Generative AI + No-code – perfect storm?

okay so my esteemed guests tonight so so James you might as well start since

you’ve got the mic so James from Deputy hi uh I’m guest slash

uh venue host Tonight uh so I’m James staff software engineer at Deputy

um and for those who don’t know Deputy is a uh sort of workforce management

um platform time in attendance and fostering scheduling and things like that for

shift workers and businesses around the world and you’re mostly an Android guy

yes I don’t really pitch and call myself at Deputy yes but in

the past I’ve done Hardware interfaces apis desktop apps

mobile great very good thank you and Adam Adam is the CEO and founder of

buzzy oh cool thanks thanks David

um yeah Adam from Buzzy uh founder what do I call myself uh husband father gig

Surfer um founder of Buzzy Buzzy’s a no code platform I’m a geek so I created a no

code platform um we use generative AI to take an idea

to create turn it into a brief into a data model into a design which is a

figma file and then turn it into a working either responsive application

full stack by the web or native in about 15 minutes

so we’re all about accelerating things and making it easy for you to climb the

mountain and I’ll talk about that a bit later very good so IOS and Android

the um yeah so we used react native under the covers for the native ads so

both IOS and Android and who knows what other platforms whatever supported whether we do TV and windows and stuff

down the road but yeah at this point of time just IOS and Android right very

good so um the folks who are online with muted you so

um we get a clear run but if you want to put your hand up then we can actually go to questions at some stage so

um I think internally here I think we’re kind of you look like a pretty orderly crowd so I think what we’ll do is we can

take take any questions or or you know comments on the Fly because that’ll that’ll make it richer and um hopefully

we get a bifo somewhere along the way right so so best just um next slide please

this was this was really a dodgy thing I just did quickly just to kind of like frame up the different types of things

that have been happening as I mentioned there you know there’s a bunch of tools that have been around like notion and their table that are actually allowing people

to get a lot of stuff done and even prior to that you know whether it was if this then that or

um you know zapier there’s a lot of stuff that’s been you know around sort of doing things without having to write

code but in those in that case you know they’re just connecting different systems and using apis so

didn’t want to sort of focus on any of those sorts of things MailChimp is would you ever think that

mailchimps like no code well kind of like back in the day used to write HTML code and uh that allowed you to do not

much at all because then you have to figure out then how to actually get it into a SMTP mail message and do all that

and then when tools like MailChimp came along that actually turned that into a consumer usable thing or even better

than a power user so that’s that’s an example of that and my company contextual is a little bit like that now

for in-app engagement so the version that we do at the moment we allow once

our sdks inside our iOS Android or web app then you can actually do

guides and tips and tours on the fly so we’re not trying to be a no-code platform for developers or to displaced

developers we’re trying to free up developers to push things that just like with mailchimpings you know instead of

actually getting the dev to do the hard coding of HTML in this particular case instead of getting the div to do the

hard coding of Engagement inside the application tool tips you know Pop-Up Videos targeting that it particularly

uses particular points on that Journey you can use us as the platform for that and I’ll talk a little bit about that

later on then you’ve got the web Builders which I mentioned briefly bubble and Diplo Etc

public everybody fuzzy in there as well so well I put you down there which was

absolutely what we do wear this one oh you do web as well okay my apologies

um so and then we then we get into Dev tools as well too and I probably missed some other some other categories but

that’s where we start to start to see you know obviously the the AI is starting to bleed in as the as The Iliad

so definitely Adam’s on the front end of certainly Ai and no code that’s that’s

great to have him here tonight and we’ll also talk to this chap here I’m sure you’ll have a few pointers on that and

at least you’ll provide some more color on copilot and things like that as well too so

is everybody okay with that does anybody think I’ve missed out a massive category that we should be talking about

right cool all right so let’s have a look at Adam’s uh value prop is a some dude on Dick

clock I’m allergic apparently you can just ask AI to build apps for you so I’m using to do that is

Buzzy dot Buzz it’s basically just lets you turn figma designs into full stack web or mobile apps and it even generates

the figma designs for you so you really don’t have to do anything Buzzy will even give you these giant component libraries that the AI can use to work

off of and then you just go over to your app here there’s nothing here right now and you go to plugins and download the

Buzzy plugin I have this AI assistant right here and I’m just going to ask it really quick create an app to host local

basketball competitions I’m going to send that and it’s going to start generating me a brief and boom it’s done

generating it’s finished it called it hoop Zone which I kind of love and then it gives a whole list of functions user

roles team captain admin user functions schedule games scores tons of stuff let

me go over the data model and I click generate data model and boom would generate a bunch of stuff different uh data models for standings scores games

competition types things like that and we go over to markup and we click generate app design wow okay it

generated 52 screens 70 components 150 Fields all of this completely automatically and then I just go over

here and I’m gonna click publish and back behind it on the actual figment page you can see it just generated a ton

of those screens all that completely automatically it’s wild Publishers do a QR code and I can just start testing it

my dad this is crazy like everything just works there are Fields there are forms

go to games oh my gosh that’s awesome if you want to give it a try you can go into the link in my bio it’s a little

workshop website you can go to the tools Tab and you’ll see a link to it

what we’re doing is we start off with the user prompt currently we’re using uh chat GPT IPOs GBP tool we then use that

to generate the brief which think of the brief it’s like a functional spec you can iterate on that you can change it

within take that and we generate a data model so one of the things we understood like we’ve had our figma plug-in that

can be converted into working either web or native for a while we understood that

designers and ux don’t always understand or find it hard to do things like the data modeling and then we take that

um we take that data model the brief and a UI kit that’s in figma that is a best

practices UI kit that’s responsive and everything and then we’re using AI to define the application or the app

definition which is everything about that application but we’re not using AI to generate the code so the benefit of

that is that once we put that into the Buzzy runtime we can then turn that into a website or into a react native

application on the fly so we’re not using chat EBT to do the actual coding

as such but we’re using it to describe the application and that just means that

as a somebody who’s creating the app you don’t have to maintain all that code you do need to pay us to keep the runtime

there but so it’s not for everybody but it’s massive speed of helping you know

if the goal that’s what I was saying before the goal is to help you climb to the top of the mountain

um some mountains are small we might get you all the way there some some mountains like let’s say if we’re going

to go to Kosciuszko maybe we’ll get you three quarters of the way there if you’re mounting that you need a prime is

Everest maybe we’ll get you to base camp but you better bring the Specialists and your

pickaxes and everything like that to get to the summit so and by that the

Specialists I mean that’s where you do need to bring in your developers and people to do stuff so it’s on a case-by-case basis so let’s let’s just

kind of sort of um look at this practically so I want to do an app that actually

shows me the locations of vivid David Sydney spots

how would I how would I start I would actually describe give me a map based application yeah I would start with that

prompt of you know create me an application Google and we should try this out

um and you know these are the things each of the location perhaps has an image

um it’s obviously got an address it might have times that that showing is going to be on

um you also might want to provide things like reviews so allow people to put in reviews and then that will be your initial

prompt and from that we’ll create that functional spec with a bridge but then you can add it and go on is that

literally going and looking up a bunch of templates that you’ve got or is it actually describing the data model and

you’re working backwards from the data model so so we were actually kind of Blown

Away by what chat GPT could do and understand that we asked it that really

it’s just clever prompt engineering under the covers so we started this journey we built the Buzzy Tech over

several years um and a couple of years ago we put the figma wrapper around it and then for us

because of the way that we just really trying to get to this data this app definition point that we can then turn

into something um we really looked at how we could use um the llm to generate help generate

that app definition so once we go that functional spec um

the the llm was like insanely good at defining a data model like scaryly good

so and you know it really surprised me I didn’t think it would like it sometimes gets stuff wrong and there’s subtleties

in in you know how you name things like in data relationships and sometimes it gets confused but in general was was

like you know I I did a Computing science degree and I can look at an application and break it into a data

model in my head but for a designer who wasn’t trained as such or doesn’t think that way that’s quite a challenge and it

could do an insane job of of generating ultimately the metadata so you can ask

it for a data model for a lot of different type of applications and you can say you know give this to me and

Json or SQL or whatever it is it does a pretty good job so James have you seen anything similar

around that’s that’s basically going super nuts can I say that

nobody knows I have no idea beginning to end yeah so you go right up to

publishing stage right yeah yeah and that’s because of that history you’ve had building it before

and so when you were building a you were building a an app creator platform then

at some stage you said okay let’s take a from figma to that and now you’re using the GB team to go yeah so we’ve always

been about how do we get like we started with life as like an ugly no code platform it was predominantly used by

business people for creating apps in a matter of minutes so there’s like ugly back-end apps that you could say Hey you

know you would go in and you would drag some Fields onto a screen and say I want to capture this data so people doing things like

um mapping applications for field reporting and stuff like that that you know you see the people wandering around

like reading meters and things like that um we did we started with life like actually doing things like sporting apps

and all stuff like that um so we’re already we what we were

creating was basically a platform that went all the way from concept through to

actual live deployment um so now what we did is we now allow you to generate an app and then that’s

deployed into like an Enterprise scale architecture so until kubernetes infrastructure that could be scaled

horizontally vertically um there that you know it’s all single tenants so your own database cluster

that again can be scaled separately you can break things into microservices um if need be so

um yeah so right like we’ve been doing a lot of work with people around and we show them what we’re doing and they said

they’ve seen people doing bits of this with AI um maybe you know generating the design

into Sigma component has been no way going all the way end to end and actually getting something running on

the cloud in a short period of time

yeah I guess yes and no um as Adam said I think there’s a lot of

people that have been doing bits and pieces of this for for quite some time

um and yeah I I yes I have to assume that there’s

probably a few um people out there who were as excited as you guys were like we’ve

been building this thing for however long and now suddenly llms are just going to make our life

so much easier like they augment what you’re even doing a really great sort of um

no pilot I guess to what you’re doing now that’s exactly like we’ve always just seen it as an accelerator

um like we’re seeing applications that are getting built with this that like I said if you’re climbing a big mountain

we’ll only get you part of the way there you need to bring in the heavy lifters and your developers to complete things

because no shying away from that I think we’re a long way from AI being able to

do absolutely everything you need somebody at the home who’s driving this thing and directing it and prompting it

and doing the coding bits to plug the gaps so I think that yeah so I think

that’s what we’re saying it’s not a it’s not one size fits all and you can’t do absolutely everything with it yeah I

think the probably the super interesting thing about all this to me is you mentioned

before how MailChimp changed um the

like exactly what a product person could do and then what they needed to hand

over them into a development to get them to take the reins from their MailChimp shifted that

um and over the years we’ve seen more and more products and I think from my maybe naive maybe not you but I

think for me this is just another one of those big shifts in that direction

um where I don’t think it’s gonna magically take away the Need For

Engineers overnight um but I think like what an MVP looks

like now is much different to what an MVP looks like five and ten 15 years ago because

of how much you can get done without needing to bring in Specialists so

and I think what’s radically changed too like not that I’ve made a study of MailChimp but I would thoroughly expect

them to move away from templates to to generative stuff as well too and I think notion seems to be doing is there

anybody a notion fanatic here the notion seemed to be bringing a bunch of stuff in really quickly which is unfortunate

for me because I’m still on Evernote but such is life yeah so yeah you see a lot

of augmentation that are actually in things Beyond just content creation it’s certainly around the design side of it

and so it seems like these guys are on the front end of that so what’s what’s really been super compelling for you

what are you seeing coming through as a developer and I’m going to ask I’m going to ask a bunch of questions in a second

just in regards to the landscape but um yeah what are you seeing that’s super cool for you as a senior developer

what’s super cool for somebody who’s uh you know a junior developer and what’s

cool for people who don’t go we already know we’ve already heard about it I’ll start with Junior I think for

junior developers it’s probably something that’s very exciting if you

are cautiously using these sort of tools I think

we all know chat jpt makes things up it’s just the nature of ROMs that’s

always going to happen I don’t think that’s going to change anytime soon um so

I think as as these models can take larger context windows and you can feed more real-time

data into it things like that they will become a little bit more truthy um but I think the danger to me for yeah

for a junior engineer or someone without a com site background or something is

you you can be led astray and it seems so legit what it’s saying

um so that’s one thing but I would absolutely recommend a junior to be used

in this sort of tools because they’re exciting and then as you go further and further I think to me it’s a big

multiplier on just how quickly I can get work done because instead of spending 10

minutes Googling um some part of a standard library that I use

I’ve been using for 10 years but I only have to use it every few months and kind of forget tiny little details so you go

to a quick Refresher um so instead of spending 10 minutes now I spent 10 seconds get it and then I

know it’s live because I’m like that’s that’s wrong yeah you can spot with hallucination exactly and I I think the

where the tools really really shine is sort of the more Scene you get probably

the more you could output from it because you don’t need to double check and you can tell it that it’s a bit

dirty layer yes right yeah all right so let’s let’s go to the next slide

the one that I oh not him again

all right okay so this is a guy this is a guy called Matt Welsh and uh you know he’s

he’s been pretty lazy he’s LED teams at Google and apple and he’s a professor or

was a professor at Harvard of computer science so you might know what he’s talking about I don’t know

um but anyway uh he says AI coding assistants such as co-pilots are only scratching the surface it seems totally

obvious to me that of course all programs in the future will be will ultimately be written by AIS with humans

relegated to at best a supervisory role okay so

um he’s I’ve included the link there to the original to the original text so it’s it’s not like it’s a puff piece or

just you know an opinion it’s literally a guy who sort of knows what he’s

talking about so have you guys kind of agree with that and you want to dare to make a bet on a timeline or any of those

sorts of things I won’t make it better on a timeline um but in I guess in many regards I kind of

agree um I just think that the timeline is maybe a little bit further than what people were assuming I think everyone’s

very excited about chat um but I think that

if you think about what he said there you know my the owner of the company from my first

job he started programming when with Punch Cards Right Punch Cards and for

him he went right up to he’s probably still programming today but but him it’s

just each thing is a new language and I think that natural language programming is just

another new language I’ll be at a very cool one and it makes it a lot more accessible to everyone but I think that

there’s always going to be people that don’t care about tech they just want to

live their life do other things and they don’t really care to do the programming side of it so

yeah well I agree with this but I also think that the supervisory role maybe is

not the best way of putting it I think that it’s just going to be a different type of programming

Adam might argue that we’re already you are already turning people into supervisors here it sounds yeah I don’t

know if I agree with that statement um I think like having played around with

you know the technology and sort of understanding that you do need humans in that supervisory role

I think one of the biggest challenges you know that I see is that specifically like with code generation and app

generation is when you need to debug something

it’s damn hard now if you’re looking at co-generation um if you generate

you know thousands of lines of code and then you need to enhance it without

understanding that code that’s a huge challenge that I don’t know how we’re going to solve that

now like I can I kind of look at it from a point of view have you ever had an argument with Siri

we’re going like okay please call David I’ll be calling Mary no David

oh what what time did you want to do that and no David called David and it’s

like I think that if you’re giving instructions to the AI ultimately to get to set a goal for it to create an

application at some point of time you have to have granular control over what that application is either to enhance it

or to debug it so I don’t know how that’s gonna like to say that everything is going to just be created by Ai and

you’re going to be able to issue some instruction doesn’t the instruction then just become the code

like like I think the coding level will go up but you’re still going to have to have a way to granularly control those

elements like you know maybe today you’ve got co-pilot and you know that’s going to help you to a certain degree

and it right so if we’re using copilot for our developers and it’s it’s awesome um you know we talked about the the

figma file becomes the code for us in our applications but that’s still the code

um and then if we take it up a level and if so we’re busy we’re adding a new feature at the moment that you don’t

need the figma file you just write the prompt and it creates the app for you how do you change that like oh no I just

want to move that button over there move the button to the left and it’s like you know at some point in time you really

just want to take that button and just move it there or there and change its color and and change it becomes cloning

so I think coding will change um I don’t know if all coding will be

done by AI um I do think this will need to be regulated um because I think there’s a lot of

other reasons for that but that’s kind of maybe I don’t fully understand where he’s coming from and he’s way smarter

than me but yeah but look I think regulation is a really

tricky topic and I’ve lived in to you know you can say at one level it’s

the cat’s out of the bag and you know I kind of look at it like knives and

nuclear weapons you know they awesome Tech um that can be used for a lot of good

and for a lot of bad um and there are rules out there that

say okay right you can’t carry a knife of this particular size and you know on your body or whatever bring it into a

class in school it’s got to be regulated so I think AI is incredibly powerful this is just my take by the way you

don’t have to agree with me but I thought he was going to talk about regulation of coders but yeah so I want

to and this is the kind of unfair unfun part of being a moderator of this

stuff particularly with the AI ones that I’ve done I want to keep it on topic just in terms of practical things because there is lots of side topics

that are super important but it’s being beyond the scope of this particular thing so I don’t know what what you’re

asking me but just what we’ll do is we’ll we’ll just stick on the code side of it for now if that’s okay so

um so Lex Friedman sits with Sam Alton

under two and a half hours that feels like three thousand hours and asks a question wouldn’t that mean that

there’ll be needs for much fewer programmers in the world and Auckland says I think the world is

going to find out that if you can have 10 times as much code at the same price I don’t know what that you’re saying

um you can you you can just use it to write even more code I think there is a supply issue and I was asked the

question today doesn’t that need more code bloat let us know expand it’s not about more bloat inside an existing code

but you can basically solve more problems on the planet before maybe we off the planet as well so just that that

kind of question there so throw it to you guys you agree disagree I’ll jump in I’ll listen to a podcast

recently um and there was an analogy about lawyers and there were there was a

village somewhere in the US and there was a lawyer and he was kind of not very busy and another lawyer moved into town

and all of a sudden things got really busy so I think the way I kind of look at that analogy is that everyone was suing

everybody and I think if the more ability that we give people to write applications there

is a finite resources of of coders that exist at this point of time I think the last time I checked was

something like 0.03 of the world population the more ability for people to create

applications the more requirement there are going to be for coders to come and fill the gaps I think

we get to do the really fun stuff and not the boring stuff that’s my take

mine is boring because it’s exactly the same as yours I think um yeah totally agree I think it just

this whole thing is going to just change how

it’ll kind of just free up people who are doing um work today in software development

that tomorrow will be done by AI it’s not that those people disappear and

have to go find new careers I think they just get to start solving more interesting problems um and they just kind of shift into I

mean there’s with every new technology there’s always going to be some limitation right and then the Builders

of that technology live sort of on the edge of those limitations and listen to

feedback from from their consumers think okay how do we improve this how do we expand upon all that sort of thing so I

think that that’s yeah I think that’s where it’s that’s the direction so I

kind of feel as though you two have kind of um contradicted yourselves so you said

before that the debugging’s hard but in fact at the moment the AI all that’s really

good at doing is generating code that it’s seen before and debugging is not part of that so we can assume that and

maybe it won’t be llms to do this but we’ll we so we’re seeing so Palm 2

down to from Google allows you to have your own instance allows you to absorb

as many tokens as you will say there are limits but we will get to a particular point with the more tokens could consume

your code base and then you could actually have have the AI doing maintenance so that’s

where I was getting to you know maintenance debugging that’s the boring stuff and you would hope that it’ll be

able to do that but you said okay well maintenance and debuggings are the hard bit so that’s the unfortunate boring

part of most development jobs particularly if you take over somebody else’s code um I’ll I’ll take this to start with and

then I’ll put it over but I think that in my mind right it’s I don’t think that

these tools aren’t going to be able to handle Depot and things like that I think that’s coming very very quickly to

be honest um I wouldn’t be surprised if um Pope pilot can handle some pretty

impressive refactoring and debugging issues by the end of this year

um but I think again going to like what’s the edge of these tools and their

limitations I think it just means that okay you’re no longer going to have to debug this one particular thing and the

tool will do that for you but there’s another issue over here the tool does not knowing how to debug that and

then that’s and I think that’s what I mean by we just will have to start solving more interesting problems and

new problems that we probably can’t even comprehend right now um yeah

yeah that kind of this is a complex topic um but I think

I kind of look it a bit like like a repo so if you’re coding and you use like hey

repo on GitHub or whatever it is you get it from um you kind of expect that code to work

and hopefully you’ve never had to debug it there’s a few times where I’ve had to get into the repos and debug stuff and

that’s really hard so I think like what I see what I was trying to say before is that

we kind of like what is doing is providing more and more out of books the similar concept to repos that you can

then just tie together with specific code that you will still need to debug and hopefully not have to understand

what’s in those repos I think there’s different ways starting to like react native plugins for your stuff yeah there

could be plugins there could be node.js or there could be whatever language that you know whatever whatever repository

that’s out there on GitHub or or in other that you use um it’s just like that thing is like a

black box that you hope Works per the spec and I think

what like whether what I was talking about in terms of the fun stuff and the debugging like I’d rather talk about

okay right how do we get that one that one that one side together with some code whether you’re using AI to help tie

that stuff together or you having to use your own um coding skills I think that’s where it’s going to get you know really

interesting um from a point of view of um there’s just so much capacity that we

have to understand a chunk of code and I

agree with you I think that that AIS as they increase the window size and everything will have more capacity but

the question is how do you tell it what to do how do you tell it that there’s actually a bug over there and how does it then

work that out I think there’s a huge challenge doesn’t it generate unit tests based on

[Music] testing but um I think that yeah like

I don’t know if I trust it this may be part of the part of the part

of the problem like I’ve seen like I went in and I’ve grown okay right go and create me an application for this in chat GPT and it’s going you know out

comes the application now write this for it and then I’ve gone okay is this code robust and he goes no here are five

things that it does that it needs to add in I said can you please add it in he goes no find a developer no really yeah

so and I’m sure it’ll get better at closing those games but I think it’s kind of like you get into the situation

where it’s just creating more and more code and then eventually we need to find a developer and you’re going oh my God

like there’s now 20 lines of code over here that I would never know when the world has ever seen except the AI yeah

and it doesn’t we don’t know the instruction to give it to tell it to fix that particular bug yeah it kind of gets

into an interesting conundrum okay great thanks next thanks best

oh okay so this is my proposal which is sort of to

do with you know if you’re going to write more code um then if you’ve got a if you’ve got

off a lot of offshore developers and if you if you do if you’ve done a lot of offshore assuring like I have over the

last 20 years you kiss you kiss a lot of frogs and most of them are frogs um every now and again you get lucky

um you know a thousand offshoring companies you know are always getting my email and and Linkedin and so on and so

forth they will be the big losers out of this because you’ll actually get a lot more done

um without without having to sort of go through the offshoring process until until you get to that kind of like base

camp issue or further halfway up the mountain issue am I completely wrong on

this statement I have I have some thoughts on this

um that feels like the intuitive answer

that they will lose out in this but actually who’s to say that

companies that run these um like offshore software houses choose to say that they’re not going to use

um or the generative AI tools to upskill their Engineers

um and end up producing some pretty amazing stuff exactly and not just that

but like communication which I know many companies that I’ve worked for have

tried offshoring at various sizes and one of the biggest problems was

communication lines miscommunication with specifications code Handover things

like that um if generated AI can solve a lot of those communication problems

maybe they’ll maybe these guys become winners um yeah just that I have no idea what’s

going to happen that’s just yeah I just thought maybe that’s an interesting point we did an interesting exercise recently

which is we were hiring for Dev and we basically said you kind of must used you must use some

sort of AI in the process you know we didn’t want anybody to sort of create a

an uneven landscape so we just said you know really you must do that but we’re still able to figure out who were the

who were the more intelligent developers because we’re asking them to it’s like the old open book exams in the sense

that you you I kind of that’s the right terminal but you know you’ve got all the enough information but how you apply

that particular Implement and so that’s again that comes down to offshores or Junior devs who are intelligent enough

or understand enough of the you know the molecular level stuff or the macro in order to apply it at an application via

anyway that’s just not