AI Engineering Teams: Hiring, Skills, and Judgment

Amir: On this episode of the show, I have with me Bali Raghavan. He is co-founder and CTO at Hero Safety, and we're gonna be talking about building your team in the world and age of AI, and talk about how engineering teams are being affected and the role AI's playing in shaping those and building those teams.

Talk about real world physical problems and how AI is helping solve those, and tech for good. Bali, thanks for taking the time.

Bali: Hey, Ron. Nice to meet you. Thanks for allotting in time yourself, and it's good to be on the show.

Amir: Absolutely. All right. Hero Safety. Uh, tell us what you guys do over there.

Bali: So, we are building the end-to-end operating system for public safety. And when I say public safety it encompasses everything from what our law enforcement agencies are doing, police departments and municipalities and cities are doing to keep their communities safe.

But you can also extend this beyond law enforcement to places like how are we keeping people safe in our neighborhoods? How are we keeping our people safe in malls and retail locations? How are we keeping people safe in stadiums? We are building an end-to-end AI-enabled operating system for for this which will hopefully change the way in which people view public safety and the way in which access to public safety works today.

Amir: Absolutely. Super cool. All right, so this episode we'll be talking about building teams in the age and world of AI. I guess a good starting point would be is obviously you've been an engineering leader before, so we'll bring in some of the past of how things have changed for you.

But right now you guys are a s- a startup. W- what does engineering... H- how's the engineering team set up currently?

Bali: We we have a bunch of engineers on the team. And when we started we set it up the the traditional way, right? That we've all come to know about during the last several decades of how companies have been built.

We start out with an engineering team with a bunch of engineers. And we start out with having a couple of product managers on our team and a couple of operations folks since it's a real world problem that we are dealing with. So a very tried, tested setup is what we started out with.

But in the world of AI everything is changing. The definition of who is a software engineer and who is a programmer who is a software developer is changing on a day-by-day basis, and it has definitely changed where our designers are coding our operators are coding, our program- our product managers are coding.

So everyone is coding. There used to be a time when the engineer definition used to be can you take problems write computer software in order to translate those problems into executable code? And can that all come together to realize the product? That definition is quickly changing in the world of AI.

So if you ask me today what does our engineering team look like? I think the entire team is our engineering team because everyone is coding.

Amir: Absolutely. Do you c- I guess when you think about that and, obviously you guys are a certain size right now, a- and theoretically, To a certain size you might be able to keep things a certain way, but the notion of everyone is coding, is that something you're gonna try to keep as, as you scale, let's say, 50 people, 100 people whatever that is, i- is that something you think you're gonna try to keep as part of the fabric of the company?

Bali: I think there's no reason not to keep that, right? 'Cause one thing AI has done in my opinion is a lot of intelligence has become commodity now. What used to take a lot of specialized training and experience for you to know how to do things, it has become fairly commodity and programming has become commodity.

Writing code has become commodity. You can... you don't even need to know a programming language. There used to be a time when you used to test people when you interview engineering candidates. You'd, you used to test them on whether they're well-versed in Java, whether they're well-versed in C++ whether they're well-versed in Golang.

Nowadays you don't need that because there is an abstraction layer where your your programming skills have become fairly a commodity out there. So what has become really helpful are all the other things that you were looking for. Do you have people who can ideate?

Do you have people who can iterate quickly and try to break do- break down ideas into a way in which you can converge on solutions quickly? Do you have people who are good at execution? Do you have great problem solvers who can connect the dots to solve problems out there?

And on the engineering side obviously not all the skills that come with your engineering background have disappeared. What AI has done is it has amplified the necessity for some of the skills that were part of the portfolio of skill sets that went into calling yourself an engineer.

Certain things like systems engineering, certain things like architecture, certain things like are you able to pinpoint whether what the AI is doing is right or wrong? Those things have become way more valuable today, and that's what I'm looking for when I hire engineers nowadays.

Amir: Interesting. W- which I think, i- it's interesting when we talk about the in- the profession of software engineering, I think the software development piece I think sometimes gets overstressed as to, it is being automated. It simply might not be the most valuable place as you've described.

I guess if you were to go one layer deeper and talk about there are other people outside of engineering writing some kind of code. Obviously, and you guys are building pretty mission-critical. It has to be very fault-tolerant. W- when you look at the shipped product and you look at the role everyone that plays outside of engineering How much of that code how much of that code is the basis, meaning it's actually moving into, the workflow of the engineering team?

Or how much of that is used to help bridge requirements, iterations, et cetera? 'Cause I guess for me, when I look at it, is that I think there's a... doesn't always have to be code that's shipped coming out of, Agentic. It could be facilitating the code that's being shipped.

Bali: Yeah. Yeah I think you you will see a wide variety of things that are happening.

Even if you look at my own team you'll see a wide variety of things that are happening. There are things around oh, are we are we generating the right designs? There is like a lot of AI-based design development that is happening out there. Are we are we understanding the requirements and are we framing the framing the problems correctly?

There is a lot of things that are happening out there. Are we if you look at our go-to-market side are we using AI to figure out the go-to-market strategy and the lead gen piece of it? There is AI kicking in there. Are we creating support bots that can essentially offload a lot of support requests that are coming coming in related to our product from our users?

There is a good bunch of AI that is being used there. Are we doing simple things like you need to create a fairly straightforward feature page? Are are there things like, hey, you want to create a profile image on a user page? There are, there is code that is going into production there.

But then there is like mission critical pieces where there is a good amount of fairly deep software development that happens. So you're going to see a wide spectrum of tasks and activities where AI is kicking in. And what I would say is there used to be a time not too long back even 12 months back, 24 months back, where for some of the tasks that I mentioned, you needed engineering help.

You needed engineering bandwidth to make it happen. And the amount of bandwidth that you need from your core engineering team for making some of these things happen has drastically gone down because people are able to take things to the finish line without engineering bandwidth being required.

Amir: Absolutely. I love that. I guess when you look at the role of the engineer, as you've described it, it is evolving. I spoke to another CTO, and they mentioned he's had to help his team get better at asking questions because that's becoming more of the crucial skill set versus obviously writing code, because you need to ask the right questions to g- generate the right requirements and make sure that you understand what's being developed is on track to meeting those requirements.

When it comes to the skill sets that you're seeing, you mentioned, some of the non-coding ones. What are some of the skill sets that you're seeing that are becoming more valuable in this age of using agentic?

Bali: Yeah asking questions, asking the right questions that's always existed out there, but it's way more amplified in the agentic world that we live in because it's at the end of the day it's an inference engine.

It's going to it's going to do what you ask it to do. But are you asking it to do the right thing? And are you asking it and prompting it in a way where it can produce the right outputs is obviously like super important out there. The other thing that I don't see AI replacing for a very long time or maybe forever is judgment, right?

Or are there can you make a judgment on how to do things and whether you're doing the right thing, whether this is the right solution to the problem, even before you prompt prompt the AI engine out there. That is becoming very important. Can you make-- can you use judgment?

Can you make the right judgment calls? Thinking from first principles, and this is where problem-solving and ideation start to come into the picture. Are you thinking from first principles or are you just trying to get it done quickly is another skill that is becoming very important.

And then the other side of the spectrum is now that you have figured out what is the most efficient way to prompt the system and generate outputs, how do you know whether what it is doing is right or wrong? How do you know what the output is going to do to the rest of the system?

How do you know whether it is not doing something that it should not be doing? So those kind of skills where you can almost look for a needle in a haystack, right? Because code is being spit out at rates that have never been seen before. There's like lots of lines of code that are being spit out on a minute-by-minute basis.

So understanding what this code is doing to your systems is where a lot of software engineering and computer science background is becoming very helpful, and it's like a good skill set to have. And how do you how do you make people efficient out there?

And efficiency comes in various forms, right? For some of the, Some of the stuff that you're doing. Are you being efficient in terms of how the entire team works together in an AI world? And are you creating the harnesses? Are you creating the products? Are you creating the processes so that people can remain efficient?

Cost efficiency is another thing. AI is like now becoming like a fairly significant piece of the spend for every company, big or small. So are you being efficient in terms of how AI resources are being spent and is the return on investment still worth it? Are you thinking about those things?

Or is this like blindly going and using the latest and greatest models available and doing what you want to do, where to change like a single line of code, you're spending like hundreds of dollars? Those are the kind of skills that are becoming really important in the new world.

Amir: Absolutely. Wh-when it comes to your team and hiring, obviously in this case could be engineers, not engineers, 'cause everyone in the organization is using AI. But specifically when it comes to engineering have you seen any shifts in how you are assessing and what you're looking for in that interview?

Obviously... and also, are you still using, coding as a assessment form?

Bali: Yes. It's it's funny, right? Even in even in my own engineering process there's been like a radical shift over the last, I would say, 12 months, not too long back where there used to be a time when when you interview engineering candidates there were technologies developed there was proctoring to make sure that people are not gaming the interview process or they're not using AI tools to game the interview process.

But if you look at how engineering interviews have changed now, and this is so true even when I interview our candidates now though the bug has become a feature. If you are not using AI then it's a problem, right? Because that's what you're going to be doing on a day-to-day basis.

So this I would say there's been a radical shift in how the engineering interview process works where you want people to use AI. You want to extract those signals which are different than what you were trying to extract previously. Previously, you were obviously evaluating candidates for problem-solving skills and soft skills and other things.

Those things still continue, but you are also evaluating them for their ability to program and for their ability to code. And yes you still want to do that, but you view it with a different lens. Are you using the right AI tools and the AI systems in order to code? And it has amplified the other things around, like problem-solving, around system design around architectural skills, around first principles thinking.

Those things have become amplified because coding and code generation has become fairly boilerplate.

Amir: Absolutely. It's interesting. In your company, obviously, you guys are dealing with some mission-critical applications. It has a component of the physical world.

You guys are tackling and solving problems that, have a different level of complexity because of that physical component. When it does come to hiring you'd love to have somebody come in with that experience. It's not always gonna happen. When it does come to hiring for the type of people you do need to join the company, does anything change or, is...

Are you still looking for the same things, or does the type of, you know, product you're building require a slightly different lens?

Bali: You-you're going to need both generalists and specialists, right? There are, We work on a wide variety of problems out there. Some of the some of the frameworks around hiring generalists and specialists hasn't actually changed if you look at it.

It's just that what what skill sets that, you-you're expecting out of the generalist bucket and what skill sets are you are expecting out of the specialist bucket is what has what has changed, I would say. From a generalist perspective full stack engineers can they can they work across the stack?

But now in the AI world, can they use the right AI tools and can they solve the right problems is definitely what you are looking for. In the specialist bucket you're still looking for specialists, right? For example, we we work a lot on computer vision and AI systems and we work a lot on not just cloud AI, but we also do quite a bit of AI on the edge.

People who have who have dealt with problems like that, can they tune GPUs? Can they extract the mo- extract the maximum out of GPU pipelines on the edge? Can they process massive amounts of data? And ha- do they have the knowledge the nitty-gritty, as well as the knacks of extracting the best out of the AI tools that are available out there?

How do they fine-tune the system? How do they train the system? If we have to develop custom models do they have do they have experience with that? Do they deeply understand how AI systems work, not just at the superficial level, but under the hood how do these things work and where the- where are the bottlenecks?

Those things are definitely very valuable nowadays. The need for generalists and specialists hasn't gone away. But the lens with which you're viewing the skill sets that they bring to the table has definitely changed.

Amir: Absolutely. Yeah and I guess, Hero Safety itself we're talking about a public safety platform.

It's AI native. When you're looking at the role of AI, I think it was maybe a year or two ago people I think there was a... We'll leave names off, but there was a company that makes doorbells and has cameras and they thought that it could potentially be a solution to helping facilitate public safety as a part of the overall.

There was a backlash for that. ... AI itself has a little bit of a positive and a negative connotation. When it comes to public safety and AI, talk to us a little bit about the role AI plays i- in terms of, what its role is and are you seeing... What's the reception you're seeing of AI and its role within that public safety?

Bali: Yeah, the interesting piece is a lot of folks in public safety they're aware of what is happening around them in the world of AI and how AI is transforming other industries. So you go talk to folks in public safety, you'll be surprised that they definitely are well-versed.

They some of them can talk at the level of an Opus model versus a Fable model. So they're pretty well-versed out there. But technology hasn't penetrated or looked at the public safety space at a level it could have, right? Compared to some of the other consumer-facing areas where technology has penetrated a lot.

There's not been a very high level of technology penetration. Not to say that it hasn't happened. You have the Axons of the world you have some of the other companies of the world where there are technology systems that exist out there, but it's extremely fragmented and the penetration is not at the same level that in other other verticals out there.

There is definitely curiosity and there is definitely intent to adopt AI systems because most of the departments or most of the most of the people who deal with public safety are the-- it is a highly reactive labor-driven system. They can't keep up with the load.

They can't keep up with the sophistication. They don't have enough budget to essentially do all the things that they should be doing. But they're on the hook for keeping their community safe. Which is which is a fantastic noble thing that they're chasing.

But things are like way more manual than what in some of the other real world spaces out there. And that's where there is curiosity and there is intent to adopt. But at the same time, and rightly there is also skepticism, right? Is AI going to result in the wrong decisions?

Is it going to is it going to invade privacy? There are a bunch of things that fall into the right bucket of healthy skepticism, and those are the constraints with which we want to operate.

Amir: Absolutely. It's interesting 'cause when we start talking about, we're talking about public safety and we're talking about AI, there's also this, the flip opposite story that people will call 911 for help and they're gonna be on hold because there's just not enough people.

And that's a discussion point, and I feel like if, as you said, there's a performance booster, if we can actually get to those calls and help triage, eval- whatever that is, that should move the needle. Which I think sometimes when we talk about AI, I think it's we wanna hold the computer to a higher degree of standard at times, but also then the productivity gain might come with some of its own error.

But there's also still error that's happening. It's not like humans don't make errors they work- there's, However good you are at your job, you're s- still susceptible. I guess when you look at it and you look at this gap of there's a constraint, people in the medical field are facing the same thing because people feel comfortable when they're directly dealing with a human.

There's still a, a phase of comfort here. I guess a question for you is it comfort? Is it just early and we haven't, as humans, we haven't had a chance to see enough results to feel confident? Is that what we're seeing or is it something else?

Bali: I think it's early and it'll take time to to get adopted and for people to feel comfortable with with the use cases that are AI-enabled, right?

And we have seen this over and over again. When before the telephone came in your option was to go walk to the person that you want to talk to. And then the telephone came in. There was like a level of skepticism around "Hey this is now remote communication. I can't even see the person.

Is this going to be good or bad?" But over time it it did get adopted. Same with like internet, same with smartphones same with like any of the technologies that we have seen out there. And I think AI will go through the same cycle. What will be interesting to see are the use cases where people get comfortable and realize that hey, this is part and parcel of how things work.

A classic a classic example is Amazon, right? When I was I was there and we were launching the jewelry store or even e-even the TV store as an example. Those were the days where you had very established jewelers who were selling gold and diamond rings, and you had those Circuit Cities of the world that were selling these large screen TVs.

And there was this question of "Hey can you really spend tens of thousands of dollars on on, on a diamond ring without without seeing it and order it online and expect that it's going to show up at your doorsteps? Can you l- really order a, a fifty-inch television set and you're going to pour thousands of dollars into it without seeing it in person?"

And time only proved that oh people got more and more comfortable with doing that with some of the online stores. And I'm sure that AI will see a similar cycle.

Amir: Absolutely.

Bali, I appreciate your time. Thanks for coming on and sharing with us.

If somebody does have a follow-up question, what's a good way of connecting?

Bali: LinkedIn is the best way to connect. I I made a call a long time back, and I'm happy about it that I'm not on Facebook, I'm not on Instagram. I'm a LinkedIn person do connect with me on LinkedIn.

Bali Raghavan if you if you search me up on LinkedIn you should be able to find find my profile and send me send me a connect request.

Amir: Awesome. Bali, appreciate your time. Thank you.

Bali: Thank you. Thanks, Amir. Have a great

Amir: day. All right. Absolutely.

Bali: See you then.

Amir: All right.

That's it for the episode. Be back again, different guest, different topic. Until then, two things. One, Bali talked to us about how AI has affected him and his team, and how they go about building, and they apparently have AI touching nearly every aspect of the enterprise, not just engineering teams, but other disciplines as well.

And also the fact that they are working on a physical problem and how AI is helping within that, specifically public safety. And obviously we hear a lot about this, and now with the introduction to AI, we might see some unique and interesting solutions Bali mentioned. So exciting to keep an eye out.

And also share this episode with anyone else that you think might be interested in it. I'd appreciate that. Also subscribe, comment. Leave me a review on Apple Podcasts if you can. Until next time, thank you and goodbye.