Beyond The Code

Maulik Sailor (00:10)
Hi, I'm here Malik Sela, founder and CEO of Notchup.com. We are building an AI co-pilot for engineering managers to automate their people ⁓ ops. And today I've got two very interesting guests for us to talk about engineering management in general, from two very different perspectives. Both of these guests brings wealth of...

⁓ experience ⁓ from their individual background as well as for the role that they had up. And today we're going to mainly talk about how as an engineering leader, what are the challenges that they face ⁓ in their day-to-day activities, know, short term, mid term and long term. How do they really go about ⁓ managing the, what does it mean by engineering management? You know, what elements are there? You know, what are the challenges do you see?

what are the tools out there in the market currently at the moment, which makes their job easy. ⁓ So without further delays, let me first introduce Robert. Robert, I happened to meet him very briefly actually in SF. I'm not sure if you remember me or not, but we spoke for a brief moment as one of the event. I like what you spoke back then and I was like, okay, I want to get you somehow in one of these podcasts that I do. And I'm really glad that you could make it today.

So to our audience, Robert is currently VP of Engineering at Sugars CRM. ⁓ I'll let Robert introduce himself in a minute. And then we have a second guest, ⁓ Ajit Beharra. He's also an engineering manager. He's currently heading up a larger team at Meta, but his experience mainly is on enterprise side. But today, both of them ⁓

almost in a similar role, but from a very different perspective, is going to talk about the challenges that they face. So why don't we start with you, Robert? Robert, it will be great if you can ⁓ introduce yourself and talk about your day-to-day activities as an

manager. Sorry, as a VP of engineering. Actually, you know what? That will be great, because I...

Robert G (02:18)
Sure. Sure.

Maulik Sailor (02:23)
I understand there are different hierarchies between engineering managers and VP of engineering, but depending on the company side, that definition can blur. So would be great to also understand the differences in the two roles. Yeah.

Robert G (02:36)
Okay, so

I'll start from the VP of engineering role. I'm Robert Gonzalez. I've been with SugarCRM for a little over, actually about 13 and a half years now. Started off as an engineer, as a senior platform engineer within SugarCRM and worked up through leadership into management, now into the executive ranks. I've been in tech for a little over 20 years.

I've enjoyed my entire journey to where I am today. And one thing I will say right now is that engineering leadership today is at an inflection point. We're in one of those periods of time where ⁓ we're experiencing change like we haven't experienced in a very long time. And it's creating a lot of excitement and a lot of energy. For some people, some anxiety, and that's a... ⁓

It's an exciting time to be in, for some it's a scary time to be in, and I'm having a really good time in the midst of both of those spaces.

Maulik Sailor (03:37)
That's great. That's great, Robert. ⁓ Ajit, ⁓ would be great if you can introduce yourself and your background. And again, it be great for us to understand your day-to-day activities, you as an engineering manager. What does it entail? And to be honest, you work for a large ⁓ enterprise, right? ⁓ So from that perspective, there might be some blending of... ⁓

Ajit (03:56)
Bye.

Maulik Sailor (04:02)
⁓ the definition and the roles of responsibility from one level to another. So would be great to hear from you ⁓ about your role, your background and at different companies, do they call it same or do they call it different?

Ajit (04:19)
Yeah, maybe I can start with my background. So I have a very engineering centric background. I have been ⁓ working in this lives tech companies for about 15 years and I moved to people management. I've been leading engineering teams for quite a few years now. ⁓ My experience spans from learning the like

say in this case because today we are discussing about AI, ⁓ specifically JNAI, how it can improve engineering team operations, how can it enable them further. So I have a good understanding of language models and then as part of my, I'm also actually doing my MBA at Berkeley Haas at the moment, just wrapping up. And as part of that,

I have been doing some independent studies on how AI can be leveraged for decision making especially in the people side. So all the perspective I bring to this discussion is from those. I'm not defending my current company meta in this discussion. But as the discussion goes I can share more of my experience.

Maulik Sailor (05:39)
Yeah, that will be great. Great. So one thing you touch upon is about people management. But generally, ⁓ across the board as an engineering leadership, when we talk about, and again, Robert, feel free to correct me if you think that I'm not heading in the right direction. But generally, in engineering leadership, you have four aspects to look after. One is your technology.

like the piece of technology that you are responsible to deliver ⁓ as per your business priorities and so on. Second is your people side. You need a bunch of people to deliver that piece of technology or to work and then create that technology and make it available for the business. So that's second to people side. Then you have the process that, okay, you have the people, but you need to take them through a particular process

to create that output that you need to create. Generally, when you talk about process, we talk about dev-op practices and agile delivery and all of that, but is there something else that we should be talking about? And then the fourth element is financial management. It's going to cost you ⁓ a few things, like whatever the budgets are, to run the team that you are responsible for.

Plus, you know, any process that you implement, there might be some tooling involved in that process, right? So there are costs involved in that, right? And with this AI coming in the mix now, then okay, where do you deploy the AI tools? Where do you deploy the human, you know, in that? So I think in my opinion, these are like the remit of engineering leadership or management. Would you agree? Am I missing anything here?

Robert G (07:35)
I would agree with that. mean, when you look at generally when you look at the areas that we impact the business, it's going to be in those four general areas, right? In the world of like lean manufacturing or practical problem solving, you have those broken down into the four Ms, man, method, material. You just kind of covered all of that. And as engineering leaders, we have to be cognizant of all of that, right? Our technology needs to be

It needs to be the right solution for the customers that are leaning on us. It needs to be built right so that way it maintains proper levels of security and quality. It needs to be built in a way that is priced right for our customers to consume it. And it needs to be built in an efficient manner by the people that are building it in order for it to be saleable. Like it needs to be an efficiently developed ⁓ tool that the company can back with its own investment. So I think those are four key areas of the role of any leader to be aware of what's going on there.

Pajit?

Maulik Sailor (08:32)
would you want to add something?

Ajit (08:34)
Yeah

Yeah, think so that's why I feel like what you guys said in those four areas, I think in all of those areas, Gen.i can be leveraged to make it more, for example, say if you want to reduce cost for consumers. Now you are spending for people also on tool tooling on many, this is a cost kind of part of and then when you leverage Gen.i the right way, then the talent you retain are

the productivity gain you get that will affect the cost as well. So, you can discuss more along those lines.

Maulik Sailor (09:13)
Okay, great. Wonderful. So, you know, so my background has been kind of a product. So I started into engineering, ended up coming in becoming a product manager and I've been in a product for a long time. And then I also perform like engineering leader role, ⁓ kind of a CP or CTO kind of hybrid role for a startup. So I was responsible for product as well as for technology and engineering practices. ⁓

But generally when you're talking to product, in the product you talk like three flavors of a product manager. A product manager who is more user focused, a product manager who is more technology focused, you know, technical product manager, or a product manager who's more market or commercial product. And you call it like commercial product manager or product marketing manager, right? So kind of that's where you have like three different flavors within that. Do you think in the engineering management,

you also have these kind of flavors ⁓ of an engineering leader. Somebody who's focusing more on people side, somebody who's focusing more on technology side, somebody who's focusing more on the financial and strategic side. Have you seen something like that in the market?

Robert G (10:31)
So I haven't seen that the way that I've seen it in PM. ⁓ I know exactly what you're talking about in the world of product management. You have product managers who are ⁓ heavily tech slanted. have product managers that are very ⁓ product or user centric. And then you have those that are geared towards sort of the financial side of things. On the engineering side, I've actually seen that, but I've seen it in a different way. So I've seen tech forward engineering leaders.

and user or product forward engineering leaders that tend to sit at the same level, right? As you start to get up into like director level and VP level, you start to have people that lay in place in those particular areas. One easy way to look at that is, you know, platform or backend versus front end where your backend is generally not as concerned with the users and what the users have in their hands as much as what happens for the users with the products that they're using. And the front end is much more user centric.

The higher up in the organization you go, within technology anyways, can take at the CTO level. They're interested in technology, they're interested in users, and they're very interested in the financial implications of those choices to the business. And the financial implications cover the breadth of the business. It's the cost of people, the cost of tools, it's the cost of getting your product to market, it's the cost of transition and digital transformation. There's a lot of costs that you need to take into account. But generally, I believe that sits

the cost part of it sits at a higher level in engineering, at least in my experience, whereas the focus on product and the focus, or the focus on users and the focus on technology has a tendency to be richer at a little bit of a lower level.

Maulik Sailor (12:12)
⁓ I've got a few bunch of questions actually over there. I'm going to do a little bit side talk from our core topic. Oftentimes in many companies, you have product reporting under CTO or VP of engineering. In some organizations, you could have a CPO and a head of engineering reporting under the CPO.

where some organization will keep it completely separate. You will have a CPU running the whole product management and then you will have a CPU running the whole technology management. Have you seen this? Do you think there's a benefit of doing one way or the another way? ⁓ Or that could be really a problem in doing one particular setup. What does your experience say?

Robert G (13:03)
Ajit, do you want that one first? I have an answer for this, but I want to give it to Ajit first.

Ajit (13:04)
Now, you got

it. think I can maybe cover for the first question based on my experience. But I think this is something maybe Robert you can check.

Robert G (13:15)
Okay, I'll step into it. So I've actually worked in organizations where engineering and PM were under the same umbrella. And I've worked in organizations where PM and engineering were separate. And I think there's positives and areas for improvement in both of those structures. One area that I think is a strong positive when you have PM and engineering under the same umbrella, in our case, it was all under the CTO ⁓ umbrella. ⁓

One of the areas I think that showed the most benefit was there was a lot of synergy in the teams because they were basically embedded into the same organization. We had the same head who was ⁓ sharing with us the same strategy. We were all aligned on that strategy. We were all working together and we were all rowing the boat in the same direction without any confusion and any lapse in that conversation. ⁓ However, when you have strong leaders like a CTO and a CPO that have their own organizations and those leaders

are pushing and encouraging collaboration, you get the same effect. It's not as rapid because you're not in everything together, but you do get the same effect. the organizational structure that I'm in right now, we have a CTO, we have a CPO, and both of those organizations are ⁓ very well aligned and working towards the same initiatives in really strong collaboration in our own separate verticals. Ajit, go ahead if you have something else to add to that.

Ajit (14:44)
The first question that Malik asked about in the engineering management is there leanings of engineering managers? Somebody is more technology inclined, somebody is more people inclined. In the tech companies I see that there is ⁓ like depending on people's inclination, we'll find engineering managers who kind of have more hands on approach.

to learn what the team is exactly working on and also how they can enable the team even on the technology standpoint. But there are other teams where the engineering managers focus more on the people part, leaving the technical work to tech leads or the engineers in the team. But as Robert mentioned, as you go higher up in the, like go to directors and VPs,

⁓ this kind of split is not there much even if it's a very highly technical area so they might have strong engineering or technical background but the hands on work definitely gets lot reduced.

Maulik Sailor (15:50)
Wonderful. ⁓ I want to drag the conversation back to our core topic, Apologies for taking this ⁓ sidestep. But generally, sometimes on our audience, we have people who are growing up their career and they may want this clarity that, okay, how can they map that background into their future goals? Hence, thought this sidestep will be interesting. ⁓ Coming back to the core of our topic, right? About engineering leadership.

There's a lot of AI buzz happening across the board. Any part of the business, you can imagine any workflow or any function you have, any task that you need to be doing within the organization. There is an AI tool for that or there is an AI agent for that. There is an AI workflow for that. ⁓ Or if there's nothing, then you still have AI automation platform like NITN, ⁓

which you can basically execute a workflow, right? ⁓ Now, going back to our original four segment, technology, people, process, and strategy and finance, where do you see the most impact of all this AI revolution that we are currently living into? ⁓ Why don't we start with you, and Jit?

Ajit (17:18)
think starting with the technology. if you think about, especially in the ⁓ software development or software infrastructure. So previously the way people used to write software is like encoding some rules, routines and concrete instructions and then storing that as like software. But with AI that

paradigm is totally changing. like some people even dub it as like software 2.0 where you are not the software itself the network or the language model itself is the software and the the what we are storing is the weights.

So this is no longer like you don't have to explicitly say this unit to do this because now we can use natural languages to do it or you can encode the patterns into these models to do this. I think that's where the biggest change is gonna happen first will happen first because a lot of this software 1.0 is gonna move to software 2.0, but I think that will change how engineers work in the team but on the team scale like engineering team scale

how we can like adopt the narrative AI into engineering processes, ⁓ hiring, performance management, those areas are also going to change significantly in coming years.

Maulik Sailor (18:53)
I'm glad you said that because that's what we are hoping to do ⁓ with Notchup. But anyway, Robert, where do see the most impact between these four segments that we added?

Robert G (19:08)
So I tend to look at this at a little bit of a higher level. ⁓ I agree with Ajit that you're going to see tremendous impact in the software engineering space in particular. You're going to see tremendous impact in the core of the work that we do, which is producing code, producing product. But at a higher level, I tend to think that the biggest impact is going to be on process. And it's going to be all the things that Ajit just said, hiring, performance management.

strategy, ⁓ even to the lower levels like sprint management and code review and architecture and meetings. You're seeing this in meetings a lot, Where you can hop into a meeting, not have any additional software anywhere on your computer. In 30 minutes, your meeting ends. In 31 minutes after you started that meeting, you have a full recap with summarization and transcript of your meeting that can then feed into your other AI systems that build your context, that feed the model, that tell your systems, this is what you needed to know.

from that, right? If you look at the way that we've done business up to this point, and you look at the way we're doing business now, AI is having, in my opinion, the biggest impact in all of the processes that we follow, whether it's your financial planning or your budgeting or your team planning or, as Ajit said, your hiring or your performance management. All of those areas of the business are being positively impacted.

by AI. It's making things faster. It's making things clearer. It's surfacing the right data at the right time when you need it. And it's putting you in a position to have insights where before you needed to go find those insights out of a pool of data. The AI tooling that we're using now is doing that part for you. So your job becomes a lot easier and a lot faster.

Maulik Sailor (20:54)
Okay, now I'm quite familiar with the engineering side, right? You you have your GitHub co-pilot, you know, you have your cursor, you have your, you know, the wipe coding, you know, the lovable and the template and all those stuff, right? ⁓ I heard a saying that, you know, there's no feature for a front-end developer, you know, would you agree?

Robert G (21:17)
I would not know. So I'm a huge proponent of keeping the human in the human resource. And the reason for that is there's one element I believe that humans have that computers haven't mastered yet. And I don't know that they're going to. And that's the element of emotion. The element that you have to be able to connect with your customers, connect with your market, connect with your coworkers. Like you can put 50 AI agents together and tell them to go build a product. They're absolutely going go knock it out of the park.

If you can go tell them to build a product that's going to make their users feel good about themselves, it's very hard for them to do that because they don't understand the nature of feeling, right? There's pieces of the process that people still need to be a part of. I believe that AI is very good at doing a lot of the heavy lifting that we've been doing over years as engineers. But at the end of the day, I think we are still the gatekeepers of what makes its way into the hands of the customer. Whether you're looking at a coding agent,

or a code review tool or a diagramming agent or whatever AI tool you're looking at, if you don't put human eyes on the output of it, you're taking a big risk in my opinion. I think that the jobs we do as engineers are gonna change. Like Ajit was saying, I don't need people to go and write code anymore as much, but I still need people to make sure that the code that's being written wherever it's being written from passes muster.

It needs to be secure. needs to be high quality. It needs to be scalable. It needs to be the right application of logic. And while AI can do a lot of that work, the gatekeeper there is the person, in my opinion.

Maulik Sailor (22:51)
That's it.

Ajit (22:52)
I think I agree with Robert on this. I think AI is not going to fully replace engineers, but it's definitely changed the way they work significantly. For example, like if you go back like 30 to 40 years or even further back, we used to write like very low level languages. Slowly that has moved up. are writing. So today the most used language for AI is like Python.

It's a long journey from the low level language to here today. But with AI, I think that will shift towards the natural languages. So, what that means is the engineers, so somebody with domain knowledge, they would still be in charge to make sure that.

the intent is right like to build a product what we are building is it connecting to the customers they need to make sure so that also means that they are no longer the traditional engineers anymore so every engineering work will have little bit of product management thinking into it

So I think it's kind of the abstraction level is slowly going up. So today maybe still they're going to do a lot of validation and testing after the software is integrated, being generated from AI. But even that will slowly move from manual work to AI. So a lot of validation testing will be done by AI. But the whole system will still be set up by engineers. And the whole intent and the

goal, what we are building that still needs to be set up by humans. So, I think that is the direction I think it is going toward.

Maulik Sailor (24:38)
I want ⁓ to touch on the other side, apart from technical. So strategic and financial side. Now strategy and finance, it's not really a set of rules. It's a lot of guesswork. It's a lot of experience, gut feeling, data that you are analyzing, but you're analyzing ⁓ not in a way that everybody else will be analyzing.

companies with the same sets of data with competing product will still have a very different strategic direction that they may pursue, right? Do you think there's a role for AI to play there?

Robert G (25:21)
Yes, I think there's a role for AI to play in everything we do. mean, for any of us that have used AI in any capacity, we can see the power that it brings to the table and the things that we do. And something you just said right now is actually critical to this conversation is if I have a data set and you have a data set and it's the same set and you and I are even in the same product space, you and I are likely going to grab different insights from that data the same way that AI works. You have a model, you have another model.

you send the same data to it, you're gonna get different results. Hell, you might send the same set of data to the same model twice and get different results from it. It is our job to use all the tools at our disposal to be able to do the jobs we need to do effectively and efficiently. And I think AI plays very well in that space of strategic initiatives, building out your forward strategy, your product strategy, your tech strategy, and every other strategy that you might have. I think AI has a home there.

Maulik Sailor (26:20)
Okay. Have you seen the tool currently in the market which is working at that level?

Robert G (26:27)
So I haven't dug deep into like strategy specific tools, but I've played around with what the various LLMs could do with things that I want to do. Like you can feed a lot of data into these things and ask for a lot of different outcomes. I know like the Anthropic series of LLMs are very good. Open AIs is very good. ⁓ I've played around with Gemini and what Google has brought forward with their models in the past. I have my favorites, but I haven't looked at a specific tool for this yet.

Maulik Sailor (26:58)
I did.

Ajit (27:00)
Yeah, think so now these language models, they come with that deep thinking mode, which can be really used for coming up with strategy and if you're incorporating a lot of data to make decisions. just going taking a step back like to your original question, like what kind of role AI is going to play in strategy. So I think humans have

a drawback in our thinking because if you ⁓ think about like all the noise there in the data like when the volume of data is really high when the data is conflicting humans are not really good at decision making ⁓ similarly there is a lot of also bias

not that AI doesn't have bias but AI can be easily rectified but humans the system the way it is built is gonna like accumulate bias and noise over time. So it's not easy to ⁓ remove like in AI systems you can do if you want to if you intend to take action. So when you have a lot of data and

when the data is conflicting or it can actually make overwhelming for humans to make decisions, I AI ⁓ plays a role. So when you're going back and making strategy decisions and you start with data, so the fast scan that needs to happen can happen through AI, but later on humans should be at the level when they take that deliberate decision making based on that extracted data.

Maulik Sailor (28:35)
Okay.

So overall, what we are seeing here is that current state of AI is, I think we all agree that it's super powerful, it's super useful. And there are a lot of, the LLMs in particular, I have got a lot of advanced capabilities right now. But the problem is a lot of, I mean, they may come back with different flavors of output.

⁓ based on what you feed to them. And so there needs to be that human judgment and the human control required both on the technology side as well as on strategic side and so on to make that decision. So eventually the decision making is still with the humans, but then the actual processing of the underlying data or background information, you can basically do it through LLMs or any other AI model.

Ajit (29:29)
So in the first phase like we can reduce that variability in answers of LLMs by setting guardrails or setting rubrics. So that would give us more filtered or more refined data and it won't actually overwhelm with all the raw data you have. And looking at that data I think it's easier to take decisions for humans than starting with the initial set. ⁓

Maulik Sailor (29:57)
Cool. ⁓ Coming back to the other pillar, the people and process. Now, people is a difficult one. There's a process side. I have an interesting thought on the people side. I'll reserve that towards the end, but let's just talk of the process. Now, process, you can again split in two sides. Your technical process where you running, let's say, your CI, CD process to produce the output.

But then you also have other admin and compliance and necessary evils that you have to do. For example, reviews. I wouldn't say necessary evils, but it's something that not everybody enjoys. ⁓ Your hiding and fighting, upscaling, all those things.

Do ⁓ you enjoy this, first of all? If not, are you deploying anything AI to take care of all this? I don't know to use the phrase necessary evidence, but the things that you don't enjoy, particularly on the process side, people in process side, are you deploying anything at the moment or are you aware of anything in the market?

Robert G (31:16)
So I can speak to that a little bit. ⁓ There are elements of management that are people oriented that I think are very enjoyable for people who are for managers that are people first. Let let me say that I personally am a people first manager. So things like coaching, mentoring, career pathing, ⁓ guiding, correcting those things. I actually take a great deal of enjoyment in that. So I have not looked to offload any of that to

AI tools. What I have looked to AI tools for though are patterns in staff that might need to raise certain things to me so I can see them. Right? So there's a lot of software on the market today. ⁓ Right? You particularly if you look at like productivity tools, that's a great one to start with. You look at a thing like ⁓ DX or linear, I think it's linear B or jellyfish in those spaces that look at data for your teams, that for your people, and they surface patterns that you can look at.

Those patterns can range from anything like excessive throughput to diminish throughput to higher defect counts to slower turnaround times on PRs, right? Those all tell a story. That data tells a story. What I like AI to do for me is not just give me the data, but to give me the story along with the insight so I have actions that I can take. From there though, I like being able to take those actions myself. I like having the conversation. I like hearing the other side of the story because a lot of times data can give you

a portion of the story, but in order to get the full story, you got to actually talk to a person. You got to ask them, hey, why is your output going down over the last two weeks? I have people that in my organization, they've had family trauma. They've had political turmoil in the part of the world that they're in. And they're either without power for parts of the day, or they have to pack up their family and leave on a moment's notice. Those kind of things impact a person's ability to do the job that you're asking them to do.

So you can't just look at a number and say, you know what, that's a red flag. That's an indicator. And we need to be able to use those indicators in the work that we do. For the people side of that, for me, I don't want to use tools for the people parts. I do want to use tools to get me closer to that. But when it comes to the actual handling of people, whether it's career pathing, correction, promotion, rewarding, I actually like doing that part myself.

Maulik Sailor (33:36)
Yeah, no, it's a point. know, one thing I remembered when you mentioned that, know, sometimes, you know, I'm familiar with all these tools like jelly beans and all stacks and the others, which gives you all this data that, okay, you know, raises red flag, okay, there's something not, or a particular team member is not productive enough and all. You know, oftentimes, some teams, are underperforming, you know, and...

you will just see an issue that, okay, they are a little bit slower than the rest of the team. Some of their velocity doesn't make sense. There's not really any obvious problem. Probably most people are fine. But when you dig deeper, and the limitations of those tools is that they only look into your engineering stack. But when you connect the people stack, then you can see, okay, you know what? A particular team, they're underperforming.

purely because they just simply have a conflicting behavioral style or a working style, right? Which is not reflected in the ⁓ engineering stuff that they do, right? Sometimes, and you can rectify that by literally just like moving people from one project to another, you know? But generally people are not self-aware that, this is the way I'm working.

and this is the way the other person is working. You you try to talk, you talk to each other, maybe, ⁓ you know, maybe you're personally in conflict with each other. You understand, sometimes you put on with it and so on. But you know, once you connect the people side, that's when you start seeing those ⁓ hidden, ⁓ you know, data or hidden insights that are generally not available in a lot of engineering tooling, in my opinion. So that's something that we discovered in

you know, some of this stuff that we have been doing, that oftentimes is not like, okay, people are underperforming. It's not because they are underperformer. It's just they are put into a wrong, ⁓ in a wrong team, actually. ⁓ Sometimes, you know, when you talk about coaching and mentoring and upskilling them, sometimes you give them the upskilling to your team. ⁓ They will do it. You you ask them, they'll do it, okay, fine, part of my PDP, I'll get it done.

you know, blah, blah. But they're not really motivated to do that because their long term career goal may not be aligned with that. Right. But you may have somebody else in a different team whose long term goals may actually align with your long, you know, your strategic priorities and so on. So it might be better to move that person into your team and assign that learning to that person. Right. So these are some of the insights that ⁓ we discovered, you know, in some of the pilots that we did.

that is beyond the obvious, things that you cannot really see. And for us, why this is exciting is because we were trying to build all of these things, the data points and everything internally and trying to create models for us to figure everything out. And it was like usual, taking us to do forever. And then last year, said, okay, instead of us trying to build everything, why don't we connect LLMs to this one, generative AI, right?

So we connected all the data sources into LLM and suddenly like, know, the model works amazingly well, right? So, I mean, and it can do a lot more correlations than even what we were thinking about, right? So for me, just that shift from our internal models to, you know, some of like, you know, OpenAI in particular and Llama ⁓ literally changed the way our platform works. So I'm like really, really positive about that. ⁓

Now, when dealing with people, there's something we touched upon earlier. AI is generally probabilistic system. It's not a deterministic system. And when you're dealing with people, there are a lot of regulations and complaints. You can't just hire and fire. There might be you're making your decision based on ⁓ information, which seems true, but may not be true. ⁓

Or maybe ⁓ the AI is not really able to handle, I think Robert, you touched upon earlier, AI is not able to handle ⁓ the human side of it as well ⁓ as a team leader should be able to do. What's your general advice? So if you were using some tools right now, I'm not sure what tools you are currently using, but if you are, what are your advice or

a few tips or guidance you can give to other similar leaders like yourself.

Robert G (38:32)
So the bits of advice I would give are maintain the level of curiosity you had as you rose through the ranks to leadership, right? Engineers are curious people. Engineers are driven people. Maintain that because as technology evolves, there's gonna be a lot of opportunity for us to do some really cool experimentation and trials with some things that are coming out that haven't even been invented yet, right? Every day, some new form of technology is being invented that we get to try. Stay curious, keep trying.

At the same time, keep people at the center of what it is you're doing, because I still believe, and I will hold onto this, that people are a very important part of the process and the businesses that we're building. While I do believe that AI can make great strides in doing a lot of work that has historically been done by people through just general labor, there are still elements of the people factor that we will never be able to get rid of. And so my advice is...

Keep people at the center of what you're doing and trust that people can grok what they need to, to be able to make really good decisions with the technologies that we're using in that. If you can encourage alignment with your strategies, you can unleash a level of productivity and a level of curiosity that we have not seen before. So strive for alignment, stay curious, keep people in the loop.

Maulik Sailor (39:55)
Thank you.

Ajit (39:56)
Did you guys see the recent MIT study came up about adoption of LLMs in this engineering teams how it's like 90 % of those trials have been entered in failure. So I think maybe this question kind of relates to that. feel I think I have not actually gone through the reasoning because just the numbers I have looked at. But I think the I think it's up to the leaders to

Find the fit and trade-offs like where LLMs or regenerative AI creates advantage versus where traditional means continues. For example compliance or very safety critical things. You are not going to leave it to AI at this moment. You can use it to some extent but not leave it to AI. ⁓

Yeah, I think that's a big question. So when creating strategy for the team, think the setups and the fit, those are two aspects you need to look at really well.

Maulik Sailor (41:05)
Okay, cool, wonderful. Now I'm just conscious of time, right? And I did say that I have a question that I'm going to reserve towards the end, right? And I'm gonna bring it over here, right? Okay, be prepared for that, right? I've not asked this question ⁓ to any of the guests yet. ⁓ And it's fairly new things that we have been evaluating recently, okay?

How do you guys think about, you know, like as part of your engineering teams, like the people side, you know, how do you think about employing digital labor to work alongside the actual humans? Right? Right. That's one side as an employer, right? Or as a, you know, team person in charge of running a team, right? But also as an engineer yourself,

Do you think you would actually would like to clone yourself as an AI clone, which can do most of the mundane tasks for you so that you as an individual can focus on high level, high value productive ⁓ activities? What do you guys think?

Ajit (42:21)
Maybe I can go first. So I think the question like can an AI agent be one of the engineers in your team? I think that's the starting point. Whether you want to clone or not, that's slightly different. But can we have an AI agent and engineer in the team? I think it's definitely going to be beneficial for not just

Maulik Sailor (42:23)
Yeah.

Robert G (42:23)
Go ahead, Edgy.

Ajit (42:48)
it's for the team and for the organization. For example, think about like the team is loaded with a lot of on-call tasks today. A lot of these teams who who interface with systems that needs to be up all the time. You can have ⁓ this kind of as one of the engineers who does all this triaging on-call first work. Second, so I have seen a lot of teams

⁓ the days

What happens is like suppose they have PRs, they're submitting commits and they don't have people actually doing reviews in time. And that kind of takes a lot of productivity. But there also you can have this person doing a first level scan, giving some scores for somebody else to come and do a final review, which would be much faster than if you wait for other people. So I'm saying that there are some gaps or opportunities for this application of AI agents in the teams.

Robert G (43:52)
Okay, so I'm gonna give you my opinion. I've had an opinion for a while now that AI is your additional team member. They're another part of your team or a seat in the team, And my more recent opinion has grown into, I could see engineers turning into orchestrators of agents. And what I mean by that is I've used AI so far in most areas of SDLC.

all the way from the idea phase that PM employees through the support phase of defect remediation. I can absolutely see a world in which an engineer's job is to orchestrate the full development of something from idea phase through the supporting of that as a singular person that manages a group or orchestrates a group of AI agents. I don't know that we're there yet. I think right now I could absolutely see a

Like an architecting agent, can see a code assist agent. I can see a code remediation agent, a code review agent, a documentation agent being on the team where it offloads all of the busy work and the mundane work that engineers have done in the past. Cause I'm going to ask, I'll ask engineers watching this right now. How much do you enjoy writing unit tests? How much do you enjoy writing your documentation and covering your API? How much do you enjoy creating architecture diagrams?

Like I don't think that's the most enjoyable part of what we do. I'm sure we, some of us enjoy it, but most of us enjoy solving those really hard problems using our minds and throwing some code at it or some other pieces at it at some point in the process, right? The problem solving part of it is where we excel. The additional layers of effort that we put into it, those are the parts that most of us don't want to do regularly. And that's where agents come in and help unload a lot of that off our plate.

so we can focus on the problem solving pieces. So to your first question, employing an agent as a digital teammate, I can see that, I 100 % can see that. As it relates to cloning myself, and I'm gonna take you at your word there, that term. No, I don't see myself as wanting to clone me because I think I can barely put up with myself as it is. But the truth is, I think there's tools out there that

handle the busy work for us and unleash us to do the work that we want to do.

Maulik Sailor (46:17)
Yeah, cool. So you know the cloning part, know, you have tools which can like, you know, the PR and code reviews and you know, unit testing. I think that I've seen some AI tools demo ⁓ which can do it for you. But you know, the reason why I was getting intrigued about cloning is because see we as ⁓ humans, all of us have personal experiences, right? Which is why we all are different, right? And secondly,

we all have a piece of knowledge that we know nobody else know in the world. Right. So if I were to clone my agent, then my agent is going to be different than what's out there. It's going to be always be like, whatever the public or internet knowledge is on top of that, there will be a small Delta, you know, which is what will make my clone a little bit unique. And so it's for everybody. Right. So then depending on the task at hand,

the task might be better performed by my clone than generic AI agent. That was my thinking that, if I were to replicate myself as a clone, why would I want to do it? Because I think I have a piece of knowledge which is not there on the internet. And so is everybody else. So every one of us are actually a huge knowledge.

⁓ you know, repository, which is not online right now. Right, it's not in any online database.

Robert G (47:50)
So.

Ajit (47:52)
So question.

Robert G (47:53)
So

what you're saying is that we become the context engine, right? That's what you're saying? Yeah. Go ahead, Ajit.

Maulik Sailor (47:57)
Correct, yeah, yeah, basically.

Ajit (48:00)
So, more like the digital twin of you that you are kind of envisioning. Will it have the autonomy to take action on your behalf?

Maulik Sailor (48:11)
I don't know, but I would say not 100%, not 100%. But things like that, know, like, look, oftentimes I find myself, you know, class for times, right? That, okay, I need to attend two meetings, you two team members need my attention. I can't do that, right? So I'll say, okay, fine. Okay, Claude, you go there, get the notes, give me the insight, right? ⁓ If it's about this, make sure you get this as an outcome. You know, I can just pre-program that, right? And I can attend the other one.

Right? Whatever, right? I mean, that's just a thought, you know. ⁓

Ajit (48:44)
So it's more like passive

work that they can represent you.

Maulik Sailor (48:48)
Passive

work, also like, know, for example, you know, if I take off ⁓ my background, my career, I've been involved in a lot of early stage zero to one startups many, many times. Right. And I have seen that a lot of times, like, you they are all different, you know, I've worked in different segments, but that is a, there's a piece of work, which is kind of same in all the startup, regardless of the domain or whatever.

And I'm like, know, why do I have to keep doing this over and over every time, you know, there's a new startup I'm doing, right? Instead, I can just have my clone do that for

That's an example. Why MyClone? Because MyClone will do the way I want to do it, not the way chat GPT will tell it to do.

You see the difference, right? ⁓ And as a human, we are all a little bit flawed, right? We are all flawed. So my agent will also be a little bit flawed. It won't be perfect. And that is what will make it more like me. It's my extension. You're still working with me, but it's my extension.

Ajit (50:00)
think there will be a black mirror episode and how things go wrong from here. ⁓

Robert G (50:07)
I was thinking the same thing, Ajit. That's coming straight out of Black Mirror, man.

Maulik Sailor (50:12)
Yeah. But what you know that

Ajit (50:14)
But I definitely see

how it can be useful. see some people actually send their digital selves to Zoom meetings. But I have not seen it taking real action on behalf of them rather than just taking notes and listening. But to capture your full personality, you have to give it full memory of everything you have done in the past. So it can then extrapolate.

Maulik Sailor (50:22)
Yeah.

Yeah, I've been...

Yeah, so my

thought is to flip around the whole. So something that I'm experimenting right now, right, is to flip the whole LLM concept. So right now, LLM is basically you have all these models trained up by Entropic and Chess GPT and Google and everybody else. They have the base model working. We go in at the user, and then we basically give over prompt. And the prompt gives the response based on the training that has happened.

I'm looking to flip it around the other way, where the database or the underlying knowledge or information is from me, is mine, something that I am creating and I am basically uploading. It may not be online, it may be my historic project that I have done, maybe I have to just recreate some of the artifact. ⁓ I give the prompt, ⁓ I basically train it to give the response, okay, if somebody asks my clone about this particular thing, make sure you reply in this way.

the internet when they talk to me, you know, my clone will give the response that I would give as a human.

Ajit (51:45)
Yeah, yeah, yeah. You know, like what you're saying, it can be directly applied to the previous discussion we had. So for example, you have all the history for your team provided to the agent, like it's stored. So then when you start a new project, it can say, hey, based on your past history, you have been making this kind of mistakes. Why don't you start with having redundancies or some kind of contingency plans for this kind of situations? And same for the individuals, like us.

gave us that kind of...

Maulik Sailor (52:16)
Yeah, so I had this talk recently, you me and Sachin, we also spoke about that. That, know, if we were to have a clone, you know, what would we want our clone to do basically? And yes, I would like it to do everything, you know, and for me to relax on a beach, you know, but that's not possible. and then I will get really bored, right? So ⁓ like, you know, what, you know, what can we have this clone to do? And

At this moment, we don't have all the answers, but it's something that I've been recently experimenting. That, OK, if I were to do this, how would I do it?

Ajit (52:53)
So OpenAI, I think some of these ⁓ companies also are building, they want to store all the information you provide, all the interaction you had, so that they can customize, personalize the answer for you. Yeah.

Maulik Sailor (53:07)
Yeah, but that's exactly the point. Because if

I tell OpenAI about my, like, you know, whatever knowledge I have, it's going to become a public knowledge. And I don't want my unique experiences to be a public knowledge. You know, because that's what makes us unique, right? Just imagine, right? The reason you are in a role, ⁓ the job or role you are currently doing is because you as a person, you know, whenever you interviewed for that role, ⁓ they have probably made a decision that your background, your experience, your knowledge is relevant to

Ajit (53:11)
yeah yeah yes true yeah

Yeah.

Maulik Sailor (53:37)
the project that they are doing, right? ⁓ Yes, there might be some other candidates or somebody else as well, but they selected you because you're like a unique take kind of match with what they were looking for.

Ajit (53:50)
This is a good start up idea. You can have some enclave where you can store the personal memory or like all the data and then you have LLF.

Robert G (53:50)
Hey, Malik?

Maulik Sailor (53:59)
Robert, I believe you

want to go.

Robert G (54:01)
I need to drop because I have another engagement I need to get to right now.

Maulik Sailor (54:06)
Cool, nobody's nobody. Robert, lovely to have you, but I need to catch up with you properly separately, so I'll drop you a message. But Robert, I'll let you go. Thanks a lot for joining this live. I think it was a very interesting topic. We are still running more than an hour, so it just shows how interesting the conversation is going.

Ajit (54:07)
Thanks.

Robert G (54:12)
Very good.

Indeed, it was a good conversation. Thanks, Ajit. It was very nice meeting you. Sashim, take care. Bye, Malik.

Maulik Sailor (54:29)
Thank you. Thank you, Robert. Thank you. Yes, yes, did. Like you were saying something.

Ajit (54:29)
Thanks, John.

Yeah was saying this could be a good startup idea because what you saying is true that you don't have to provide all this information to any company. Not even like if you build a company nobody is gonna like nobody will like to provide this information but if it can be separated it's stored so that the company whoever is providing the service they don't have access and the people can use it as this as their digital twin.

Maulik Sailor (55:01)
Yeah, correct. Right. So anyways, you know, it's a fairly new concept we've been we've been experimenting. So hopefully we'll have something available soon. That's what I can say. Yeah, cool. Anyways, I think just mindful of the time here as well. You know, thanks a lot for joining today. You know, it was really wonderful for you to have you on our podcast. You know, great insight, you know, all the very best with your MBA.

that you are doing at Berkeley. I'll be next up actually fairly soon, possibly next month. So hopefully it would be great to see you in person actually.

Ajit (55:36)
Yeah sure ⁓ yeah we can chat up chat up I chat up this. Thanks Malik thanks Sachin.

Maulik Sailor (55:40)
Cool. All right, cool. All right.

Thanks a lot, Ajit. That's it, folks, for today. We'll call it a day for today. And thanks a lot for everybody who joined in ⁓ on this podcast. Thanks a lot. Thank you. Bye.

‍