Bhanudas Tanaka Settle in for a fascinating voyage through time with our guest, Bhanudas Tanaka, who has experienced the transformative journey of computer programming from its early days right up to the AI-powered present. Let’s whisk you into an intimate conversation where Bhanudas recounts how he reignited his mathematician father’s passion for programming, a story that’s sure to inspire. 

But what does the future of programming hold? Not one to rest on past laurels, Bhanudas shares his insights on the seismic shifts brought about by AI in the programming landscape. Think of metadata-aware tools and AI agents that monitor the org continuously – the future of Salesforce development is here. Bhanudas details the minimum skills required to jump onto the AI bandwagon and the importance of a hands-on approach. He paints an exciting picture of the potential of GPT and Flow GPT in generating code and metadata, opening up a world of possibilities. Join us as we navigate the riveting evolution of programming and AI through the eyes of Bhanudas.

Show Highlights:

  • Discussion on Bhanudas’s professional programming experiences, including creating a shipment manager for a local shipping company.
  • Overview of the significant evolution of language models and AI over the past three decades, with a focus on the progression from text-based Eliza to current AI capabilities.
  • Insights on AI’s potential to revolutionize Salesforce development.
  • The future of AI in Salesforce and the potential of GPT and Flow GPT to generate code and metadata.
  • The changes in how we interact with technology, from searching in Google to prompting AI systems, and the possible future developments in this area.

Links:

Other Resources:

Resources Recommended By the Guest:

People to watch, follow, and digest:

  1. Eliezer Yudkowsky –  https://www.youtube.com/watch?v=Yd0yQ9yxSYY. 
  2. Also interviewed by Lex – https://www.youtube.com/watch?v=AaTRHFaaPG8
  3. Andrej Karpathy – https://www.youtube.com/watch?v=kCc8FmEb1nY
  4. Sam Altman – Lex interview – https://www.youtube.com/watch?v=L_Guz73e6fw

Content creators he loves:

  1. Two Minute Papers – https://www.youtube.com/@TwoMinutePaper
  2. Lex Fridman – amazing guests, very deep and thoughtful – https://www.youtube.com/@lexfridma
  3. Asianometry – tech + Asia, lots of history, well written – https://www.youtube.com/@Asianometry

Episode Transcript

Bhanudas Tanaka:
And by constructing those instructions, you could draw things like squares and rectangles and triangles, and you got your algorithms more structured, more elaborate, you drew roses with different colors. And so that was my very first experience with computers and also programming.

Julian Duque:
That’s Bhanudas Tanaka, Architect and Co-Founder at Terranox. I’m Julian Duque, your host for the Salesforce Developer Podcast, and here in the podcast, we share stories and insights from developers, for developers. Today, we’re going to talk with Bhanu about GPT for developers, but before, we’ll start just as we let off and we often do, with his early years.

Bhanudas Tanaka:
I was very lucky. My parents, we were kind of middle class and they devoted themselves to send me to a private school. And this was back in 1980, so I’m a little older, and we had a computer lab at this school. So when I was in first grades, it was 1981, we had IBM PCs and they had just come out, and there was a program called Logo.

Julian Duque:
Logo, yes, I remember Logo.

Bhanudas Tanaka:
With the turtle and there was a triangle. It was a turtle, RGB four-color, magenta, blue, whatever. And you gave it instructions: forward four pixels, right turn 90 degrees, forward eight pixels. And by constructing those instructions, you could draw things like squares and rectangles and triangles, and you got your algorithms more structured, more elaborate. You drew roses with different colors. And so that was my very first experience with computers and also programming. Back in those days, what were computers, right? There wasn’t a lot of things to do on them. So we started with light programming.

Julian Duque:
But you went directly to programming. That’s very, very lucky. My first experience, I remember using a DOS program called Banner to create banners or texts, 3D. I mean, imagine 3D text just DOS and using those printers like-

Bhanudas Tanaka:
The DOT matrix? Yeah, yeah.

Julian Duque:
The Dot matrix printers to print those banners. And I used those banners to create different messages in my room. It was nice, but I never had the opportunity as a kid to interact with programming language or a programming environment like Logo. I knew about Logo way, way after. Cool.
So you started very, very young. What happened next? So what was your evolution with computers or programming languages after that?

Bhanudas Tanaka:
I mean, I’ve always been fascinated with coding. I think I was born that way. The next memory that I have was sixth grade, going to see a movie called War Games.

Julian Duque:
War Games, oh yeah.

Bhanudas Tanaka:
And me and my buddy got home that night, it was a sleepover, and we we’re like, “Let’s spend all night. We’re going to stay up all night and we’re going to build War Games in BASIC.”

Julian Duque:
Wow.

Bhanudas Tanaka:
Because at that point, we were playing around with BASIC. Are you familiar with it?

Julian Duque:
Yes. Of course, of course.

Bhanudas Tanaka:
Yeah, yeah, yeah. With line numbers and when you think about-

Julian Duque:
Yeah, GOTO 10.

Bhanudas Tanaka:
Yeah, the things that we no longer have to deal with. We had a language that you had line numbers, and what happens when you have to insert code between line five and six? You have to now renumber your code. And if you’re using GOTOs, you’re now doubly in trouble.

Julian Duque:
Oh, yeah. Of course.

Bhanudas Tanaka:
So I started writing BASIC, but it was kind of just for fun. I remember we stayed up all night and we were able, in ASCII graphics, to print out the map of the world because that was, if you remember War Games, Whopper ran the nuclear war simulation, which had a map of the world. So we started there. We used ASCII text and blocks and colors, and that’s all we got. We had to go to sleep probably at six in the morning.

Julian Duque:
Wow, that’s amazing. And after that, you continue…

Bhanudas Tanaka:
Yeah, luckily my school had a good… Now we’re fast-forward, mid-’80s, we had a computer lab with the IBM 370s, I think, it was terminals. And that’s where I learned C. I think that was the first formal language. We also learned Pascal.

Julian Duque:
Oh, yeah. With Turbo Pascal.

Bhanudas Tanaka:
Yeah, Turbo. Borland Turbo, Pascal. Yeah.

Julian Duque:
Borland Turbo, Pascal, the blue screen. That was my first… Well, I dabbled with C as well, but I didn’t win after the typical had evolved. But with Pascal, it was my first experience. And a friend came to my house when I was 13 with a bunch of diskettes. “Let me install this in your computer.” He showed me how to write a function, the clear variables, do basic stuff, and I was not paying too much attention to it.
My dad, he was in his late 50s and he was getting sick. He passed away some years after that. But he used to program with PL/1 and those old-school technologies. And he saw me playing with Pascal and he asked me, “What are you doing?” “Oh, I’m playing with a programming language.” He sat down with me and asked me, “Okay, teach me. What are you doing?” So I went ahead and explained him the very basics that I learned from my friend.
I went out, my dad stayed on the computer playing with Pascal. When I came back that night, he was still there and he told me, “Come, sit with me.” And he built, because he was also mathematician, he built a program with a model to solve differential equations, like second-degree differential equations. And I was like, “Wow.” I mean this guy, my dad, he was sick and he hasn’t been programming for a while. And just one day with the logics he learned before, he was able to build something that worked.
For me, that was my, “This is powerful. I have to learn how to do this.” So that was my first real interaction. It was amazing.

Bhanudas Tanaka:
That’s beautiful on so many ways, right?

Julian Duque:
Yeah.

Bhanudas Tanaka:
And it’s rare. I haven’t met a lot of second-generation coders, so it’s pretty cool that your father was a coder.

Julian Duque:
Yeah. He used to tell me stories about mainframes, computers as big as a living room, and how he used to also program them with perforated cards and that stuff.

Bhanudas Tanaka:
Yeah, the card readers.

Julian Duque:
For me, that concept for me is still very, very far away.

Bhanudas Tanaka:
Right, right. But we’re doing it too, right? We talked about BASIC and line numbers. People listening are like, “What are they talking about? Line numbers?”

Julian Duque:
Yeah. I think we might say we are very lucky to be able to be exposed to these at a young age.

Bhanudas Tanaka:
Yeah.

Julian Duque:
When did that turn into your profession?

Bhanudas Tanaka:
Professional, I’d say right out of… After high school, I started to play with Microsoft Access.

Julian Duque:
Access?

Bhanudas Tanaka:
I found Access. So this is like 1994 now. I was at university and I started playing with Microsoft Access, and specifically with this new technology called Internet Information Server, ISS.

Julian Duque:
The ISS, yes, yes.

Bhanudas Tanaka:
1.0. You had these template files, you had these data files. And so I was fascinated with, I can now build a database and I can expose it on this thing called the web and it’s real time, and I can update the database here, but anyone in the world can see my updates. Today, that’s like a no-brainer.

Julian Duque:
Yeah, that’s huge now.

Bhanudas Tanaka:
But back in those days, we had discs and the closest we had to the internet back then was FidoNet, which was this distributed dial-up network back in the mid-’80s. That was one of the things we played with, which were bulletin board systems. And so you had this pretty intricate dial-up network that was global. You had these nodes that called each other every day and downloaded emails and messages and you basically could get a message around the world in two days with FidoNet. And so some of that capability was already there, but the idea that we could use MS Access and put data out on the web was just super cool.
So I can’t remember exactly how I got the job, but a local shipping company needed an update on their software, and so I just built them a shipment manager. In fact, that’s what it was called, Shipment Manager, which is the worst.

Julian Duque:
Super creative.

Bhanudas Tanaka:
I know. It’s just so creative. In Microsoft Access. And we did things like exposing tracking numbers to their customers on the web. And this is pretty advanced, right? FedEx didn’t even have any concept of this yet. It was fun. So that was the first professional work that I did.

Julian Duque:
And did you ever do anything besides working with computers or this was your thing since the beginning?

Bhanudas Tanaka:
I was an avid surfer also.

Julian Duque:
A surfer? Oh.

Bhanudas Tanaka:
In fact, my dad, I remember when I was 17, he lectured me on, “You’re going to college and you’re not going to be a pro surfer,” so I was really into surfing at the same time.

Julian Duque:
Oh. Do you still surf?

Bhanudas Tanaka:
I do occasionally. As you get older and more out of shape, it’s less enjoyable and I’m very picky now on the days that I go out.

Julian Duque:
Yeah. I might say I have a little bit of thalassophobia, so I’m afraid of the deep sea and I don’t know.

Bhanudas Tanaka:
The creatures. Well, I always tell people I have the same, but when you surf, you’re on the top, so you don’t see it really. Yeah, but when I go diving, it freaks me out a little bit. Yeah.

Julian Duque:
Oh, yeah, yeah. It’ll freak me out a lot for sure.
Who was that introduction to Salesforce for you?

Bhanudas Tanaka:
It was a nonprofit. So I was a customer first. I was hired as a technical executive to replace Razor’s Edge with Salesforce. And this was back in 2010ish, so it was back in the day when Nonprofit Starter Pack existed. It wasn’t the Success Pack yet. And me and my team implemented Salesforce. It was very painful. I learned a lot through the process, and mostly because I came as a developer. By that point, I was a full-stack dev, and so I remember my first experience with Salesforce was, “I can’t write code in production.” I just wanted to try the thing out. I was like, “I want to write eight packs.” And I’m like, “Wait, I can’t do it in prod. That’s interesting. I have to do this sandbox thing.”
So I really had to learn the ropes of Salesforce’s environment, which is, for those of us in the ecosystem that have been working with the platform, I always tell folks our knowledge of the limitations or the guardrails, the governor limits, how Salesforce works, is our value.

Julian Duque:
And this was how long ago? What year did you start with Salesforce?

Bhanudas Tanaka:
It was about 2010.

Julian Duque:
2010?

Bhanudas Tanaka:
Yeah.

Julian Duque:
This was a long time ago.

Bhanudas Tanaka:
It was a while ago, yeah. Everything was much smaller. We didn’t have Marketing Cloud, they had no Slack, no Quip. So I feel like I was in the, and people would probably argue who were here a decade before me when they first started, but I feel like it was the first wave. It still felt small.
And then entering into the mid-2015s, there was another, through all those acquisitions, I feel like Salesforce did another boom and it’s now huge in my opinion. We were joking earlier that no one knows the whole system.

Julian Duque:
No, it’s impossible.

Bhanudas Tanaka:
It is.

Julian Duque:
It’s massive. And have you done full-time mostly Salesforce or there have been some other technologies that you have been using?

Bhanudas Tanaka:
In the last 13 years, I think we dabbled with a bit of Drupal and the normal stack that you find in nonprofits that connect to Salesforce, but I would never focus myself there. I’m a full supporter of putting everything on platform. That’s why you pay a lot for it. Use it.

Julian Duque:
Yep, totally. That’s great. So let’s talk about what we came to talk about, which is AI and this new boom, the generative AIs, GPT, all of these revolutions we’re living right now.
How you started with AI, how was that first experience for you?

Bhanudas Tanaka:
All my life I wanted to learn Lisp, which I never did.

Julian Duque:
Oh, Lisp.

Bhanudas Tanaka:
So I think in the back of my mind or somewhere inside of me, I’ve always wanted to play with this. The first time that I saw the machine interacting like a human was with ELIZA. There was a, I don’t know what you would call it, it was a text-based thing called ELIZA that was I guess built in the ’60s and it made its way to PC. I think it was ported to BASIC and we had a copy of it and we would play with it and it would start up and it says, I think she would say, “Elucidate your thoughts.”
That was her opening prompt to you, which I think was interesting is ELIZA prompted you versus all our other, the modern interactions, we prompt the machine, but I’ve thought about that a lot lately. What if they start prompting me? Like my Google device, when I go home, it recognizes me. So it actually prompts me when I walk into the kitchen and it hasn’t seen me yet for that day. It’ll say, “Good morning, Bhan.” So maybe it is prompting me, but ELIZA was the first time, and this was back in the ’80s, so it was very rudimentary and I can’t remember how accurate it was, but it was amusing. It was the first time we thought, “Okay, it’s talking to us.”
And in those days, we were having some really low-quality audio. If you remember, there was some voice or text-to-voice, but it was pretty horrible. So they started to become a little more human, I guess, or more human.

Julian Duque:
Yeah, more humanized. Right now, when I see where this technology is going, I get really fast into the uncanny valley because these videos or images that are generated by the AI or the text-to-speech, now it’s starting to get difficult to recognize if it’s generated by an AI or if it’s real. We are getting into very interesting territory now.

Bhanudas Tanaka:
I mean, the whole term AI, I do appreciate that, at least in certain circles, we’re calling it GPT or LLMs. I think AI is such a general term. It has the word intelligence in it and I don’t know if we’re there yet. I think it has modeled human, human language. I think it’s even good with foreign languages also, not just English. It has indexed a lot of other languages.

Julian Duque:
I have interacted in Spanish with it and the results are amazing.

Bhanudas Tanaka:
Really?

Julian Duque:
And I can even prompt it to use certain specific dialects of Spanish because it’s very different. The Spanish that we speak in Colombia is different than the Spanish that folks in Argentina speak. And if I prompt ChatGPT, “Okay, tell me something in Colombian Spanish,” it does a pretty good job, which is incredible.

Bhanudas Tanaka:
Interesting. I wonder if it could go back in time too. So some of the common prompting patterns, as a thing, “As an Argentinian, back in 1750, write this.”

Julian Duque:
Wow.

Bhanudas Tanaka:
I’ve never thought of this, but it’d be interesting to find out.

Julian Duque:
Yeah, because language evolves as well.

Bhanudas Tanaka:
It does, it does. Right.

Julian Duque:
That’s an interesting experiment to do.

Bhanudas Tanaka:
Yeah, yeah.

Julian Duque:
Do you use GPT in your day-to-day job?

Bhanudas Tanaka:
I do.

Julian Duque:
Or do you do it just for fun?

Bhanudas Tanaka:
I do it both. I’m currently very fascinated by it. I think the fun parts of it is testing it. I think currently, what people do a lot of, they’re like, “Oh, does the AI do this?” And it is like, “Just try it.”
Last night, we were having a debate about something and so we fired up ChatGPT, we fired up Anthropic, and then we fired up Bard, and we just compared what the results were, and we found that GPT was very conservative. It kind of took offense to what we were asking, and then Anthropic was just straight-up. It just answered the question. And then Bard was just a little confusing. Well, not Bard. I’m sorry. It’s Google Search with AI or whatever that is.
So I think the fun part is playing. For me, also the fun part is consuming a lot of YouTubes. There’s a guy called Two Minute Papers. I love Two Minute Papers because what he does is he condenses academic papers and he does it, not in only two minutes, it’s a catchy name, he’s normally six to eight minutes but-

Julian Duque:
Still, it’s short. Yeah, yeah.

Bhanudas Tanaka:
It’s still short. It’s still short, right. He consumes it for us. He’s a PhD and he brings it down in layman’s terms and he has great graphics and I really point to Two Minute Papers as a YouTube channel that spurred my interest in machine learning and what was coming out because all of this, what we see today, started a decade ago and has been evolving, like the introduction of GPT itself and the development of these large language models really allows us to get where we are today.
And those are… And I say inventions, I mean neural networks have been around for a very long time. I remember when I was a teenager at a science fair, someone did a project on neural networks. That’s 1990 and that’s underpinning our language models. So consuming YouTubes, that’s my form of entertainment and it tends to be all along AI and learning about what’s new because literally every day, there’s something new, there’s a new development, new news.

Julian Duque:
Yeah. I have seen a lot of sentiment of FOMO, fear of missing out, from people. They want to get into these LLMs, AI, machine learning, but they get this kind of challenge because you think that you might need to know a lot of math and science and it’s going to be difficult and you don’t even try to get into that domain and you don’t do anything, but to start experimenting with these technologies like GPT, you don’t need to understand how they work. If you want to maybe of course do a research or build one or implement one, sure. But to use it, you don’t have to get into that.
What’s the skill, the minimum skill that we need to have to start using technologies like GPT?

Bhanudas Tanaka:
Yeah, that’s a great question. I can approach that from two directions. One, my own experience of being able to engineer prompts and get it to do what I want it to do. That’s based on being in technology for 30 years and coding for, I guess if I do the math, 40, 42 years I’ve been writing code. I’ve been making machines do stuff. So I have an innate way of formulating, “I want you to do this,” but with specific words, specific phrasing.
So I think one of the things that folks should learn is how to engineer prompts because if you are more of a junior and new to this work, it’s not immediately clear how to make the machine do what you want it to do because you actually don’t know how to think through the steps. You still have to make the machine go through steps, or let’s just say GPT, to produce something. You can’t just give it a very generic thing and have it get it right and there’s a lot of tuning. There’s a lot of looping.
So I think for beginners, for people who are interested in it, using specific prompts, going to YouTube, getting research, there’s a lot of classes now online that will teach you how to do prompt engineering, the same way that we had to learn how to search with Google. We phrased things differently with Google, it’s not even natural how we’ve, over the last 20-plus years, to get results.
So I think over time, it’s just going to become second nature, and I really think the current version of GPT is kind of search on steroids. It’s search-plus. It gives you results of things. So say you’re looking for a recipe, you don’t have to search for a specific recipe on Google. You can give GPT a list of ingredients and it can kind of combine that altogether and make something for you. Or if you did the reverse, you’re like, “Okay, I want to make shepherd’s pie,” you would say something like, “As a James Beard Award-winning chef, make me the best shepherd’s pie, give me the ingredients, list out the tools that I need, and then give me instruction with precise timing and instruction on how to make this dish.”

Julian Duque:
Wow. One thing that I have seen that is very interesting is folks that are using GPT for meal planning, like, “Okay, I need to eat this amount of calories a day, give me a week meal planning and give me recipes.” And they’re even losing weight with these technologies. The amount of use cases that I have seen is incredible to me.

Bhanudas Tanaka:
Yeah, yeah. That actually makes me think I should ask it now. I need to lose some weight.

Julian Duque:
Yeah, me too!

Bhanudas Tanaka:
“Create me a weekly meal plan with these types of food.” That could be really interesting.

Julian Duque:
Yeah. So I’m using it at work of course, for my hobbies, but I think I still need to use it more for some other use cases.
Do you think… Well, we were talking about how to search in Google. For me, searching in Google is a basic skill now. If you are working with computers, like most of the people do, you have a good advantage if you know how to search. Do you think these GPT or prompting, because obviously other technologies are available and they’re going to be available, but prompting is becoming or will become a basic skill?

Bhanudas Tanaka:
For now, I think it is becoming a basic skill. And I say for now because in my limited experience, I feel like we’re at the start of something different in computing and we don’t know where that’s going to go yet. I’m sure there’s many smart people who have a vision of that, but for me, I don’t know what that’s going to look like.
I do see how it’s moving towards, if you watch Star Trek for instance, and you watch how engineering interacts with the machine, with the Enterprise, it’s like, “Status check,” and then it talks back, versus how we do it, which is we go and look at governor limits and we maybe type a little bit of Apex to see where we’re at or there’s a dashboard. It’s very clunky how we interact. So I envision that in the future, natural language will play a role, but maybe it’s just going to be how we speak. You know?

Julian Duque:
And this will change the way how we interact with computers, like user interface. I think this is going to be a new area of a study for UX engineers, for example. How can they start building interfaces or building experiences that rely on AI or GPT or these technologies to improve how to interact with a computer? It’s very science fiction for me, like these movies where people are just waving hands and controlling computers and just talking and they do what you say, it’s becoming a reality super fast. I think we’re in the middle of a paradigm shift.

Bhanudas Tanaka:
Oh, absolutely. I accidentally came upon smart devices at home over the past years. So my entire house has smart devices in every room and all the lights are controlled by those devices, our doors are controlled by the devices. So when you go into a room, the actual light switches don’t work. And my wife initially pushed back.
But what it has taught us is how do you prompt a machine? Because at first, it feels weird. You feel like you’re commanding, but that’s how it works. I walk into a room and I say, “Turn on lights,” or I can, from another room say, “Turn off all the lights in the kitchen.” So there’s a way that I don’t have to physically push. I have to think and say it, and I could be in any room and think, “Oh, I left the front door unlocked. Lock the front door.” So it gives you this flexibility that breaks down the physical boundaries.
And the reason we did it initially was because we were renting and if anyone who has rented, your lighting is always horrible. So we replaced, we would basically just have lamps and things, but then you have the problem of the switches. So we then had these Phillips Hue switches all over, but then eventually, we just went with smart devices.

Julian Duque:
Nice. I love it. As a telecommunications engineer, I remember when I was back in college, they were imagining the future. So these telecommunications companies, we are going to live in a connected future with the smart devices and they draw a very beautiful picture of… And it’s very accurate what we’re experiencing right now.
I have smart devices at home. I love that technology, but at the same time, I think about the privacy and security concerns behind that because obviously somebody that’s smarter than the builders of those technologies or more smarter than me can break into the security of those devices. We have cameras to monitor our pets. Who might be watching us? We don’t know. It’s fascinating that we have these advances, but also make us be more aware about the potential security and privacy concerns that are behind.
For you, what are those concerns around technologies like GPT? Do you feel positive that the future is going to be amazing or are you a little bit scared like some people with this?

Bhanudas Tanaka:
I once had a friend who described me as, I think he would call it frolicking in the daisies. It’s my personality to not worry about certain things, which I think is a character flaw of mine.
Yeah, I hear all the points and I think they’re all valid. I think at a macro level, I get concerned. What is the data being used for? Are we just being monetized? The whole model of freemium, I think we’ve realized as a society, doesn’t work because the humans become the product. But beyond that, I am mostly just excited to see where we’re going to go next.

Julian Duque:
Yes. I’ve heard some other podcasts and interviews. People try to predict what’s coming next. How do you imagine GPT or these technologies are going to be in the next year or five years? Because in just a couple of months, we have seen an exponential growth in usage. So for you, how is the immediate future for these, like next year?

Bhanudas Tanaka:
Yeah. I hope in a year, we get access to models that are trained specifically on the platform, Salesforce platform, because currently we’re using models that have been shared with other languages so there’re, at times, a lot of cross-errors basically. Apex looks like Java, so sometimes it responds with methods that we don’t have that only exist in Java, and I think something that is trained directly on all of our Apex and all of our config is going to be really smart. I mean, think about the reinforcement learning of every time we compile, it’s saying whether we were right or wrong.
There’s a new, new as in the last week, released publicly, it’s an OpenAI tool called Code Interpreter. And what that does is it builds Python code to do the thing you want it to do. It will also create a test class and if the tests fail, it will continue to adjust the code until the tests pass automatically. So now we’re seeing AI fold on itself.
I think in the next year, we’re going to see a lot more of that. AI’s running in loops on itself, multiple AIs working together to verify, and really on the Salesforce side, specifically the Salesforce developers, all we’re missing is that thing to compile. Maybe this is our interim product. My assumption is Salesforce is going to do this, so that’s why I’m not rushing out to go build this tool. I assume Salesforce is building it because it’s a no-brainer.

Julian Duque:
Yeah, that’s what we are aiming with Einstein GPT, and the whole other Service GPT, Cloud GPT, and not only on the user side of things because for users, they’re going to be a very, very great, a step over what we are doing today, but also for developers and there will be some developer tooling that are coming that it definitely will improve that workflow.
I’m dreaming about this co-generation tools that are org-aware so they know about your org, your metadata, the other classes, so they can easily recommend you how to call other things that you already have on your org so that they learn, train the model with your data. I think that’s going to be very, very powerful.

Bhanudas Tanaka:
For it to be metadata-aware.

Julian Duque:
Yes, exactly.

Bhanudas Tanaka:
So you can speak in generic terms about your account object and it knows what fields you have, what the types are, even of validation rules. Then you can start to think, well, you could probably write test harnesses that are AI-based, AI agents that will continuously monitor and run things against your org and develop new ways of breaking it because it can just keep on running. Right?

Julian Duque:
Yep. But if you’re writing a lightning web component, it’ll be aware about Apex classes that already-

Bhanudas Tanaka:
It can do that. GitHub Copilot, if you have multiple, and I’ve done this… So my normal workflow is I will, if I’m starting something brand new, I’ll use ChatGPT. I think it’s good at just fleshing out boilerplate and just creating a class file. Then I will put it into GitHub Copilot because that’s in VS Code, now I’m working in the org, so I can push that into the org.
Then Copilot will be aware of code if the window is open. So in this scenario, I will create the Apex class and keep that open, then I’ll create a shell of an LWC, and then I’ll start commenting out in lines what I want it to do. So specifically, in the JavaScript control of the LWC, I can say, “Invoke this Apex method,” and then it will build the header. It’s an include, right, I think?

Julian Duque:
Yeah, it’ll import.

Bhanudas Tanaka:
Yeah, the import, with the decorator. So it will import that Apex class. It knows about it because it’s open and I’ve tested this and it has it correctly. Now, the part that didn’t work was passing variables back and forth.

Julian Duque:
The binding, yes.

Bhanudas Tanaka:
Yeah, it broke down there. So I was slightly frustrated. I think the end result, I tried to build a very simple LWC using just purely GPT technology, and I would say I got 50% of the way there without doing anything, and it compiled, but it didn’t work. Nothing happened.

Julian Duque:
But still, getting the 50% is…

Bhanudas Tanaka:
It saved me 20 minutes of work and it was fun. I think that’s the thing that I want people to experience. I often tell folks, every day, it gives me a moment of like, “Whoa, that was cool.”

Julian Duque:
Yeah, I do a lot of live coding on our YouTube channel, and most of that is using JavaScript, and Copilot is very smart with JavaScript, so I sometimes don’t worry about it, and I just let it auto complete and it does a really great job. So I can’t wait until we have that highly trained in Apex and LWC and also FlowGPT that is also coming. It’ll be a good near future.

Bhanudas Tanaka:
For sure, yeah. What will be interesting is will it become heavier? And I just thought about this when you mentioned Flow. So as a coding developer, I can write a trigger or some update really fast, and if I go to Flow, it feels cumbersome. I have to point and click, I have to do all those things, and I wonder if at a certain point, would prompt engineering, like say in Flow, feel heavier than doing the flow because you’re step-by-step instructing the machine fetch records, now loop on that, or whether FlowGPT would be so higher level that you could instruct it and it would generate code just like how it works today?

Julian Duque:
I think that’s the way, yes.

Bhanudas Tanaka:
They’re generating metadata, yeah. That’s another thing to test. I think it was Copilot. One of the, maybe it was GPT, can generate metadata. I had it build me an object in metadata and it did it. So that was another realm of when you’re creating a brand new org and you have to build all these objects, it’s cumbersome to point and click, but if you could just train it with natural language and just elaborate on it. In fact, you could probably even just take the requirements document and throw it in there. Now, that’s interesting, right?

Julian Duque:
Yeah. Give me an object with this field, these relationships, and I have the metadata. [inaudible]

Bhanudas Tanaka:
You train your business analysts to write requirements that are clear enough that you could put back into the AI, which will then just build the thing. I love that because it reinforces documentation.
The one cool side effect of Copilot being the way it is that I prompt it with comments. So now my code has a ton of comments.

Julian Duque:
It’s documented, yeah.

Bhanudas Tanaka:
And it also does the inverse. You can highlight code and say, “Write my comment.” But I actually think the natural way to do it is you’re writing your comments because those are your instructions to tell GPT or Copilot to do its next steps.

Julian Duque:
And then you iterate and fix of course. Beautiful. This is fascinating, and for me, I’m excited about what we can do today, so I can’t imagine what we can do tomorrow with this. This has been one of the recent technologies that has me hyped, really, really hyped.

Bhanudas Tanaka:
Yeah, I agree. Yeah. I will spend probably eight to 10 hours consuming content about this world. And to your point earlier, I’m not a mathematician. I don’t know Lips, I don’t know any of the underpinning technologies, but to me that’s similar to all of us in the ecosystem. I’ve never seen the code of Salesforce. I heard that you can actually run it on a laptop, which I think is fascinating, but I don’t know anything about it. I don’t know anything about its architecture, underlying from what we are exposed to. So it’s very similar. It’s the age of platforms.
Us in the knowledge, I guess the knowledge workforce, we’ve already been trained in a way. I think some of the questions that we get from people about AI is like, “Well, things are going to be different and junior people will no longer have jobs,” but the reality is we’re already doing it. And what Salesforce has done is opened up technology work to a whole separate set of people. Folks who don’t have traditional computer science degrees are building gigantic solutions on a tech platform that’s based in technology.
So the exciting thing is what’s next? Maybe English majors are going to take over technology next because they’re just good with words, better than computer science people. I don’t know.

Julian Duque:
Exactly. Those have the right skills now.

Bhanudas Tanaka:
Right. But I guess the story of that is we’ve already been through a shift like this. Yeah, again, the daisies, but we don’t have to be that worried I think. We don’t know what’s next.

Julian Duque:
Yeah. Folks asking about the same thing, “Aren’t you worried? This is going to take over our jobs,” and it is not the AI that is taking over our jobs; it’s the people that know how to use the AI that are going to take over our jobs. So our task is to learn these AI technologies, how to work with them and use it.

Bhanudas Tanaka:
Yeah, yeah. And hopefully if you’re heading in this direction, you enjoy the work. I think we are lucky. We’ve plugged a passion into a way to have a job or to make a living, and I hope that for everyone to find their passion, and if you can stick with it and hopefully you can make a living from it, you’ll be great at it.

Julian Duque:
Totally. Totally true. Speaking of passion, to change topics a little bit, besides surfing, what are your other hobbies? What do you like to do? What’s that thing that keeps you also alive?

Bhanudas Tanaka:
Golf.

Julian Duque:
Golf?

Bhanudas Tanaka:
Golf is actually my current hobby.

Julian Duque:
Wow.

Bhanudas Tanaka:
Yeah, I forgot to mention that. I don’t know why. I golfed yesterday actually, and I’m golfing tomorrow. I travel with my clubs and it’s just the greatest thing. For me, you can do it for a very long time. It’s low-impact generally. My father’s 77, we golf once a week together. He sometimes still beats me. It’s great. And you never master it. It’s kind of like technology because technology is always changing, so we never master it. It’s the same techniques, the same patterns, but you’ll never master it because there’s so many variables that are at play.

Julian Duque:
Beautiful. I will never guess, coming from this extreme activity, which is surf, to this peaceful one, which is golf. Nice. Awesome.
Well, Bhanu, thank you very much for joining us on this podcast, and I hope all the folks that are listening to us start getting into this AI journey as well.

Bhanudas Tanaka:
Thank you, Julian. It’s been great.

Julian Duque:
Of course.
And that’s all. If you want to learn more about this show, head on to developer.salesforce.com/podcast where you can hear all the episodes and read the show notes. Thank you, everybody, and talk to you the next time.

 Get notified of new episodes with the new Salesforce Developers Slack app.