Ilan Peleg on How AI Coding Tools Are Disrupting Backend DevOps Workflows

Hey guys, thanks for the throw. We're here with Ellan Peleg, who's the CEO of Light Run, and we're talking about the impact that all these AI coding tools are starting to have on our backend DevOps workflow processes. Ellan, welcome to show.

Thank you so much. Thank you for hosting me, Mike. Great to see you again.

Good to see you as well. I think what we're talking about is not wholly unexpected. I mean, sometimes when you put more stuff on the front end of something, it's eventually gonna break the back end of something.

And what we seem to be seeing is a lot more code is being generated and it's moving through those pipelines. It's not clear to me that all that code is making it into production, but it seems to be stressing our workflows and our and the teams that manage that. So what are you seeing and what are the challenges?

Super interesting. So actually, you know, what we've seen is that, you know, AI C cell has literally become the new reality, right? I think, uh, Dora claims that Dora recent sent a Google report claims that around 90% of the developers out there already adopted some sort of AI system, you know, coding tools, coding agents, and rights.

Um, so, you know, developers simply now spend less time writing code themselves, but rather more like prompting, right? By debugging by, uh, by coding. And then the bottleneck is moving ahead to, you know, reviewing the code, debugging, fixing, verifying.

Uh, in fact, the world also claims, uh, that 60 to 70% of developers time is now spent actually in debugging, fixing air code and fixing their code. Um, and the bottleneck again, is moving to the runtime side of things. Um, and basically, you know, on the moment the code suddenly, you know, meets real world conditions, real world, you know, life and Southern Code is not siloed anymore and is part of an overarching system, including the runtime, including scale, customer data, potentially operations, APIs, third party dependencies and so forth.

So this is where, uh, you know, suddenly things start, you know, break. Um, on the contrary, the past year has been also the most expensive year in software outages. Speaking of, you know, the major outage of CrowdStrike of last year, and just, I'm not sure exactly when this one is going to be broadcasted, but last, last month, we, we've seen, uh, these major outages coming from the big IPO scale such as AWS, uh, the shut down the world, at least for a few hours.

Same with, uh, standard Azure and so forth and so forth. So both trends kind of, you know, contradict and conflict. Um, and this is exactly the gap that we're filling here at lighton.

We help, you know, fix issues at the pace of ai. So fixing issues and specifically runtime issues should be also streamlined. And, um, this gap is actually widening.

And this exactly, you know, where light one is has been super helpful. Troubleshooting, remediates remediate software when software is already operational and are potentially serving customers at scale, uh, we should move faster. Um, 'cause otherwise this bottleneck is widening and we need to help engineering teams fix issues at the same pace.

They actually develop code. The, so developing code is, is not the bottleneck anymore, but whether making sure it's running and operating right? So to your point, it seems like I hear the same things and developers are spending more time reading code than writing it, but the problem with debugging it is they didn't generate it and they don't really understand how it was constructed.

So how do we help them debug something That, that's a great one. Actually, to your point, understandability becomes a key challenge. Uh, I mean, eventually when you know, wheels hit the road, this is the challenge.

Like, it's not me in my siloed, you know, dev machine writing my code in, in the id, but rather, you know, code suddenly, you know, should operate at scale. Uh, and there's so many different requirements of how software should operate at scale. Uh, so key element is, uh, supporting the runtime context.

Actually, this is one of our key announcements, uh, uh, which is us democratizing runtime context to the wider ecosystem. Uh, speaking specifically of, of the coding or testing phase, imagine that each time developer is introducing a new piece of code, potentially even generated by, um, you know, AI or AI code system tool, uh, he will have runtime signals eventually letting him know how the software is going eventually to behave at the wrong time. So it's not only about writing code that works, but it's about writing code that runs at scale resiliently.

Um, so I think like helping engineering teams not only debug the software, but even before debugging is happening, making sure that the software they write doesn't really work on their machine, but whether it fully operates and is resilient, uh, to sustain the scale, sustain, you know, so many potential, you know, vulnerabilities, potential bugs, potential, um, outages is something that should be kept in mind in the hearts of every developer outer, making sure it is not just, you know, shipping code, but making sure that the code will eventually run and operate smoothly at the runtime. So do you think developers are spending more time actually doing that? Because it seems like, you know, they're writing more code and they're enjoying the productivity benefit, but it's not clear to me that they're using that time to go back in and optimize the code or for that matter look for vulnerabilities.

Yeah. In, in fact, Dora also claims that developers using AI code assistance also introduced 41% more bugs. Mm-hmm.

So, you know, as you move fast, you also break things. People tend to say that even before the GNA AI era. Uh, so it's, it's definitely happening.

And I think the key here is not only about embracing AI or mainly productivity gains, but making sure you are embracing AI in a resilient, safe manner. So quality is becoming, you know, the next frontier. So how do you make sure that developers, sorry, but are not becoming lazy and simply not accountable, or, you know, how eventually the software that is about to serve your customers and help you achieve your business goals eventually will help you in achieving your business goals rather than, you know, just developers shipping much more lines of code because potentially they might become a bit lazy and, and reliant on those ai, you know, magical tools.

It, it's truly magical out there, but this magic eventually can somehow, um, you know, introduce some quality issues. In fact, this is what the study, you know, all of the studies out there already prove. Uh, so again, in order to eliminate this one, developers engineering team should be equipped fully the runtime context and how eventually the code their shipping truly is, is about to behave when, you know, uh, when, when the he, the wheels, uh, hit the road, right?

The other thing that seems to be happening too is that as more of this code gets into production environments and it's verbose the cost of running, it seems to get more expensive. 'cause we're processing things that we probably didn't need to in the past. So our cost gonna go up.

Definitely, uh, one reason is simply having introducing much more lines of code, much more compute, much more, you know, code that is that eventually is running. The other thing is AI also, or embedding AI agents as part of your sort of delivery life cycle or embedding much more AI generated code, also introduces like much more non-deterministic behavior that is a bit unpredictable, which can definitely raise your costs of maintaining, scaling your environments. Um, so definitely it's, it's another huge concern.

Again, it ties back to, to my previous point, which is the operate side or the runtime is becoming the site. We should all be careful now because it's quite clear that, you know, AI already reshaped how fast code is being generated. Now we need to make sure that, uh, at the runtime you are maintaining your resilience metrics cost, um, and obviously security.

So all of those operational objectives should be kept and even, and, and, and you should also expect to hire, I would say, uh, standards there because, you know, as AI already proved all of us, it can help us move faster. It should, and we should expect to have better customer experience, better resilience. So shortening, you know, processes at the upside, um, both from cost, but also from, you know, resilience, uptime, troubleshooting time, um, security remediation, all of those, uh, we as consumers for, you know, softwares running in the world, right?

Consumers of those digital assets, we definitely expect higher standards over there as well. Mm-hmm. Um, what's to be done about all this?

I mean, I can't put the genie back in the bottle. These guys are gonna be using AI coding tools. So what do we need to do on the back end to kind of absorb all that?

Um, so I would say much more guard those and governance, um, and, you know, quality gates in again, as, as you move fast, you wanna make sure that you keep moving fast, but you know, not breaking, you know, things. Um, um, while doing so, uh, what we started, you know, to see is that the new reality is actually a reality where code is cheap and bug are bugs are expensive. Uh, quite funny, but it's the, it's, it's actually the reality.

Um, so I think CIOs, uh, engineering leaders and whoever is responsible for the SDLC must take proactive actions, um, in order to verify that eventually the software delivers higher standards of, you know, quality and resilience. Um, and I'm not sure, I'm not sure we were there yet. 'cause as mentioned before, you know, coding has been all like streamlined.

You write code in minutes, um, but eventually when you know soft, sorry, but when s**t hits the fans, um, what happens is that it takes you days, weeks to respond, at least in when calculate fd, right? You literally need to onboard tens or even hundreds of engineers in a world group them together to make sure that, you know, software is, is being remediated. It shouldn't be that way in the gen AI era.

Mm-hmm. It should be streamlined. And, and if code is being written in minutes, it should be fixed in minutes.

Hmm. Will we therefore need more AI on the backend to make up for the deficiencies of the AI on the front end? And what might that look like?

Um, I think taking a deeper look at, you know, the SDLC and different playbooks that exist in enterprises or, you know, software organizations already adopted in the past, each one of those should be automated using the power of ai. It might be a very custom built, you know, internal, um, you know, playbooks, internal policies, internal governance, all of those that are still manual should be streamlined with the power of ai, everything from the front end up until the backend, um, automations, DevOps, your SDLC. Um, so, so my recommendation, you know, CIOs, uh, VP engineerings and, and generally speaking, um, engineering leaders and whoever, again oversees the SBLC, use AI to automate and streamline all of those processes across, across the board, uh, from frontend to backend digital experience, everything from code to production, definitely.

Mm-hmm. So we're coming up on the end of the year, which means we'll be gearing up for the new year. So what's your crystal ball telling you that 2026 is gonna be like, I believe that 2026 is more about the market technology in that, you know, AI literally changed the landscape, especially for engineering and especially for engineers, especially for, you know, the, the SDLC, uh, there's a new term out there, which is called a DLC agent, uh, development life cycle.

So agents are really conquering the SDLC and are really creating a new form of, you know, generating software. But those also introduce, again, much more risk and much more like non-deterministic behavior. So enterprise and software organizations will simply be concurred by more and more, you know, agents, farms running tens of thousands, hundreds of thousands of agents should be now orchestrated, should be now governed, and should also, um, introduce their, you know, reliability standards.

So I think, um, day zero was more around embracing AI for productivity gains. Day one, I, I mean 2026 will be more around governing those and making sure they eventually help us in achieving our business goals rather than degradating the quality of our software. All right, folks, you heard it here.

Hey, if we don't put the governance tools in place, then this AI stuff's gonna quickly become too much of a good thing. Hey, Ilan, thanks for being on the show. Thank you so much, Mike.

All right. And back to you guys in the studio.