Demystifying AI Engineering with Tejas Kumar – Techstrong Unplugged EP44

Welcome back to Textron Unplugged. My name is Cassandra Chin, and today we have Tejas Kumar. Hi, Can you introduce yourself?

Yeah, I'm Tejas Kumar. I've been building on the web for over 20 years of places like Versal and Spotify and Zeta and more. And today I spoke at info web shift.

We talked about AI engineering, what AI engineering even is, how it, how it exists today, what problems it has and how it might evolve tomorrow. Do you wanna talk a little bit about your talk and like the AI engineering? Yeah.

I think one thing that people tend to not fully understand is what an AI engineer even is. And I've asked so many people in preparing to come here, like what, um, do they think they could be AI engineers? What is AI engineering?

And a lot of folks think they don't, they feel imposter syndrome and they go, I can't do this because I don't know, machine learning, I don't know Python, I haven't gone to university. And the truth is, AI engineering has nothing to do with that. It's, it's entirely different.

It's talking to an API and getting a response and using that to solve a problem. Um, and the AI side of that is you're interacting via an API with large language models or some other model. Right?

Um, and I think that's something that people don't know, but once you tell them, like if they can just like speak to something over HTTP, get an answer and then use it, that's all it takes to be an AI engineer in a, in the most simplistic terms, they go like, oh wow, I could now, and it eliminates some of that imposter syndrome. Have there always been AI engineers? That's such a good question.

Um, I think so, yeah. Depending on the kind of ai, like previously we had games like Pacman where there's AI in there, right? Like the ghosts are trying to chase you, and when you eat the little cherries, then the ghosts turn blue and like they, they avoid you.

That's ai. But it's rule-based ai. So programmers knew the rules of the game and they wrote algorithms to enact them.

Some would call that ai. So there have always been AI engineers, but it's evolved over time. The AI engineering we're talking about today is to do with generative ai, which usually includes generating something, images, video, text, and generative models like um, GPD three, gp P four, et cetera, these models.

Is there really a big difference between like a software developer and an AI engineer? It's a great question. I think there one is general and one is specific.

There's so many millions and millions, even billions probably of software developers in the world. But a fraction of that would be front-end developers, right? Because there's backend developers, there's um, DevOps engineers is specializations.

So software developers broad and there's specializations inside. And so AI engineer, I think of it as like a single specialization within software developers. So Would using copilot qualify you as an AI engineer?

I don't think so. In, in this case, you're a developer assisted by ai, but you're not engineering anything that uses a large language model. So you're, in this case, you're not an AI engineer, you're an AI user.

'cause you're using it, you're not making anything, you're not what is engineering. It's problem solving through the application of science. So you're not actually solving a problem with the code.

You know, you're not writing, in fact the code is solving your problem. So yeah, you're more a user than an engineer. That's an interesting perspective that like developers are now users.

Yeah, definitely. But the cool thing is we don't have to be like, it's a, I was speaking with someone from Microsoft earlier and we were exploring the question like, does, does ai, will AI replace engineers? You know, um, and I just, I don't think that's ever possible.

'cause like we can, we come up with the ideas and then we can build, we can build them. And then when it comes to actually shipping them, like there's always a review process. Even now there's code review and at some point we may have large language models and AI that can review its own pull requests and stuff.

But I feel like you always need human beings, you know? So like all that to say, yeah, we've become users, but we still build and in fact it accelerates the way we build. Like we're still a creative aspect, which humans have.

Exactly. And I think that's kind of beautiful 'cause humans are just always irrational. And I think as AI continues to grow, it's so interesting.

We often like have a dislike of AI generated content. I, I don't know that I do, but when I talk to a lot of folks and I show them a picture and I'm like, Hey, this picture was generated with ai, they immediately like it less, they ascribe less value to it. And I think we do that also for the code written.

And so how can we replace the human element that way, you know? That is interesting. Yeah.

Like how people view AI generated pictures differently. Do you view AI generated things less, um, valuably as you do human generated thing? Personally I do.

Why is that? Um, I think I value someone's time they put into it. Not just their time, but like it's their own creative process.

Yeah. Yeah. I think we need a word for that.

I don't think, I think we don't have it, but when somebody puts in time, effort and, and attention, what is that called? You know, we don't have a word for that, but that's the differentiator between AI made stuff and human made stuff. Yeah.

Yeah. I, I don't know if I agree. Um, I once got in trouble for suggesting a future that people didn't want.

I said, imagine the year 2050 where instead of like wa waiting for the new Marvel movie to come out where it's the Avengers, whatever, you come back home from work, you sit down and you just write, I wanna watch a movie with Spider-Man and the Hulk and Batman enter. And then it's just made for you, like instantly. And it's feature length, it's two and a half hours, it tells a great story.

And then tomorrow you come back from, from work and you're right, I want a sequel. And it just like exists and it's perfect. You can't tell the difference.

I I was just talking about this with some friends. We were doing a thought experiment and they were like, I would hate that. And I was like, why?

At the end of the day you're getting your entertainment. You know, like what's the problem? The problem is humans aren't involved.

But why does that matter? Like, there's a lot of, and this is something people know, there's a lot of stuff on the web today that looks human generated but is not like accounts on Twitter, accounts on Reddit. If you post on Reddit and you get a reply to your comment, there's a high chance this is a bot.

You just don't, it just doesn't look like a bot on the topic Of generating movies to watch. Yeah. Would you change your perspective if you were generating a cartoon for your kids to watch?

That's a good question. Like maybe we'd be a bit more picky if our young ones like that's their education. Yeah.

Yeah. I don't, I might change, I might like it if, if, because then I can choose what to show my kids, you know, maybe I don't want them to watch. Um, Rick and Morty, imagine like, they just like find, they're like, oh, this thing looks like a cartoon, but it's Rick and Morty.

I would not want children to watch that. So I think another topic you're passionate about is health. Yeah.

Yeah, totally. Um, I come from, I'm just not a healthy person. I mean, I, I am now more, but um, I've got a, a disease right where I, I don't know, um, if you know the term hemophilia, I never stop bleeding once I start either internally or externally or whatever.

And so a lot of the reason I'm alive today literally is because of tech, because of, so, because of web engineering, what I do. Um, because as a, as a kid to do anything to like, to lift a heavy object or to walk around with a backpack, to go up some stairs to go to the bathroom, all of this would cause me to bleed internally. And then I wouldn't stop.

And at some point I would pass out 'cause I have a lot of blood loss. The only thing I could do was this without like harm. And so I started by playing computer games, but I got bored and I was like, how do they make these?

And then I learned that and I slowly just learned a bunch of things about code until I got really good at it. Um, and then moved, literally got a job by learning how to code in Germany and moved to Germany. I I live in Germany now and there's a great healthcare system that takes care of me.

And I'm like living my best life. When previously I was living my worst life, it was the treatment model was, if you're dying, they'll save your life, but if not for free, if not, you have to pay. And the bills were like $12,000 a week, A week.

And so, um, it was, it was crazy 'cause like I'd wait until I'm in critical condition, go to the emergency, they saved my life and go back. It was lots of trauma for like 20 years. But then I moved to Germany by learning how to code, which is so cool.

'cause there's so many young people who just don't know what they wanna do with their life. I, and, and now, like I'm, I'm 31 years old, I've been doing this a long time. I get the occasional young person will come and ask me like, Hey, I'm finishing high school.

I'm a senior. What should I do? And ask them, what do you like?

And there's no answer. There's like, I don't know, I management, like, it's very confused. And I have never had that, 'cause I've just been writing code since I was eight years old.

It's the only thing I could do. And in so many ways it like saved me from this. Like, I don't know what I'm gonna do.

You know? And so that was before. And then I moved to Germany and I have this healthcare.

And so now I'm like, what else can I do now that my health is taken care of? How can I optimize it and maximize it further? And so that gives rise to a whole like different discussion on nutrition and sleep and supplementation and food and exercise.

And so that's outside of tech. I spend most of my time there because it's been a story of health from the beginning. Uh, can you describe more what you mean by you focus on like sleep and nutrition?

Yeah, I think as developers there's oftentimes a pressure that we give into. Um, I've been a developer for many years, right? And there's usually a performance anxiety.

Like I'm not doing enough. That comes from, I don't know, where either the company or the individual. Then there's this pressure to I need to ship something.

There's performance reviews. And when you tie all of this together, we're incentivized to ship stuff. And sometimes we won't stop.

At least with me, I don't stop until something is shipped. And in the age of remote work, like if I'm working from home, I'll start in the morning and then I'll just continue until the problem is solved. And we tend to underestimate problems.

So, oh yeah, I could just build this. No problem. Next thing you know, like it's three days have passed and you've been in the same spot.

Um, we often, especially 'cause it's so easy to get those rewards by writing code, it compiles Oh wow. And you just keep chasing those keyboard only rewards that you don't get to pursue like physical real life reward by like running the fastest mile you ever ran. And so we're incentivized to neglect our health by our, either by our organizations or by our own, like reward pursuit behavior.

Like, I'm just gonna keep building this. Wow. It's, the feedback loop with code is so fast that you're like, oh, I'm just gonna, and you get stuck in there.

And um, I feel like this is the typical developer look around, right? There's people who don't look the way they want and aren't as strong as they wanna be. And that's just the reality for us.

And so I spend a considerable amount of time looking at how can I optimize this? And that comes down to, um, being very mindful with the balance between work and the rest of my life. And so for me, I tend to work very well in like, highly focused bouts that have a time cap.

So I'll do like an hour and a half to two hours of like maximum focus. Like you can't imagine I'll leave my phone somewhere and I'm just like locked in. Um, and that's my deeply focused work for the day.

That's it. Just an hour and a half, two hours. And I call this offensive work 'cause I'm moving the needle, I'm doing the thing.

And then the rest of the day is defensive work, meaning there's all kinds of stuff flying around. There's messages from Slack and Discord and people are tweeting at me and people wanna do interviews. Just all this stuff that shows up in your day that is defensive work.

And a good combination of this is very helpful for me. So when I finish that intense work bout in my mind I'm like, great, I've moved the needle. When I finish the defensive work, I'm like, great, I've served a bunch of people.

And that just automatically, it's so beautiful. That automatically makes room for the rest of life. How do you know when to put a stop to defensive work?

That's a very good question. Um, I don't, and so I try my best to, um, avoid more defensive war. And sometimes I've gotten very good at this recently, but previously I was notoriously bad at like just closing the computer and putting it away.

But now I think, you know, what happened, I think making way for offensive work has taught me because when I finish offensive work, I just literally close everything I'm done for the day. And I think continually closing things for offensive work has reinforced the habit of make getting better at closing things for defensive work as well. One thing that puts me at ease for defensive work is Slack.

The chat app has this feature called catch up, but only on the iPhone, only on the mobile phone app, which is really awesome. It's like Tinder. So like it shows you cards of like the latest messages and you can swipe left to keep them unread until you come back.

Or you can swipe right to say, this doesn't matter. And this, I use this religiously for protecting against defensive work. Even right now, I could open my slack and open catch up and I I will just like, I need this later, I need this later, I need this later.

That doesn't matter. That doesn't matter. And then when I, when it's time to do offensive work or when it's time to start the defensive work, I'm like, look, it's all this nicely curated list that passed me curated for right now.

You know? And so a great help for buffering defensive work, uh, and stopping is to say, great, I'm just gonna use the catch up feature and come back to this later. I think what's an interesting approach, it's like saving it for later.

It's very important because if we don't, and this is so I think there's a lot of synthetic pressure that tells us we need to respond to this thing. Now, that's never true unless you're like an airplane pilot and you need to start descending or something. For us, we put rectangles on screens, like things are very seldom as time sensitive as they may seem.

And so for me, yeah, I'll come back to this later. Super helpful. Also, I do this with Slack, you can mark a message as unread and with email you could just mark it for un mark it as unread.

And I come back later and I'm like, ah, it's unread. Let's sort it out now. So, so again, to answer your question, my strategy for defensive work is I just mark things as unread and say, this is a problem for future me.

Yeah, that's a good strategy. It works. It works really well.

Yeah. Um, is this like canale with like virtual work, this kind of strategy? No.

Um, even in the physical. So I used to work in an office before and there's a lot of things that require defensive work. The worst is probably inter like, people will just like straight up show up and be like, Hey, can I get your opinion on this?

And without even knowing, maybe they've taken you out of your float, just like that. And then you've gotta go look at their thing. Or even if you reject them, you're still out of your flow.

So you've lost. So now you might as well, okay, come, I'll look at it with you. And then you come back to your desk and you're like, where was I?

And that takes time. Um, so no, I, I think it was maybe even worse in the office because when I'm at home, I don't have a person physically interrupting me and, and I don't see their face. And I think this is huge.

When we see a face, like in physically, we can't help but respond with compassion. It just, it's in human. There's a great book about this called Hope for Cynics by Dr.

Jamil Zaki, talking about scientifically proving that human beings are just naturally good. It's an excellent piece of literature and it's very true. Like if I, if I see a face asking me for help, the chance that I will stop what I'm doing and help them is exponentially higher than if somebody just writes me on Slack, Hey, can you help me?

I'm like, no unread, you know? Um, so the problem is better. I feel, at least for me, um, remotely.

That's interesting. Um, I, for me, like I have a dog and when he looks at me, it's like I kind of quit what I'm doing and just Because of the face. And they, they're master manipulators.

Like they know how to, like, they'll do the thing where they tilt their head to the side, they master the art of like, please, you know, they Yeah, yeah. Totally. Like there's some similarities there.

Yeah, definitely. We, because dogs are just mammals. We can look at any mammal that has a face and feel empathy for it.

For real. We can't. And if we can't, we maybe have some type of sickness.

Um, but yeah, anything with a face cats, you can see them. Some people say, I'm a dog person, not a cat person. But if they look at a cat's face, they can infer like, what does it feel like?

Yeah, That's interesting. 'cause I've never considered myself a dog or cat person. Yeah.

I actually have both. I'm a mammal person. I'm, I'm in the same camp.

I just, if it's living and it's not parasitic, let's go. I hate mosquitoes. That's the only thing I can't handle.

Yeah. I think we've had a really good chat today, so thank you Tejas. My pleasure.

My pleasure. It's a pleasure talking to you. Yeah.