that was really interesting is that,~ um, ~you have to be nice with AI polite, et cetera. ~And I knew that ~I knew there was papers about that, but,~ uh, ~it's,~ uh, ~proved ~because, ~because AI is trained on the ~wall. ~Web and there is nasty people on the web.

So if you're not nice, you will have bad answer. And if you are nice, you will have the answer ~from the, ~from the nice people. ~Um, ~but except that, I think for what you describe is what it's called a rag retrieval, augmented generation. ~Uh, and so is ~if you want, for example, to make, ~I don't know, ~a chat bot for customer support of your company, you need.

To feed,~ uh, ~the large language model with the information from your company because the LM is, is trained on the whole web and we don't know the information about the company. So, ~uh,~ you upload this,~ uh, ~data, it's got in chunks,~ uh, ~because AI doesn't [00:15:00] have a limited context and then it's vectorized and Send to a vector database and so that allows to search and to ask question in natural language You don't have to say the exact same word is searched ~by similar by by ~by semantic similarity and these vectors are arrays of Multidimensional arrays and their numbers and so if this number are close them it's similar.

So ~If you, ~if you talk about,~ uh,~ a Fox Terrier, it will know that we talk about dogs, et cetera. And, ~you know, ~it's, you don't have to say the exact word, the

Noel: Yeah, ~nice ~nice. So ~it sounds like ~it sounds like in general ~Yes, it is like there that ~it's helping at least that search across the language Barrier, ~you know or~

Alexandra: Oh, yes, ~you can, ~you can do,~ uh, uh, ~the documentation. ~You, ~the information that you upload ~in the, ~in the vector database can be in one language and you can ask question in another language ~that, ~that works. Yes.

Noel: that's pretty ~cool. ~Cool. ~Um, Oh, Elizabeth's asking if you can flip your hair back again, if it's not too much trouble. Thank you. I appreciate it. Um, no.~ Okay. So let's zoom out a [00:16:00] little bit. ~Uh, so like ~it feels to me and ~maybe,~ maybe this is ~a, like~ a vast over assumption, but I would suspect that devs that are.

Largely front end focused have historically not been in charge of search or haven't had to do much search implementation. Do you think that's changing now with the capacity to like, you know, ~up like ~point at a docs website or like point at some end point that your front end can spit out and like, okay, now ~we've, ~we've built.

Vectorized search database. ~However, however, that's, ~however, that's built it. We can then like pump into a model and I can answer questions in a more natural way. Do you think that that's actually shifting? ~Who is working on, like,~ who is solving this problem? Or do you think that that's still the purview of backend most of the time?

Alexandra: ~I, ~I don't know. I think that I was saying just at the beginning, ~it's, uh, ~front end devs don't know that there is opportunity for them

Noel: Hmm.

Alexandra: by learning AI ~because of, ~because it's new and people think that is ~a ~machine learning, et cetera. And now with the large language models, it's like, ~uh,~ driving a car. You don't [00:17:00] have to go ~in the, you know, ~in the motor, et cetera, in the mechanics.

So that's the research engineers. And so you can interact with ~API ~APIs and that's what front end engineers do. So they should do it. So, ~um,~ no, for. ~Uh, ~uploading to a database, et cetera, you don't need,~ uh, ~backend. And there is a virtual ai, SDK, which is really good, and it's in JavaScript. You can do,~ uh, ~many thing where there is also a, uh, long chain and thing.

But with the virtual ai, SDK ~and the, ~and the open ai,~ uh, ~API, ~you can, ~you can do many things. ~Uh, ~so effectively, when you are interested by AI, you begin diving into back end tools. That's what I'm doing with Node ~and, ~and with Python too, because I'm really interested. But if you don't want ~to, ~to do nothing with, with back end and you're a front end engineer, You can also just work with APIs, ~uh, and, ~and,~ uh, ~there is also some low code tools,~ uh, ~that can,~ uh, ~provide [00:18:00] an API.

So ~you can do, ~you can do many things. ~Uh, and, and, ~and especially ~as, ~as I say, there is an opportunity because,~ uh, ~there is a gap. There is nobody doing both.

Noel: ~Mm hmm.~

Alexandra: ~I~

Noel: ~Yeah. No, no, for sure. For sure. Do you think, do you think that they're the SDKs, the tools and it may be more broadly, I think this is a better question. Do you, ~do you think that the abstractions that we are providing with tools,~ um, ~like Lang chain and,~ uh, whatever, what's ~what's Facebook's,~ uh,~ Lama index, ~like,~ do you think that those are the correct place to be jumping in for front

Alexandra: was thinking so because ~I, I've been, ~I've been pulling my hair out, ~you know, because, ~because there is no path. And so you have to find your path, especially if you are a React developer and everything is in Python. So sometimes I jump into Python. Oh, it's an easy language. Yes, but I don't master it. So if I don't know how to do something, I'm stuck with what I do.

And then I come back to JavaScript and there is no, no examples, et cetera. So I begin with. Yes, with long chain, et cetera. But now I'm building,~ uh,~ this product ~and, and, ~and for that, I realize exactly what it takes. And so I'm working,~ uh,~ [00:19:00] mostly with the APIs, with Vercel AI SDK. And you realize. At the end, when you build an application,~ uh,~ it's got all the hype now with the agents, etc.

But what, yes, first, what is an agent? An agent,~ uh,~ work with tools and he will decide which tool to use depending on what happens. So you give him a lot of power. But mostly,~ uh,~ most of the time, you know ~what's Uh, what, uh, ~which tool it needs to use. You don't necessarily need an agent. And another thing, ~what is, ~which is important is finally we are providing solutions,~ uh,~ and AI is a node in the solution, ~you know, ~and you have to use it at the last moment, all what you can do without AI, do it without AI.

And at the end, the important is the data. It's always,~ uh,~ data input process and data output. So at the moment, if you need to, you have the right data and you need to analyze this data, then AI is perfect for that. Or you need to generate,~ uh,~ text or images with,~ uh,~ AI based [00:20:00] on the data you give them, then it's perfect.

~Uh, ~but yes, I think understanding Open EA,~ um, uh, ~API and the rest of the SDKs. Really good. ~Um, ~and then if you need something more powerful, then yes, long chain. And there is new frameworks all the time too, so there is , everything is moving. ~It's, ~it's even not in, in version alpha,~ so,~ but it's using production.

So yeah,

Noel: ~I think, ~I think I agree. ~I'm, I'm ~part of me is a little bit skeptical that there will be a ton of, ~um, ~practical use cases that ~front end like ~front end focus devs would need to lean on something like Lang chain for, I mean, maybe this is even beyond front end devs, like just most application

Alexandra: depending on the complexity ~of the, ~of the app.

Noel: Right. Yeah. ~What's ~what's needed? ~Um, ~I mean, I think that there are cases, but, ~you know, we always use the examples like, oh, well, ~I think the examples you always see are~ like, they're like~ text prompts. ~Like it's always, ~it's always chat. It's like, okay, well, if a user asks about the weather in a given location, ~it's like,~ okay, well, then we can give the chain the functionality to go.

Call a weather endpoint or [00:21:00] just return to the user immediately. ~That's like, ~I think that that ~is,~ is the case for chat. Like it's cool. If you can introduce all these capabilities and motalities and logic flows, but outside of chat, ~if you're not, ~if one's not building a chat app, ~I think, I don't know, ~it can be kind of hard to conceptualize.

~These like highly conditional, like ~the need for complex orchestration

Alexandra: ~yeah, then you need, ~if there is a complex orchestration, then you need this,~ uh,~ these frameworks. ~Uh, but. ~But yes, what I advise, and this is not what I have done everywhere is yes, begin by the APIs, see what you can do with it, build something,~ uh,~ takes over sale SDK. And one thing was the rest of the SDK that I love.

And I think is a huge opportunity. I'm really fan of that is the generative UI and nobody knows about it. ~Uh, ~basically, and ~I, ~I made a small video. I have to post it. It's,~ um,~ In a chat, but it doesn't have to be a chat. A chat at the end is for, it's a way of

Noel: It's just one

Alexandra: doing things. It's a layout, you know, it can be a page where you write and you ask things, etc.

So [00:22:00] instead of rendering text, it renders text. React components. You don't generate them. You have to build them before. But I built this ~little, ~little chart where ~I'm calling, uh, ~I'm using a recipe,~ um, ~API, and a quote API. And if I ask for recipes, it's going to show me all the recipes,~ uh, ~in nice React components with the image, the link, etc.

If I ask for a quote, it's going to ~show ~Give me a quote if I ask for anything else ~is ~going to give me text and I think that for e commerce ~Uh, ~you can connect the api put all the products in a rich development Augmented generation vector database and people can ask things in natural language and the products will appear magically there so they are one click to buy.

That's what e commerce wants. And for personalization also,~ uh, ~the website can look different. depending ~on the, ~on the navigation ~and, ~and user profile, et cetera.[00:23:00] 

Noel: Yeah. Even, even on the layout. Yeah. I think,~ Yeah. I think, I think the cool, yeah. Like on, on these, ~on these generative ~like, uh, ~tools, ~like I think the, the, ~it's interesting to me. I haven't really seen it in practice outside of the like examples they always do. But yeah, I think, ~I think you're, I think that ~there ~is, ~is a potential future where it's like, ~yeah, ~an e commerce site, ~like ~one can go on and search and then dependent upon ~like ~the search terms that are actually used.

We let the model determine ~like. ~What data is displayed per result? ~Like if, if someone, you know,~

Alexandra: ~Even,~ even it can be, I imagine a blog, but the blog is from a brand and this brand has this blog because they want to sell products. And I don't know, this blog is about travel, for example. And so the user asks, Oh, I'm going ~to, ~to London this weekend. ~How, uh, should I, uh, which, ~which clothes should I take, et cetera.

And so it takes the weather and show, Oh, it's raining. ~And, ~and he shows a product, which is an umbrella, for example. And so the person can buy it directly there, ~you know, ~and it's a blog, but it's just chatting ~and, ~and they give him what they need and people buy it. ~So, yeah. Yeah. ~And you can mix and match different products there and then it's creativity again and market research and UX that will define ~how to, uh,~ how to [00:24:00] do it.

But, ~but the,~ the technology is there and it opens many, many possibilities and it accelerates the sales. ~It's,~

Noel: ~Yeah, there's, I think that there is, again, I'm, I'm not, ~I'm not a hundred percent bought into that. That is the future we will end up seeing, but I think it's possible. ~Like there's, ~there is potential that this could. We could end up ~in, ~in worlds where ~it's like, yeah, ~the behavior that a user's trying to evoke from a app or a website or ~whatever, ~whatever abstraction we want to make is ~like~ driven pretty heavily by.

Models somewhere making decisions and trying to infer what the user is trying to do.

Alexandra: yeah, the website would become much more interactive ~and, and, ~and I think if it accelerates sales. It will come ~because, ~because the goal is always to sell something. ~So, yeah,~

Noel: No, ~for sure. ~For sure. I mean, ~the, ~the developer in me ~that is ~that ~did like, you know,~ has always had to debug lots of state and figure out user issues. And how did this person break? This thing is like kind of pet terrified of this notion. It's like,~ uh,~ I don't know what a user is seeing when they're encountering an error ~is, ~is, you know, tricky, but like, it will probably [00:25:00] have.

Tools that are better at analyzing those things and, and data will flow around, you know, front end monitoring tools, stuff like that. ~Um, ~could be, it could be super nice, I guess.

Alexandra: It's not super complicated. It's just the function calling from OpenAI. So if you ask recipe, he knows that. So you have the schema that is, and he's asking, he's going to call these components. ~And~

Noel: Yeah. You can look at the logs and it

Alexandra: It's quite controlled, ~you know, ~it's not the agents that do 100 things, et cetera. This is quite controlled.

Noel: Yeah, totally.

Alexandra: can be more complex.

Noel: Yeah, yeah, I agree is okay. So I feel like we've covered a lot, ~I guess, kind of~ to wrap a little bit. ~If,~ if there are devs listening that ~have kind of~ their front end devs, ~they've done a little bit of probably maybe,~ maybe they've looked at the,~ um,~ you know, like opening I APIs and stuff a little bit, but haven't.

~Haven't ~played with them too much. What would you recommend for them? If they've got an hour to ~kind of like~ figure out what tools they could put on their tool belt, where would you implore them to kind of jump in and explore?

Alexandra: Yeah. So OpenAI API and the Vercel AI SDK, ~it's, uh, ~it's really great. ~And, ~and this generative [00:26:00] UI work with the React components, which is annoying because you can only use Next. js and I'm a fan of Remix.

Noel: ~Yeah. ~Yeah. Right. ~Right. Yeah.~

Alexandra: And,~ um,~ but I think it, yes, it's remix is going to have also write several components. So I think there is,~ um,~ yes, it's a best to begin in my opinion.

~Uh, ~and then try to build something with that. And then look like chain and orchestration frameworks,~ uh,~ if something is more complex. And he has also this,~ uh,~ low code and local tools. So the developer generally, ah, no, I don't want no code, but I myself diving in some of them. For example, there is one named relevance and it's very developer friendly because you can interface code.

You can, I even, I've been. Even be able to create npm packages for function. I want to use and use them there So it's input output ~and you and ~and you have blocks of code and they have an sdk also And so it's ~uh, ~it's interesting and at the end you have an api so you can integrate it There is this [00:27:00] one.

There is also lang base that is not ~Uh, ~released,~ uh,~ but ~I, ~I have access to the beta, which is quite simple to use, and you can create what it's called pipes, which are,~ um, um, ~you put your front, ~all the, ~all the, all what you want to put to the, with the LLM and,~ uh, all, ~all the,~ um,~ settings. And then you deploy an API so you can reuse it, et cetera.

You can also do Retrieval Augmented Generation with it and all that. It's a flexible,~ um, um, ~shareable, reusable. And also if you work with a team, perhaps you have people that are not coders, but that will work in the prompt. ~Uh, ~and they can release another version. But the API doesn't change. So the developer doesn't have to say, Hey, ~oh, you know, uh, ~the prompt engineer have to come here and change the prompt,~ uh, ~and go into the code or send to me, ~you know, ~with these tools, ~they can, uh, ~there is collaboration.

And ~there is, ~there is many tools like that, that are interesting. And so even as developer is interesting ~to, ~to look at them,~ uh, ~ because you can use it with code too.

Noel: ~Yeah, this is, this is an aside and again, we should wrap up, but~ I think it [00:28:00] will ~be, it'll ~be interesting to watch and see like, if these kind of bespoke tools that are these like kind of workflow generators that are ~like, they're ~built around ~like ~prompting and LLMs as their main thing compared to like the ones of old, right?

Like the Zapiers and stuff. I think it'll be interesting to see ~like, ~Which of those ends up being the kinds of winners in the space? Like are the old ones going to end up being able to implement enough where it's like, okay cool They have ~like ~integrations to all these things and now they've also got steps in there that can do ~like ~smart ~You know, ~prompt AI based stuff.

~Um, ~just an observation. ~Like, ~I think it'll be interesting to see ~where we like, what, like where, ~who ends up winning in these. I'm sure there'll be a bunch of them.

Alexandra: ruiners. It's,~ uh, ~Yes, by building that you decide what to use and

Noel: What's better

Alexandra: yeah,

Noel: for a given case. Cool. Was there ~any, ~anything else you want to,~ uh, kind of, uh, ~suggest listeners check out as we wrap here

Alexandra: just use it and learn it and because I think there is a There is many opportunities to build things or to find new jobs, because there is really high paying jobs ~in the, in the, ~in the space,~ uh, with, ~with AI.[00:29:00] 

Noel: Mm.

Alexandra: and yes, don't be afraid of it. We have to go with it. There is no choice. I, yes, one thing it's, there is no, I like it or I don't like it or I'm against, it's there.

~So, ~we have to go with it anyway. Thank you.

Noel: Right. ~Right.~

Alexandra: yeah,

Noel: The cat is out of the bag. Cool. Well,~ uh,~ thank you so much for chatting with me, Alexander. ~It was, it was, ~it was a pleasure.

Alexandra: and same for me.

Noel: Take it easy.

Alexandra: Thanks.