From Windows Kernel to AI-Native Platforms: Scaling Engineering Teams and Systems with Puneet Saraswat

In this special season 3 finale of ShipTalk, host Dewan Ahmed sits down with Puneet Saraswat, VP of Engineering at Harness, for an inspiring conversation about engineering leadership, system design, and the evolution of software delivery.

What You'll Learn:

Career Journey: Follow Puneet's path from discovering programming in a hot computer lab in North India to building Windows 7 kernel components at Microsoft, then transitioning to Azure's distributed systems, and finally joining Harness as one of their first 15 engineers.

Production Systems at Scale: Get insights into building reliable systems that serve millions of users, including lessons learned from Windows kernel development, Azure's global infrastructure challenges, and the evolution from monolithic to distributed architectures.

AI-Native Software Delivery: Explore how AI is transforming the software development lifecycle beyond just code generation - from automated pipeline debugging to intelligent remediation and best practice recommendations.

Engineering Leadership: Discover strategies for scaling engineering teams, balancing velocity with security and compliance, and transitioning from individual contributor to management roles.

Beyond the Code: Learn about Puneet's passion for indoor climbing 🧗‍♂️  and how lessons from physical challenges apply to navigating corporate growth and continuous learning.

Key Takeaways:

• Why fundamental computer science concepts matter more than ever in the age of rapid technological change
• How to build "cellular architectures" that prevent regional failures from cascading globally
• The importance of hiring generalists with strong analytical skills and growth mindsets
• Strategies for managing the balance between AI-accelerated development and production reliability
• Practical advice for engineers considering the transition to management

Perfect for software engineers, engineering managers, DevOps professionals, and anyone interested in the intersection of AI and software delivery platforms.

Guest Bio: Puneet Saraswat is VP of Engineering at Harness, where he leads platform engineering and SRE teams. With over 15 years of experience, he previously spent nearly a decade at Microsoft working on Windows kernel development and Azure infrastructure. He holds a degree from IIT Kanpur and is passionate about building scalable systems and developing engineering talent.

Transcript

WEBVTT

1
00:00:03.280 --> 00:00:19.300
Dewan Ahmed: Hello, Hello, welcome to the ship. Talk. Podcast today is season finale of our season. 3. And we have a very special guest with us. Puneet puneet is the Vp. Of engineering at harness. Welcome punit.

2
00:00:20.110 --> 00:00:23.123
Puneet Saraswat: Yeah, thanks, Devon. I'm excited to be here

3
00:00:23.980 --> 00:00:28.050
Dewan Ahmed: Likewise super excited. I like the swag you have on you.

4
00:00:28.450 --> 00:00:32.610
Puneet Saraswat: Yeah, we have the same. I think we both bleeding hardness today.

5
00:00:32.610 --> 00:00:39.699
Dewan Ahmed: Yes, yes, I know you mentioned. It's a pretty warm in the office, so I I appreciate the dedication to to put on the swag.

6
00:00:40.250 --> 00:01:01.379
Puneet Saraswat: Yeah, like, when I started it was all cloudy and then, now I see it's it's kind of all sunny. That's barrier for you. But yeah, I like, I love the slack. I've been at harness for 7 years. So we we get so much swags. And like, it's it's kind of the the joke in the office, like every day you can repeat, you can have harness, swag, and don't don't repeat, you know. Like so

7
00:01:02.010 --> 00:01:05.169
Puneet Saraswat: yeah, but that's kind of life, you know. Harness.

8
00:01:06.340 --> 00:01:28.830
Dewan Ahmed: Yeah, yeah, absolutely. And I have this in. My wardrobe is full of Ibm red hat, Ivan. And now harness. So from socks to T-shirts to hoodies. I have a backpack. I have a luggage so I have to actually find like where I don't have a water bottle. So I have to now find out where I don't have a a swag on a on a product, and then, probably like, get that next time.

9
00:01:29.320 --> 00:01:51.749
Puneet Saraswat: And maybe you're aware of it like it. We have this life at hardness, slack channel, and where people, you know go on vacations, and they'll like all kind of exotic places they'll put like hardness bottle for, and the swag there, and you know, like hardness has been there. So I see like so much company pride in the in the the team there, and swag goes a long way building that you know.

10
00:01:52.290 --> 00:02:13.779
Dewan Ahmed: Absolutely, and I can already sense the passion you have for the company and for engineering. But we want to learn where that passion 1st start like, where did the passion for engineering or problem solving start? Was it at Iit? Was it way back before then? Would you guide our listeners to your journey into Tech.

11
00:02:14.860 --> 00:02:21.159
Puneet Saraswat: Yeah, I think the so. My journey started like very early. So in the school.

12
00:02:21.661 --> 00:02:30.599
Puneet Saraswat: I think it started like 7th or 8th grade when I got to use the computer for the 1st time, and that was like in the

13
00:02:30.730 --> 00:02:43.979
Puneet Saraswat: in the lab we had in our school, and that used to be like one of the. And like, I come from India up. So North India. So it's like a hot region. So the computer lab was one of the the

14
00:02:44.850 --> 00:02:48.750
Puneet Saraswat: that sought after place of rescue

15
00:02:48.960 --> 00:03:05.769
Puneet Saraswat: from the summer heat there, and that's where, like all us kids will will go there and then be in the computer room and program. And we started programming with basic and primary thing was, okay, this is is the the room you kind of escape from the heat.

16
00:03:06.430 --> 00:03:17.860
Puneet Saraswat: And early I found, like the programming was like the kind of feedback you get and instant feedback. And what you create, you can see the the results of it like that

17
00:03:18.030 --> 00:03:19.580
Puneet Saraswat: gave me that kind of

18
00:03:20.106 --> 00:03:30.799
Puneet Saraswat: early no interest in programming, right? Like that's where, like on the summer, I remember in the 8th grade took a summer project to build a

19
00:03:30.920 --> 00:03:35.160
Puneet Saraswat: school reporting system which is like report card building system.

20
00:03:35.450 --> 00:03:51.249
Puneet Saraswat: And we we programmed on basic with one another friend of mine and and it was super fun, you know. So this whole ability to program and see the result. And I trade, and that kind of focused me up in intact.

21
00:03:51.580 --> 00:03:55.199
Puneet Saraswat: And of course, like Iit was a formative time

22
00:03:55.640 --> 00:04:02.460
Puneet Saraswat: we can talk more about it. But like once I went into, it's like it's like very high caliber.

23
00:04:03.730 --> 00:04:18.149
Puneet Saraswat: People from all over India, you know. Come at one place, and you know, like you have a lot of fun and learning, and you know, like, so that kind of also cemented like next level of that's where I get got into comm science

24
00:04:18.795 --> 00:04:24.260
Puneet Saraswat: did some AI machine learning? Those were that was like around

25
00:04:24.590 --> 00:04:27.730
Puneet Saraswat: more than 20 years ago, like, still, we were doing

26
00:04:28.537 --> 00:04:33.732
Puneet Saraswat: things like background segmentation, and my thesis, and and

27
00:04:34.640 --> 00:04:48.420
Puneet Saraswat: the the final year thesis project was, you know, image segmentation of like moving parts of the bodies, and you know, it was featured extraction on the images and computer vision and machine learning intersection of that.

28
00:04:49.380 --> 00:05:03.799
Dewan Ahmed: Well, and and that was like way before there was a hyper on machine learning and AI, and even in academia was it common for folks to to use AI and machine learning in projects back, then.

29
00:05:03.940 --> 00:05:17.499
Puneet Saraswat: No, it was not that popular but it was coming along it. It has not, captured the the public imagination, as you know, but in the academic circles still machine learning

30
00:05:17.700 --> 00:05:35.010
Puneet Saraswat: neural networks there was no deep neural networks were not the thing, but we were learning artificial intelligence used to be the introductory course in in Iit Kanpur, and there we we would learn things like feature extraction. We would learn things like

31
00:05:35.562 --> 00:05:39.540
Puneet Saraswat: background segmentation from in the in the pictures, and

32
00:05:40.292 --> 00:05:52.779
Puneet Saraswat: at that time you used to use Mat Matlab and see and division libraries were there. Opengl had come. So they were like academia had

33
00:05:53.450 --> 00:05:58.420
Puneet Saraswat: the action going on, but it had not, like really become the the thing, that which which

34
00:05:58.750 --> 00:06:01.230
Puneet Saraswat: it became in like last 3, 4 years.

35
00:06:01.680 --> 00:06:09.902
Dewan Ahmed: You mentioning? Matlab is bringing my electrical engineering days where we had a induction motor hooked up, and then we could run some

36
00:06:10.330 --> 00:06:36.139
Dewan Ahmed: simulations on a Matlab. And Matlab was a beast of a program that you really need a pretty beefy machine to run, Matlab. And also, things are not like these days, where you can just mostly use open source, and you can almost find a lot of free tools, especially in academia. You had a license, and you had to go to the lab computer to use it.

37
00:06:36.560 --> 00:06:38.639
Dewan Ahmed: Was it similar for you as well.

38
00:06:38.640 --> 00:06:43.950
Puneet Saraswat: Yeah, those are the fun days, you know, like, we used to have this computer lab again in Kanpur was a super

39
00:06:44.396 --> 00:07:10.749
Puneet Saraswat: hot weathers. And then computer Science Department is the place where you get the the respite from the heat. And we used to just lock into the in the department, and we used to have like this lab labs. Every Prof. Will have their labs, and there'll be all the the infrastructure available for you to know. So had these project like so Matlab was like

40
00:07:10.860 --> 00:07:14.433
Puneet Saraswat: one we use for this image segmentation project

41
00:07:15.565 --> 00:07:27.770
Puneet Saraswat: so we we would say, extract out the foreground object from the background, like the and I remember the the 1st one was we got a video, a series of images from our ta, and he was walking around

42
00:07:28.276 --> 00:07:36.383
Puneet Saraswat: on the campus I shot this video, and our task was to to segment him from the background, you know

43
00:07:36.940 --> 00:07:42.409
Puneet Saraswat: and then build that series of images through that segmentation. It was pretty fun project.

44
00:07:42.830 --> 00:08:05.160
Puneet Saraswat: Yeah. Those are the the days.

47
00:08:19.350 --> 00:08:28.659
Dewan Ahmed: And I think we are aging ourselves with those references. So, moving from from Iit to Microsoft, you have spent a significant amount of time at Microsoft

48
00:08:28.770 --> 00:08:36.900
Dewan Ahmed: building secure and scalable systems. What were some of your key takeaways into building those production grid systems.

49
00:08:37.650 --> 00:08:43.440
Puneet Saraswat: Yeah, I think, Microsoft was my 1st job out of college. And

50
00:08:44.034 --> 00:09:06.020
Puneet Saraswat: the 1st day I remember, like I I ended up in a in a team which was starting to build windows 7. So windows vista project got over, and there was, like all reshuffling, rethinking happening for windows 7 and windows vista. There was a lot of learnings. Microsoft was applying on towards windows 7, and that's where I got

51
00:09:06.500 --> 00:09:15.390
Puneet Saraswat: my 1st team located in the 1st 3, 4 years. I was just working on windows kernel building this networking device. Drivers.

52
00:09:15.530 --> 00:09:23.430
Puneet Saraswat: That was a lot of fun. That. That's where, like my, my training to production level systems happen. And you see, like windows that

53
00:09:23.580 --> 00:09:28.090
Puneet Saraswat: was like a the

54
00:09:28.270 --> 00:09:40.810
Puneet Saraswat: the operating system of the masses there. And what I learned there is like, there is no corner cases when you're building software for production system at that large scale.

55
00:09:41.030 --> 00:09:47.349
Puneet Saraswat: And there's some some interesting stories. And as as we went to

56
00:09:47.807 --> 00:10:02.100
Puneet Saraswat: release the release preview of windows 7. And then it went went out, and the kind of components we had a pretty good telemetry system there, where on the wild, if there is a issue which happens on a

57
00:10:02.612 --> 00:10:21.669
Puneet Saraswat: on a machine which, if you're in kernel mode programming that what you call blue screen of death or bug check systems. So, Microsoft, we had a pretty good telemetry system where anything had those hit into those bug checks will report back to the system, and it'll route to the right developer

58
00:10:22.523 --> 00:10:27.510
Puneet Saraswat: of what things have happened, and I remember, like those like

59
00:10:27.570 --> 00:10:57.380
Puneet Saraswat: sleepless nights, and week after week after we released it to Beta, where software has gone to the real users where we were sifting through all the issues getting deported. And we invest days and weeks to, you know, understand root cause, fix them. You could see the whole life cycle of this, the learning for me is in those days is there's no corner case, the the reliability of system at that scale. What like everything which can go wrong will go wrong

60
00:10:58.085 --> 00:11:10.470
Puneet Saraswat: right? So that's where we learned like attention to detail building system which you can debug after the fact, and you know, like so. But those are the fun days.

61
00:11:11.750 --> 00:11:27.779
Dewan Ahmed: You build a lot of reliable systems that those scaled. But were there any instances where your team probably worked on some sort of scale. But you hit a bottleneck. Things didn't go as planned. Could you share any of those situations.

62
00:11:28.590 --> 00:11:52.070
Puneet Saraswat: Yeah, I think. All the time, I think. The second stint for me at Microsoft was moving to the azure team. So in 2014, 2013, 2014 time, I moved to the the windows. Azure team is still was called windows azure. Later it got branded to azure, and Microsoft embraced

63
00:11:52.704 --> 00:12:04.869
Puneet Saraswat: non windows, ecosystem, linux ecosystem, very 1st class along the way. But that's where, like the some of the unlearning and relearning happened for me in in azure.

64
00:12:05.020 --> 00:12:21.320
Puneet Saraswat: where, in, say, windows, we were releasing every 2 and a half years so, and the milestones used to be 6 months milestones and your release vehicle is, you know, like you're shipping your windows bits either Cds or people are downloading

65
00:12:22.990 --> 00:12:25.130
Puneet Saraswat: the tower balls from Internet.

66
00:12:25.840 --> 00:12:37.770
Puneet Saraswat: Now, it was more like, you're doing continuous deployments. You know, your delivery is to the cloud. And then you have more control. And that's where, like lot of distributed systems.

67
00:12:38.738 --> 00:13:00.330
Puneet Saraswat: Training came for me. And this whole new paradigm of, you know, like delivery is different. The cloud delivery is different than you know, like Pre cloud era, and we built the scale was like, I was primarily looking into the the control plane of some of the the critical resources for azure.

68
00:13:00.490 --> 00:13:17.660
Puneet Saraswat: And we had to go from, you know, regional to global footprint. And how do we isolate, say, one region from other. One project, I remember was like, How do we have the blast radio? First, st of any fault limited to a given region.

69
00:13:18.030 --> 00:13:22.030
Puneet Saraswat: So what we learned is like. You cannot take dependencies on a

70
00:13:22.170 --> 00:13:29.070
Puneet Saraswat: you know, a global dependency on a project or any resource. And you know, how do you segment your

71
00:13:29.230 --> 00:13:41.480
Puneet Saraswat: your traffic and your data within the region what you could not explain is like, if a fault in, say us West region take down your service, and which is running in Australia.

72
00:13:41.720 --> 00:13:47.625
Puneet Saraswat: Alright. So with that like these days, people call it cellular architecture.

73
00:13:48.511 --> 00:13:53.289
Puneet Saraswat: those days that term was not there, but it was more like, how do you control the brass radius?

74
00:13:53.420 --> 00:14:06.479
Puneet Saraswat: And so those are the things where you see it inherently brings the scale, you know. Like, if you need more capacity, you just create another set which is like another region for you.

75
00:14:07.592 --> 00:14:14.998
Puneet Saraswat: But you, coming back to your question like things which didn't scale like all the time, you know, like our Apis were not

76
00:14:15.330 --> 00:14:43.120
Puneet Saraswat: scaling the way we were using synchronous Apis, and which were blocking the threads, and we had to kind of build this decouple system where the queue got introduced and services a new Api request will come and register the request, and it will go in the queue, and then the workers will scale, as as the number of outstanding requests increase, and those are the the early patterns which we started to learn and incorporate in our architecture.

77
00:14:43.750 --> 00:14:59.239
Dewan Ahmed: I want to touch down or touch on this point, because some of our listeners they're still in their school. There might be finally of their engineering degrees, and they might be thinking, Oh, I'm studying for my Java exam. And these design patterns. Do I actually get to use it?

78
00:14:59.380 --> 00:15:09.250
Dewan Ahmed: Yes, you do as puneez is mentioning about the the fundamentals. Fancy technologies come and go, but these fundamentals never change.

79
00:15:09.450 --> 00:15:11.279
Dewan Ahmed: Would you agree with that puneet.

80
00:15:11.590 --> 00:15:37.860
Puneet Saraswat: Yeah, absolutely. And especially like, now, there's so many building blocks of where and then architecture patterns which have emerged in last 1015 years. But yeah, like all these things matter at the Internet scale technologies, you know. Think of, like, how do you plan for the surge in the traffic? You cannot keep your capacity for for the search and the way you do it is you you decouple

81
00:15:37.980 --> 00:15:46.060
Puneet Saraswat: from like your your running systems, from your request by introducing this, this concept of queues and workers.

82
00:15:46.502 --> 00:16:03.980
Puneet Saraswat: And then you you have the flexibility of as as the work is coming the queue size increases, and you can scale your workers accordingly, and that that is like a pretty standard part pattern which has come out in the distributed systems world and the cloud services now.

83
00:16:04.546 --> 00:16:19.789
Puneet Saraswat: was not the case when we were doing windows, you know. Like, because when you're running windows, you're limited by the hardware you are running on. So this thing is in the cloud era, where your compute is. Like a lot more scalable.

84
00:16:20.308 --> 00:16:28.419
Puneet Saraswat: And how do you kind of use that compute capacity when you're not restricted to one one system, you know.

85
00:16:29.440 --> 00:16:39.610
Dewan Ahmed: It seems like you had a pretty good run at Microsoft. You're understanding the systems. You are already pretty deep into your engineering career at at Microsoft.

86
00:16:39.800 --> 00:16:40.660
Dewan Ahmed: Then

87
00:16:41.060 --> 00:16:48.019
Dewan Ahmed: what made you switch to harness like? How did you hear about harness? And what was your why, now, moment.

88
00:16:48.820 --> 00:17:14.959
Puneet Saraswat: Yeah, I think that's a interesting story, like the time I was running this team in azure, we the team was called Application gateway, and that's like serving all the like. If you're running Http workload. And in your virtual network you put application gateway in front, and it gives you kind of Ssl offload and routing capabilities Http and waf, and all that good stuff.

89
00:17:15.755 --> 00:17:27.029
Puneet Saraswat: and and we saw pretty solid growth, you know, like it like last, like 4 years we we launched the service, and we were running a fleet of like

90
00:17:27.655 --> 00:17:31.670
Puneet Saraswat: tens of thousands of virtual machines which are running these

91
00:17:32.230 --> 00:17:40.300
Puneet Saraswat: and load balancer systems. And and for the and it's still a small team which is operating it. So what I

92
00:17:40.590 --> 00:17:49.949
Puneet Saraswat: so like the operational part, and the toil was real, you know, like we've had to. We had this cadence of bi-weekly releases for our gateways.

93
00:17:50.310 --> 00:17:58.509
Puneet Saraswat: and and we had to roll it out across the fleet of, you know, like azure regions which were like dozens of region those days.

94
00:17:58.982 --> 00:18:12.299
Puneet Saraswat: And you have to plan this whole activity every 2 weeks, and what and then deployment and patching and upgrades was like as a part of the rhythm for a small team by the time you're done with one

95
00:18:12.809 --> 00:18:27.079
Puneet Saraswat: the second was was already upon us, and then, if something happens wrong and then you have to roll back you you kind of miss the cycle. And there was a lot of, you know, like operational toil and firefighting. And

96
00:18:27.080 --> 00:18:45.460
Puneet Saraswat: you can imagine, like running this fleet of tens of thousands of virtual machines, keeping them updated, keeping them like monitoring them. And so I could see, like, if we were there itself, trying to build some systems to improve the operational aspect which is primarily it was like deployments.

97
00:18:46.027 --> 00:18:50.170
Puneet Saraswat: Were a big challenge, right? So when I

98
00:18:50.650 --> 00:18:54.730
Puneet Saraswat: in 2017, when, like harness came out of stealth. And

99
00:18:55.223 --> 00:19:10.790
Puneet Saraswat: I got in touch with Rishi CTO that time. And he explained to me the problem they're trying to solve. And I could identify like that was the problem I was living day in, day out, at at azure that time.

100
00:19:11.360 --> 00:19:33.099
Puneet Saraswat: so I I could see the potential of the idea and seem like problem is for everybody not just like my team at at Microsoft and and also, I wanted to do something. In the smaller setup, you know. Startup was always the thing. So that's the time I I can decided to to join the ship

101
00:19:33.679 --> 00:19:44.849
Puneet Saraswat: and we were very small that time. It was like January of 2018. I I on boarded here like we were like around 15 engineers that time. Few beta customers.

102
00:19:45.000 --> 00:19:52.309
Puneet Saraswat: And yeah, we were just hiking around. We were, you know, building the CD as our 1st product those days.

103
00:19:54.470 --> 00:20:03.820
Dewan Ahmed: And the work you are doing like when you joined harness. How much it overlapped with the work you are doing at Microsoft, like, how much new stuff did you have to learn.

104
00:20:04.610 --> 00:20:13.949
Puneet Saraswat: Yeah, it was completely new in a way, like I was doing cloud infrastructure work So it was still a layer.

105
00:20:14.060 --> 00:20:22.069
Puneet Saraswat: I was building the components which you use to build your applications. If you will. And there were some applications we were running in the control plane there.

106
00:20:23.370 --> 00:20:27.170
Puneet Saraswat: But what hardness was hardness where we were building this?

107
00:20:28.341 --> 00:20:33.479
Puneet Saraswat: very distributed orchestration system, which you can do declarative

108
00:20:34.108 --> 00:20:39.291
Puneet Saraswat: pipelines. And we can help you build this advanced deployment

109
00:20:40.551 --> 00:20:52.109
Puneet Saraswat: strategies which you can implement with, you know, like, without going doing the the grunt work of writing those like scripts which which used to be the norm people were doing before.

110
00:20:52.340 --> 00:20:58.668
Puneet Saraswat: So the work was, primarily like the nature of work, was different. But

111
00:21:00.020 --> 00:21:15.879
Puneet Saraswat: fundamental building blocks were not, you know, like we were still building cloud services. We were deploying continuously. We were delivering through. You know, Cloud. So there were a lot of parallels. But I would say it's from the in

112
00:21:16.160 --> 00:21:23.140
Puneet Saraswat: azure. I was building the infrastructure services here. It became more like a Sas service for deployment.

113
00:21:23.890 --> 00:21:26.319
Dewan Ahmed: So one level, probably up.

114
00:21:26.320 --> 00:21:27.060
Puneet Saraswat: Yes.

115
00:21:27.480 --> 00:21:35.819
Dewan Ahmed: And so back then, team of 15 engineers. Now you're leading dozens, if not 100 of engineers.

116
00:21:35.940 --> 00:21:39.340
Dewan Ahmed: How does your day to day work.

117
00:21:39.580 --> 00:21:49.630
Dewan Ahmed: Look now. So, for example, a Vp of engineering, how do you start your day? How do you end your day? What are some common things you do.

118
00:21:50.720 --> 00:21:55.440
Puneet Saraswat: Yeah, I think, as teams say, Yvonne, like they give you

119
00:21:55.660 --> 00:22:11.760
Puneet Saraswat: tend to build like multiple teams, and which you are running and which are aligned for different part of the business. So at this point, I'm running the platform engineering team and sre team at harness and

120
00:22:12.296 --> 00:22:19.700
Puneet Saraswat: so the way our like day starts is like, okay, like a lot of work here is like a strategy and doing the alignment for

121
00:22:19.870 --> 00:22:25.110
Puneet Saraswat: what is team working on, and how like business

122
00:22:25.614 --> 00:22:38.550
Puneet Saraswat: is aligned to that. So a lot of work is around that like how we plan what projects to pick are these the projects which are kind of moving the the needle in right direction? A lot of work is like, is there cross

123
00:22:38.660 --> 00:23:05.569
Puneet Saraswat: cross team collaboration happening? These, like we are building on each other's ideas versus reinventing the the same thing in multiple places, you know. That's kind of what we do in especially in the platform teams. You build the core technology. So the the teams using you can leverage it versus like they. They have to build their own and reinvent the wheel and waste the the efforts there.

124
00:23:05.680 --> 00:23:08.790
Puneet Saraswat: So a lot of work is about

125
00:23:09.180 --> 00:23:23.600
Puneet Saraswat: understanding this internal stakeholders, understanding what their needs will be, solve them today. Versus so they, they are able to capitalize and accelerate on their delivery velocity.

126
00:23:23.760 --> 00:23:52.419
Puneet Saraswat: Also, other part of it is like, also, I have good fortune to run like, say, platform engineering. So think of this as a matter of problem, for harness harness is a platform for platform engineers or devops people and running that part for internal harness teams. So we are running our own like, if you will, instance of harness which we, from our own code repositories to our own Ci system, to CD systems to

127
00:23:52.921 --> 00:24:00.550
Puneet Saraswat: so all the code to production, what we enable our customers, we are using harness to do it for our building harness itself.

128
00:24:00.930 --> 00:24:02.000
Puneet Saraswat: So

129
00:24:02.210 --> 00:24:13.309
Puneet Saraswat: lot of the things are like what we are building, we, we are the 1st user of it. And then we have that front wave row seat of, you know, like what to build as a user in the product.

130
00:24:13.938 --> 00:24:27.219
Puneet Saraswat: Right? So a lot of work comes in, you know, like seeing what's working doubling down on that, seeing the alignment across teams making sure we are doing the right things for the business.

131
00:24:28.760 --> 00:24:43.330
Dewan Ahmed: Harness is known as the AI native software delivery platform. So you mentioned about scale, you mentioned about security. How are you solving the challenge of adding AI and automation into this software delivery platform.

132
00:24:44.320 --> 00:24:47.601
Puneet Saraswat: Yeah, I think every like you can't do things with

133
00:24:48.665 --> 00:24:51.830
Puneet Saraswat: without the AI anymore, like, there's so much

134
00:24:52.570 --> 00:24:57.630
Puneet Saraswat: shift coming. And with the the capabilities of Agent Aki. Now

135
00:24:58.283 --> 00:25:24.446
Puneet Saraswat: so think of, like the way we think about it is like with this agents available at at your fingertips. How can you make developers more productive? How do we streamline code to production part? Still, there is a lot of like time, which, estimates is like less than 40% of the time for a developer goes and actual coding. And there's a lot of

136
00:25:25.225 --> 00:25:50.760
Puneet Saraswat: innovation happening in the coding side of things with the co-pilots and cursors and windsurf, and what we are looking at is after the code is done, for the code to be available in production is still, there is a lot of process and toil involved in the system, and what we are thinking about is with like harness becoming is that platform which has all kind of tools available to help you in that life cycle of code to production.

137
00:25:50.910 --> 00:25:51.950
Puneet Saraswat: The next

138
00:25:52.468 --> 00:26:11.621
Puneet Saraswat: unlock in that will be like, how can this agent take AI or AI agents are? Gonna take away some of the work which we're doing today you're still triggering the pipeline. If the pipeline are failing for one reason or other, can agents work through it automatically remediate or

139
00:26:12.080 --> 00:26:25.590
Puneet Saraswat: or help? You understand debug, if there are security scans in your pipeline, and you're finding vulnerabilities. Can we can agents go proactively, figure out the remediation and make it for you

140
00:26:26.360 --> 00:26:36.760
Puneet Saraswat: if right like. So those are the kind of things like, can you be like help? Can these agents help you configure your sdlc pipeline?

141
00:26:37.370 --> 00:26:47.800
Puneet Saraswat: with just your intent in the natural language. Right. Also help you discover, like the state of the art.

142
00:26:48.326 --> 00:27:15.430
Puneet Saraswat: In Sdlc practices based on the usage they are seeing across the customers, and give you kind of the best in the world pipelines for you, right like. So those are the the use cases we are looking at. Right? So that that's kind of where it is like. Ultimately, it's like, with this, agents available for you in sdlc lifecycle. What are the different scenarios? We can take away the time.

143
00:27:16.340 --> 00:27:30.449
Dewan Ahmed: And I can relate to that as well, because the work I did as a devops engineer before meant that if I run a pipeline. If something's wrong, I have to look at the the pipeline logs like download the the log file and then do some sort of grip or

144
00:27:30.620 --> 00:27:51.009
Dewan Ahmed: some Linux commands to find out exactly where the issue is. Now, if my harness pipeline fails, there's harness. AI. I can click a button, and then that tells me exactly what the issue is with my pipeline. So 1, 1 more ways. How software delivery problems are being solved with. With AI.

145
00:27:51.310 --> 00:28:15.929
Puneet Saraswat: Yeah, there you go, right like pipeline becomes yet another thing for you to maintain. Like you are changing some things. Pipelines are always evolving. Can this agent take frameworks? Help you, you know, like reason through that, and figure out the things you're making a change in one place? What will be the impact on the pipeline if you're running in production, right? So yeah, there's a lot of potential to bring this agentic capabilities.

146
00:28:16.640 --> 00:28:41.619
Dewan Ahmed: So we're talking about AI, and I think it's a nice segue to discuss future trends where industry is moving, based on your experience like, how do you see engineering teams balance the velocity because they have to incorporate AI. They have to encrypt the latest library, but they also have to be compliant. They also have to make sure that their software is secure.

147
00:28:41.870 --> 00:28:46.539
Dewan Ahmed: How have you managed to balance these 2 ends?

148
00:28:47.400 --> 00:29:02.570
Puneet Saraswat: Yeah, that's a very interesting question. And Judy is out there like how things will evolve. One thing we definitely see is with all this AI. Agents and co-pilots. It has become a lot more easy for you to write a lot of code fast.

149
00:29:02.820 --> 00:29:03.860
Puneet Saraswat: Oh.

150
00:29:05.290 --> 00:29:21.170
Puneet Saraswat: but what for running a production system. It's not writing code is just one part of it like you have to make sure code is reliable and secure right? And that's where we see more code doesn't necessarily make it

151
00:29:21.340 --> 00:29:28.140
Puneet Saraswat: more reliable and more secure, if at all, it becomes even harder to understand and maintain more the code you have.

152
00:29:28.570 --> 00:29:42.369
Puneet Saraswat: So that's where, like the next level of unlock would be. How can you bring this agent take AI or AI tools to also reason through the security of the code and reliability of the code, like people are talking about like

153
00:29:42.430 --> 00:30:01.169
Puneet Saraswat: code, documentation and test authoring of the code. And all right, like all those are good signs, but ultimately running those the system in production, security, reliability is the key. And how do we? And with the more and more code coming. I think there will be more

154
00:30:01.220 --> 00:30:14.760
Puneet Saraswat: need of having this AI systems help you make sure those are like the things are secure. That means what it means like if there are more checks in the pipeline to auto remediate

155
00:30:15.557 --> 00:30:24.830
Puneet Saraswat: the security issues. And also pick figure out the the the reliability patterns

156
00:30:25.950 --> 00:30:37.659
Puneet Saraswat: early on from your production systems and recommend what in your architecture is the weakness and right like. So there is a lot of systems engineering which will need to

157
00:30:38.060 --> 00:30:43.355
Puneet Saraswat: come to the next level in the AI world. You know, coding is just one part of it.

158
00:30:44.330 --> 00:30:59.699
Dewan Ahmed: For sure. So before there were a team of software engineers writing code. And then there were a few Qa or testers. Now, with the influx of humans and machines writing code. Do you see? There's an imbalance of

159
00:31:00.200 --> 00:31:20.880
Dewan Ahmed: building code versus testing code. And and if yes, do you see, there's a potential gap for that in the future, and making sure we also have either more human testers or make sure that AI is also scaling our testing of the huge amount of code that machines are generating.

160
00:31:21.760 --> 00:31:23.710
Puneet Saraswat: Yeah, I think it will be.

161
00:31:24.740 --> 00:31:34.570
Puneet Saraswat: The case will be, AI will be writing more tests as well. Right? And that's where, like, I think, multi agent system where they can work with

162
00:31:34.994 --> 00:31:43.510
Puneet Saraswat: the coding agent and testing agent. I think that is the the natural route it will be but at the same time it will be important for us to have

163
00:31:44.088 --> 00:31:54.200
Puneet Saraswat: engineers who understand the code. Well, in case something like all these systems are still evolving. So if something goes wrong and you have a mission critical system.

164
00:31:54.947 --> 00:31:56.860
Puneet Saraswat: You have to.

165
00:31:57.544 --> 00:32:04.645
Puneet Saraswat: still be able to reason through a root cause, fix the issues in production while you're gaining velocity.

166
00:32:05.180 --> 00:32:11.630
Puneet Saraswat: the understandability and able to operate at production scale. Still, I have.

167
00:32:11.800 --> 00:32:12.780
Puneet Saraswat: Oh.

168
00:32:13.480 --> 00:32:27.590
Puneet Saraswat: I think there will be need for, like more seasoned engineers who understand the basics of the system. That's where, like system engineering running at scale with high reliability, comes in picture.

169
00:32:28.870 --> 00:32:32.829
Dewan Ahmed: Shifting gears. I was told that you're into indoor climbing.

170
00:32:34.090 --> 00:32:43.219
Dewan Ahmed: Would you like to tell our listeners and viewers about that, and a follow up to that would be, how do you relate that to climbing corporate ladders.

171
00:32:44.494 --> 00:33:10.369
Puneet Saraswat: Yeah, that that's a good analogy. I don't know. So like, if you think like, I started like indoor climbing few years ago one of my like few folks in my team at Microsoft. They used to do the indoor climbing, and like we as a group used to go and that's where I picked up, and when but I paused, for after I left Microsoft and

172
00:33:10.840 --> 00:33:21.399
Puneet Saraswat: and recently, few years ago, like I. I had a climbing gym near my home, and I started taking my daughter there, and that became like father, daughter, activity for us, you know. And

173
00:33:22.461 --> 00:33:28.198
Puneet Saraswat: and what what I see is like climbing is like, definitely

174
00:33:28.880 --> 00:33:38.589
Puneet Saraswat: it. You can go in the different parts, and obstacles could be different and keeps changing every few weeks, and then it keeps you interested and

175
00:33:39.089 --> 00:33:49.959
Puneet Saraswat: also, you don't need too much prep for it. You just land there, the people just climbing. You need the strength and becomes kind of activity for me to do with my daughter.

176
00:33:52.180 --> 00:33:53.390
Puneet Saraswat: So

177
00:33:54.210 --> 00:34:14.190
Puneet Saraswat: about the climbing ladder and all, I think you have to strategize that like, there's a lot of like you. You need big strength, and you need repetitions. You get better at it. But every time you change the course you have to strategize from one place to other. Resilience is needed. You know. You need you learn resilience. You know. Repetition helps. You

178
00:34:14.460 --> 00:34:24.379
Puneet Saraswat: build the skills and muscle. Yeah, I think those things are anyways you can apply. You know, persistence. Sometime. You you're tired, you give up.

179
00:34:24.580 --> 00:34:29.390
Puneet Saraswat: But you you take a rest and you start again. Someday you can't reach

180
00:34:29.840 --> 00:34:38.669
Puneet Saraswat: one obstacle course, and then you next time you come few days later, and you see, like you have a different approach to it.

181
00:34:38.739 --> 00:34:59.990
Puneet Saraswat: So I think those are the things that you learn, especially in the corporate ladder. I would say. I don't know if it is like one to one. But corporate ladder is more like you end up taking more and more responsibility as you learn the systems, and then people and processes. And you're able to reason through more like how you're

182
00:35:00.100 --> 00:35:10.490
Puneet Saraswat: of work impacts the business and more and more responsibility you can take. You can rally your people around the ideas which are important for for business.

183
00:35:10.600 --> 00:35:15.720
Puneet Saraswat: right? So that that world I I see it's definitely so.

184
00:35:15.830 --> 00:35:26.389
Puneet Saraswat: Always learning where you are, and then kind of up leveling yourself to operate at the next level. And this is like continuous process. It's not like

185
00:35:26.891 --> 00:35:36.550
Puneet Saraswat: you. You reach the top, and there's no top like there's no ceiling there. Not like you always. So that's where one of our culture at harness is like continuously learn and

186
00:35:37.050 --> 00:35:38.650
Puneet Saraswat: and keep getting better.

187
00:35:39.730 --> 00:35:52.019
Dewan Ahmed: Very well said, when you're hiring engineers, do you look for specialists? Do you look for generalist, or do you like T-shaped engineers, com-shaped engineers.

188
00:35:52.190 --> 00:35:56.170
Dewan Ahmed: What's your strategy for hiring engineering leaders?

189
00:35:57.540 --> 00:36:05.263
Puneet Saraswat: For us like here at harness have been like. We have seen, good success in hiring generalists. People who

190
00:36:05.840 --> 00:36:20.680
Puneet Saraswat: are able to pick up and learn new things fast, because the business keeps changing like when we started in 2017 2018, what we were building versus, what harnesses today. In this AI native world. The only

191
00:36:20.920 --> 00:36:44.713
Puneet Saraswat: constant is like change. And you have to evolve yourself. And the core skills, analytical skills able to problem solve. I think those are the and then right minded people who can, you know, like, learn new things fast and reason through the complexity and be analytical. I think that's that's kind of what I have seen. More success.

192
00:36:46.620 --> 00:36:56.220
Puneet Saraswat: but that's like there are some specialized cases where we have to look for people who have done that kind of things in past, so they can build some context. So

193
00:36:56.833 --> 00:37:09.829
Puneet Saraswat: but in general, by and large, I look for you know, generalists who who have good, really good analytical still skills and have all the the right learning and of growth mindset.

194
00:37:11.120 --> 00:37:12.340
Dewan Ahmed: Totally agreed

195
00:37:12.790 --> 00:37:32.510
Dewan Ahmed: now to to wrap up. I was told that you're learning Spanish. Could you share with us some Spanish words that our listeners and viewers me as well. I have Duolingo installed on my phone. I probably learned 10 French words, but that's it. So I'd love to learn some Spanish words from you.

196
00:37:33.320 --> 00:37:37.791
Puneet Saraswat: Yeah, yeah, I think that that's funny. Like, I think Duolingo was another such thing like, I had

197
00:37:38.170 --> 00:37:45.140
Puneet Saraswat: like, started Spanish. And like they had this really good gamification and like ability to engage

198
00:37:45.696 --> 00:38:00.439
Puneet Saraswat: the users. And they have the system of streaks, they will say, and they'll help you track the streak, and I had 400 days streak at building my Spanish skills. So

199
00:38:00.730 --> 00:38:08.428
Puneet Saraswat: yeah, like, I've learned learned few words like Hola amigos Como

200
00:38:09.480 --> 00:38:16.230
Puneet Saraswat: and manzano zombre. These are the the words which are like the 1st few, oh.

201
00:38:16.704 --> 00:38:24.740
Puneet Saraswat: sessions and you keep getting better at it. I think that I love the the platform and the gamification aspect of it, and.

202
00:38:26.260 --> 00:38:52.679
Dewan Ahmed: There's a Ted Talk from Duolingo founder as well. I'll add a link to that talk in our podcast description. And where he explained the same things. These softwares they put in tons of research into keeping users hooked onto their platform, and they use the same methodology to teach languages. So a more

203
00:38:53.070 --> 00:38:56.990
Dewan Ahmed: positive way, I guess, to to keep users engaged on screens.

204
00:38:57.450 --> 00:39:00.800
Puneet Saraswat: Yeah, this is a good example of how tech and

205
00:39:00.980 --> 00:39:13.569
Puneet Saraswat: keeping the engaged for good use cases. There are a lot of examples where we are helping people waste their time. But this is one of them where this technology is helping people upskill themselves.

206
00:39:13.790 --> 00:39:29.990
Dewan Ahmed: Absolutely. What would be your closing thoughts on engineers who are thinking of moving from IC into management? How did you move the how did you make the switch, and what would be your suggestions for them?

207
00:39:30.920 --> 00:39:37.740
Puneet Saraswat: Yeah, I think the that's a good question like, for me. Personally, it was more like I was the strongest I see in the team which I was

208
00:39:38.140 --> 00:39:59.704
Puneet Saraswat: in, and time that the team size increased and there was an unnatural place. You need a leader or manager to to formally run the team day to day. So that was my foray like and that was in azure where I was a founding engineer for the team, and I was passionate about the

209
00:40:00.670 --> 00:40:20.370
Puneet Saraswat: the problem we were solving, and I was always kind of open to to help other engineers and the team and unblock them. And you know, help them be more productive. And it. It was a very natural transition for me. As the team grew in size, and the the problem became not more significant to take up that role.

210
00:40:21.740 --> 00:40:28.640
Puneet Saraswat: for I see to become manager like if if I think you have to look for some traits like, are you?

211
00:40:29.217 --> 00:40:47.580
Puneet Saraswat: In there, for the right reasons, you enjoy working with people, it becomes like as a manager, more about the team versus yourself. So 1st thing you have to learn is team. Success is your success, and you have to bring the opportunities to them, and you have to be able to, you know, connect the

212
00:40:48.099 --> 00:41:00.270
Puneet Saraswat: the dots from, you know, business hours to individual contributions and make sure team is able to, you know, like come together for a larger objective.

213
00:41:00.792 --> 00:41:10.440
Puneet Saraswat: So if you enjoy working with people, if if you care about others, success more than yours, I think you will be successful

214
00:41:10.730 --> 00:41:11.810
Puneet Saraswat: as a manager.

215
00:41:13.550 --> 00:41:26.460
Dewan Ahmed: There you go, whether you're climbing walls or whether you're building systems, you have to connect the dots. And of course you have to find fun in the journey, because at the end this is a journey.

216
00:41:26.660 --> 00:41:36.090
Dewan Ahmed: This was the season finale of season, 3 of ship talk. Podcast we had puneet, Saraswat, the Vp. Of engineering at harness. Thank you so much. Puneet.