Harness CEO Jyoti Bansal on Why AI Coding Doesn't Help You Ship Faster

Frederic Lardinois (0:00)

We are back with another episode of The New Stack Agents, and this time we've got Jyoti Bansal with us. And Jyoti, we've done this a few times, you and I over the years.

Jyoti Bansal (0:11)

Great to see you, Frederick, yes.

Frederic Lardinois (0:12)

Same here. Same here. First time we did this, you just reminded me earlier on was in Vancouver in 2017 or 18

Jyoti Bansal (0:19)

2018, yes.

Frederic Lardinois (0:22)

Yeah, I remember you were delayed to get to that flight, and I gave you a hard time as you also reminded me, because you had just sold AppDynamics for $2.7 billion, and you spent your time waiting for an Air Canada flight to get to Vancouver. Yeah.

Jyoti Bansal (0:38)

Well, I still fly commercial, you know, that's why not, right?

Frederic Lardinois (0:43)

That's the way to go. You can afford first class now. Yes. There you go. Stepping up in the world. Absolutely. Well, the reason I really wanted to get you on the show is because you've done this for a long time. You did App Dynamics, now you're doing Harness. And when it comes to doing building software right now, everybody's using these AI tools to build the software, but it's really the delivery pipeline that's super interesting to me because I think that's where a lot of the bottlenecks are still, right? So I think we should talk about that. But before we do that, Harness is now just doing math in my head here with seven, eight years old.

Did you think about AI at the beginning or at the time, machine learning and deep learning and generative AI was on nobody's radar yet at the time?

Jyoti Bansal (1:41)

Yeah, of course, LLM's generative AI wasn't, but it's interesting and I remind everyone that when Harness was launched out of stealth, this was 2017 So the headline in our press release was Harness Brings Artificial Intelligence to Continuous Delivery. This was end of 2017 And the reason we had that, because that was the core of our CD continuous delivery solution of how we verify if a deployment is going to break something or not using AI. But AI was not LLM's at that time, it was neural net space mostly, a lot of ML. And we have been at the forefront of that. Then when we launched our CI continuous integration, we looked at, and this was 2020, of how we speed up the builds. And we looked at an AI model called Test Intelligence to speed up the builds on figuring out based on what code change, what tests you need to run, so you're not running unnecessary tests and you can speed up builds by like 4X, 5X. So we have been working on AI even pre-LLMs. And sometimes people forget that it's not that LLMs are the first time a use of AI and ML for software engineering, software delivery was possible. If it was possible before, of course, now we are in this 100X, 1000X turbocharged environment of what AI can do.

Frederic Lardinois (2:57)

Yeah. And still sometimes it feels some of the things we're trying to solve would be easily solved with just machine learning as well. And a lot of companies try to shoehorn generative AI into that world.

Jyoti Bansal (3:10)

Yeah. I think a big part is like, I always tell our teams like it's about solving the problem. Doesn't really matter how we solve it. Like some problems are best solved with agents, some problems are best solved with maybe agents are not the right things. Some things are called combination of agents and other algorithms, some things combination of ML, some things are about deterministic automation. Like if you are using AI and it takes, because AI takes time to resolve and determine something, and it could be that it's taking 30 seconds, and you just deterministically write it, it may take one second.

And does the user want to wait for one second or does the user want to wait for 30 seconds? So we also have to make the determination in some cases where is AI is the right thing to figure it out for the first time, and then you save it, so then it's deterministic and faster without AI. So AI is not the end-all thing, it's a means to what problem you want to solve.

Frederic Lardinois (4:07)

Yeah, it might be the end to all things at some point. That's a different conversation.

Now, as you started getting into adding generative AI into your product, how did you think about that? What kind of process did you go through, and when did you even make that determination that now was the right time to do that?