What if the secret to unlocking the Agentic Development Life Cycle is not a better coding model, but a smarter engineering context engine? This week on Dev Interrupted, LinearB founders Ori Keren and Dan Lines join the show to discuss the messy middle of AI adoption and the painful transition from the traditional SDLC to the Agentic Development Life Cycle. They unpack why the era of cheap AI experimentation is over, how rising token costs are forcing engineering leaders to prioritize strict business ROI, and how autonomous tools are fundamentally changing the daily workflow of developers.
Show Notes
LinearB: Learn how to transform your SDLC and build an engineering context engine at linearb.io
gitStream: Explore LinearB's workflow automation tool for routing pull requests at linearb.io/platform/gitstream
Transcript
(Disclaimer: may contain unintentionally confusing, inaccurate and/or amusing transcription errors)
[00:00:00] Ben Lloyd Pearson: Welcome to Dev Interrupted by LinearB. Today, we're doing something really awesome. Uh, we've got both LinearB founders in the room with us. We have Ori Keren, CEO and co-founder, and Dan Lines, COO and co-founder. Uh, yeah, Ori, Dan, welcome to the show.
[00:00:17] Ori Keren: Hey, great to be here. Thanks for having us
[00:00:20] Dan Lines: Thanks, BLP.
[00:00:22] Andrew Zigler: Now, the reason that the four of us are all in one discussion is 'cause LinearB has been moving really fast all year, and all four of us, we have a lot to say about it, especially you two, and about where the industry is going and what our company's been working on, but also the opportunities that you see for everybody right now in the industry.
[00:00:40] Andrew Zigler: Because something that we all talk about a lot on this show is making, like, predictions and understandings about where the industry is going, because that's the insights that we get to, to work with here at LinearB. And so I'm really excited to, to dig in to some of the stuff that y'all have been learning.
[00:00:54] Andrew Zigler: And I actually just wanna kick it off, Ori, by giving the, the first question to you, because, uh, when we talked with you at [00:01:00] the top of the year, you predicted that organizations, we would struggle to digest all of this AI-generated code. We'd be, like, under, like, a, a deluge and a flood of all of this new, uh, AI-generated content that teams are gonna struggle with. Uh, and I wanna check in with you now that we're about mid, uh, way through the year. How do you feel about that prediction? Do you think it's coming true? Uh, what do you see?
[00:01:22] Ori Keren: Yeah, I think by the way, like e-even a year ago, I said something like, uh, uh, productivity will actually dip and then it will start rising. And I th- I think this time, yeah, like people strike. Yeah, we're definitely seeing it. I think da-data-wise, our data is showing and, and it's also backed by like, uh, you can read a lot of other re-research and other papers that there's like, uh, around 2X, uh, more pull requests throughput and more like, uh...
[00:01:52] Ori Keren: So more code is like actually like hitting like the, the pipes. Uh, but yeah, definitely teams are [00:02:00] swamped because like, uh, there's not enough people to review the code, there's not enough people to test the code, there's not enough people to handle the releases. organizations are actually saying, "Yeah, like we have-- we're generating 2X more code, but the gains that we're getting are actually somewhere between 10 to 15% of productivity," even if you ask them qualitative or even if you measure something.
[00:02:25] Ori Keren: and I'm not even talking about the fact that quality is degraded a little bit. People are reporting like more failures. So yeah, I think, I think it's definitely what's happening. Uh, yeah, code generation is faster, but the rest of like the pipes and the rest of the operations that need to support, like shipping software faster are still not there.
[00:02:44] Andrew Zigler: Yeah, most companies are still in this productivity pain point that you called out last year, you called out again this year. We're still suffering through understanding how to actually make sense and, and gain all of the productivity you can from these tools.
[00:02:57] Ori Keren: Exactly
[00:02:58] Ben Lloyd Pearson: Yeah, and, and, and [00:03:00] we've, we've been covering here on Dev Interrupted a lot how we're, you know, in this messy middle of AI and, and adoption and success with it has been very lumpy. You know, some, some teams, some organizations see really great successes with it, and they learn how to replicate that across their organization, whereas others maybe only see tiny pockets of these improvements, and meanwhile, the rest of the company maybe is actually starting to look worse over time because of AI. So Dan, I'm, I'm curious to hear from you on this. You know, what are, what are you hearing from our customers and, and the engineering leaders that we talk to every day about how AI is reshaping, um, you know, their challenges and, and what they're coming to LinearB to solve? You know, has, has the concerns changed, or have they just gotten bigger as AI has ramped up?
[00:03:44] Dan Lines: Yeah, man, great question. I mean, first we gotta start with what kind of customers are we talking to? I'm talking to customers, when LinearB's talking to customers, these are the best companies in the world. These are like Fortune 500, Fortune [00:04:00] 1000 companies. These companies are dominating markets right now. And so what they're going through is more so this transformation from their SDLC to their ADLC. And I think the difference, at least from what I see from the beginning of the year, beginning of the year was like still a lot of hype, still a lot of experimentation, things like that. " Okay, we're trying it in some pockets. Okay, all of our developers are using something, Copilot, Claude, this is fun." transformed now. What I'm hearing is, "Let's get back to business. Let's get back to results. I need ROI for the business." So there's three things that they're saying to me. Thing number one, trust and proof, Ori mentioned it, with quality. They're on their journey to full autonomy, but they need to see, "Can I really release this code to prod? Are the floodgates gonna open here? Am I ready for that? I'm not fully ready. I need quality." So I hear a lot, "Quality, [00:05:00] quality, quality." That's number one. Number two, I hear, and think we're gonna talk about it more, cost versus ROI.
[00:05:06] Dan Lines: This stuff's not cheap anymore. It's not gonna stay cheap, right? We're all hearing the reports, it's gonna be expensive. So now they're saying, "Okay, I gotta make sure that my cost is intact when we're working with these agents." That's number two. And then the third thing I, I'm hearing, and we can like, uh, we can break it down, but they're almost dissecting their SDLC.
[00:05:29] Dan Lines: They know they need to get to this AI DLC, this ADLC. They're almost dissecting it. saying, "Okay, I got these, you know, assisted coding agents. Maybe I have some full autonomous, but what's my specs like? Is that AI driven? Or where's my bottleneck? Before it was like the code review."
[00:05:46] Ben Lloyd Pearson: Mm-hmm.
[00:05:47] Dan Lines: they're kind of like, "Oh, how do we deploy it?"
[00:05:49] Dan Lines: So they're kind of breaking down each part of their SDLC and saying, "Okay, this is what I need to transform. I need to unlock all these bottlenecks." it's shifted a, a lot, man. It's, it's pretty [00:06:00] crazy what's happening, but it's really exciting
[00:06:02] Ben Lloyd Pearson: Yeah. And, you know, I, you, you hit on a couple points that I wanna, I wanna j-just double-click on for a minute. Uh, and that is, you know, you mentioned quality, and I've been looking into the, to the data on this a little bit, uh, you know, just from our own customer base. And like, it, it looks like things like rework are just like plummeting at this point, as one example of where, you know, AI is-- it tends to default to generating new code rather than trying to rework an existing system. Uh, and refactoring is, is kind of a similar... And, and again, it, it just depends greatly on how you look at, how you slice the data, you know. Because across the industry, we can see the broader trend, but at the micro level, when you look at individual companies, there's, there's so many outliers where they, they, they've just-- Like, there's clearly major problems that are being, uh, introduced into their, their SDLC. Um, and then yeah, on cost, we've been, we've been covering here a lot how usage costs are exploding right now. just changed to a fully [00:07:00] usage-based pricing system. Um, I think that's gonna be the norm for most developer tooling. Uh, and z- you know, we can't just keep spending, uh, limitless amounts of tokens, you know, because it's gonna make costs just go out of control.
[00:07:12] Ben Lloyd Pearson: So yeah, I think those are really two big things that like, uh, y- it's, it's the elephant in the room. We all have to be talking about those
[00:07:19] Dan Lines: The bill is coming due
[00:07:21] Andrew Zigler: bill is coming. I talk about this so
[00:07:23] Dan Lines: All right
[00:07:23] Andrew Zigler: the show and I feel like a broken record that like this cost of using these tokens is only gonna go up. Your ability to learn and figure out where those bottlenecks are will never be more subsidized than it is right now
[00:07:36] Ori Keren: Yeah
[00:07:36] Dan Lines: That's so funny. This is probably gonna ma- make me sound... Do you remember... sound old. Do you remember when it switched from, like, cable TV to these streaming services?
[00:07:43] Andrew Zigler: Oh
[00:07:43] Dan Lines: Like net-
[00:07:44] Andrew Zigler: like
[00:07:44] Dan Lines: it was, like, super cheap. It was like, "Oh, okay, your cable bill is, like, 250 bucks." Now it's, like, 19.99 a year.
[00:07:51] Andrew Zigler: Yeah
[00:07:52] Dan Lines: bill has come due. How many streaming services do you have? It's like 500 bucks a year now. That's how I feel about what's happening here.
[00:07:59] Andrew Zigler: It's [00:08:00] exactly what's happening. It's like they're, they're, they're, they're subsidizing it. They're bringing it in low, so we're incentivized to reinvent and, and blow up all of our process and get it in there really deep just for them to hike up the price. But also that's the reality of, uh, understanding the intelligence capabilities that you're drawing down, the costs of them, when to use certain models over others, when to choose to use your own local models.
[00:08:20] Andrew Zigler: These are all really high level AI fluency, AI hygiene conversations that companies are having right now. But since we did just have this whole year, year and a half of like really heavily subsidized AI code gen, you got these teams that really went at break speed and hit these bottlenecks and found new bottlenecks in their, their, their code generation process.
[00:08:43] Andrew Zigler: And, uh, like for example, in our 2026 benchmarks, like it shows that a- agentic AI pull requests, you know, they wait five times longer to get picked up, and they get merged at less than half the rate at the human authored ones, which points at a major gap in the code review process, and [00:09:00] that creating the code is only one small step of actually getting it to deliver value to customers.
[00:09:06] Andrew Zigler: And because, uh, folks are contending with that uncomfortable reality of like, "Oh, there's bigger problems in my code base than just getting the code into existence," you know, I, I'm, I'm curious to know from you, Ori, like how have you seen teams handle these shifts in their bottlenecks? Have there been other bottlenecks that have been really prominent for you to think about as like a leader in the space?
[00:09:29] Ori Keren: Yeah. So first of all, I think, um, this is super interesting. I think there are two problems. We talked about the flood of code, and like you said, like, uh, it was very cheap. It's not gonna be cheap anymore. Uh, but, uh, it was very cheap in the last year. So think about like m- a lot of code is hitting the, the system.
[00:09:49] Ori Keren: What I'm seeing and, um, when-- what I'm hearing when I'm talking to customers, and Dan spoke about like, uh, depends on, okay, what type of customer, uh, you are. But w- [00:10:00] what I'm seeing is that, Not all the companies are ready to say, " Okay, AI generated the code. Let's, let's do an AI-based review and figure out what's happening with that, like change downstream and let it maybe go out."
[00:10:13] Ori Keren: The universe like is not ready, uh, for that yet. Maybe some companies are doing that. So what's happening right now is that senior developers across all companies are swamped. Um, they are like tired. They're underwater. They need to read all this code and approve all these changes. So I'm actually very concerned from like two things.
[00:10:37] Ori Keren: Uh, one, what's gonna be the s-something that solves this? And the second thing is like how do you, uh, not like get into a situation where a lot of knowledge is lost? Because th-those senior developers are the people that know, they have like good understanding, good gras-grasp of like, okay, this service might be in risk or, you know, this [00:11:00] tribal knowledge of where, uh, things could be broken.
[00:11:03] Ori Keren: Uh, so you deploy them now like to just review all this code so we can, make sure we ship things faster, and then we say, say to the organization, "Yeah, yeah, we are moving faster." So you lose them as knowledge centers. You don't grow the next, what we call like, uh, senior developers. Uh, that's like a high concern that I'm, uh, hearing from a lot of, uh, a lot of companies.
[00:11:24] Ori Keren: Uh, and yeah, you talked about like the fully agentic, uh, pull request. it's still the same old problem. Like if an agent wrote a code, who's gonna tap on someone's shoulder, "Hey, can you please like look at my PR because I need like to get my feature delivered"? Nobody does that. So it's, uh, the ownership problem, uh, is huge.
[00:11:45] Ori Keren: How do you solve for that problem is also big.
[00:11:47] Dan Lines: I guess junior developer is the best job in the world now.
[00:11:50] Ori Keren: Let
[00:11:50] Dan Lines: me just create all this code and give it to the senior developers to burn them out. They have all the context. They know all the intricacies of the system.
[00:11:59] Ori Keren: Yeah, I wonder[00:12:00]
[00:12:00] Dan Lines: we, and we don't wanna burn out our senior developers. That sucks.
[00:12:03] Ori Keren: And I wonder, uh, the thought that I have then is like, when will junior developers will become senior developers? Because, uh, maybe this event won't happen anymore. It's like, it's, uh, super risky, like
[00:12:17] Dan Lines: What makes a senior developer? I think you said that they have the context. They know all the areas of the system. They know the architecture. They know where the skeletons lie. That's why, I guess that's one aspect. I don't know if... Yeah. That's a great question to ponder. I, I don't know what the future is for junior developers, but they're probably, if I was a junior developer, be going like, "You have to almost all agentic.
[00:12:38] Dan Lines: That's where your value lies." I don't know if they'll evolve. Let's see.
[00:12:42] Ori Keren: But yeah, that's also relates to like another thing that you guys spoke about before. Like, I think organization know how to deploy AI now very strongly to greenfield.
[00:12:53] Andrew Zigler: Mm-hmm.
[00:12:54] Ori Keren: like, okay, if I have to build new stuff. I'm actually obsessed about [00:13:00] success stories of companies who refactored like big chunks, like big monolith, and what did they do like to get there.
[00:13:08] Ori Keren: And when companies are successful with that, that's like, uh, I think open- opens up like a, a, a true, like new level of, uh, productivity.
[00:13:17] Andrew Zigler: I
[00:13:18] Ori Keren: I'm obsessed with the, with these stories. Whenever I like, uh, ran into one of these, like, uh, I wa- I wanna understand what did they do. Usually, it's like a lot of tests, like to hold to make sure everything is working.
[00:13:30] Ori Keren: But then I'm hearing stories about project that they thought are gonna cost eight man years that they do now in two, three months, which, which th- this part is amazing.
[00:13:40] Andrew Zigler: That part is amazing. The way that, like, huge preexisting companies can transform their brownfield code bases using AI and perform things that were always sitting on a back burner, like really important security migrations and system updates and even just, like, refactoring stuff that's been sitting in a closet for a decade. I think whenever, like, a [00:14:00] company's able to have success doing, uh, using AI at scale relatively, like, autonomously, like they come in and it's, like, doing its thing, is really impressive because that speaks to that they were able to distill their domain expertise and what matters to them as engineers into a loop that is able to do something end to end for them effectively.
[00:14:20] Andrew Zigler: And that's way, way, way more impressive, I think, than people that are going out there and building, you know, brand new greenfield things. As exciting and as cool and as big as the opportunities will be for those in the future, it's much more exciting for me from an engineering standpoint to think about how those big preexisting teams with those old, you know, built up crust kind of applications managed to become agentic, managed to transform with the technology. because Ori, to your point, the catalyst between becoming a junior engineer and a senior engineer is going to be really blurry. That's what a senior engineer has. To your point, Dan, it's like they know where the skeletons are, and you only know where skeletons are by thinking about the problems a lot, [00:15:00] being in it, and being a part of the process.
[00:15:02] Andrew Zigler: And the more abstracted that junior devs are from that process and never get close to it, the less opportunities they're gonna have to grow into those senior developers. So this is also too, like, uh, a personnel and like, uh, kind of like how you go about organizing your teams thing as well. Like, AI is not just code. It impacts your team and its shape as well
[00:15:24] Ori Keren: 100%.
[00:15:27] Ben Lloyd Pearson: Yeah. So we, you know, we've been talking about cost a lot so far. and, you know, one thing that we've seen from, from some of the data that we've collected is, is that, um, there's a, it's still a big challenge for, engineering leaders to translate those costs into ROI and impact on the organization. Uh, so Dan, I want to give this question to you.
[00:15:49] Ben Lloyd Pearson: You know, as we're seeing AI like evolve and, you know, these organizations are maturing with it, you know, uh, h- how, how are people justifying the investment [00:16:00] today? You know, because year ago it was like everyone was just thinking about adoption. It's like, we just want to get developers to use it. Uh, it seems like adoption is basically universal at this point, but now we need the next thing that, that shows that that adoption is actually driving something.
[00:16:14] Ben Lloyd Pearson: So what are you hearing?
[00:16:15] Dan Lines: Yeah, it's not just fun and games anymore. It's not just, uh, "Hey, how many developers ha- have adopted?" You know, that's like the, the first stage of the journey. But like I was saying, like it... And this is actually honestly what it was before AI. It is going back to business value. This cost is increasing so much that I have to say, as I'm having my developers use agents, I have to justify that the value being shipped to production, usually for our customers, could be the cost, like the debt reduction too.
[00:16:45] Dan Lines: I think that's like a good, good thing to use, you know, uh, tokens on. But it has to say, yes, it is increasing the amount of value that we're delivering. Our cycle time is, uh, decreasing. The amount of incidents found in [00:17:00] production is going down. Our quality remains the same. This is what the best engineering organizations are already proving. I don't know if I wanna jump too far ahead, BLP, but I saw a demo. saw a demo within LinearB, where now we're able to show, okay, here's all of your agents. Here's the cost of your agents. Here's how they relate to the teams using them. Here's how they relate... I mean, this is what the customers are asking for. All right, so maybe we say like we're in this alpha be- beta stage. I don't wanna ru- ruin the surprise. This is what they need. That's why we're building it. Every single conversation is, "Okay, I gotta justify this now." Fun and games are over. So that's what I, what I'm super excited about for us, and that's what, what I'm hearing the customers need and want
[00:17:46] Ben Lloyd Pearson: Yeah. And, you know, we've been talking a lot about how, like, the patterns of behavior are changing. Like, the way you do work is, is fundamentally changing, uh, including for developers now. And, you know, and the-- there's a new ty- or there's new [00:18:00] stages in the, the life cycle of, like, a developer's day, right?
[00:18:04] Dan Lines: Yes.
[00:18:04] Ben Lloyd Pearson: They, they create,
[00:18:05] Dan Lines: I don't know what's up
[00:18:06] Ben Lloyd Pearson: time creating a prompt, and then there's some idle time while the agent is doing work. And then
[00:18:10] Dan Lines: Yeah
[00:18:11] Ben Lloyd Pearson: there's a-- Then the agent needs the human to come back and provide feedback, and there
[00:18:15] Dan Lines: Yeah.
[00:18:15] Ben Lloyd Pearson: as it waits for, for the human. You know, th-
[00:18:18] Dan Lines: crazy
[00:18:19] Ben Lloyd Pearson: these whole new dynamics now that we, we've started, we've had to start thinking about because we're, you know, we can see that, that, that AI is just fundamentally changing how we get work done day to day.
[00:18:30] Dan Lines: It's funny, I, I heard, I think O- Ori, you might have told me, I've been hearing this. There's almost like, okay, there's this classic outer loop, let's call it the outer loop, that's always existed, and you can kind of frame that by, you know, DORA metrics, cycle time, MTTR. S- that's like the outer loop. But there's this, and it's still very important, but there's like an inner loop now, that's what you de- described, BLP.
[00:18:54] Dan Lines: There's like this inner loop of a, a agentic circle that's hap- And it's like, "Is my agent [00:19:00] running? How long did it run? Oh, did it get stuck? Does it have the con-" There's something that's changed there with the, this inner, inner loop, uh, I think is, like, really interesting maybe to, like, touch on here.
[00:19:12] Ben Lloyd Pearson: Yeah. And, and I'm, and I'm curious, you know, 'cause we, we meet a lot of companies that are just sort of at all stages of AI maturity. So I'm just wondering from your perspective, like, what, what does that look like now? Like, where, where is the typical organization? Where are the, the leading organizations?
[00:19:26] Ben Lloyd Pearson: Like, what are you seeing?
[00:19:27] Dan Lines: Man, it, it, it depends on the size of the org. Like, uh, like I'm say- Okay, maybe you have two different, uh, style organizations. Were you born pre-AI or post-AI? Did you go and dominate a market before AI was here? These are like your Fortune 500s and your 1000s,
[00:19:46] Ori Keren: them now the,
[00:19:46] Dan Lines: are you
[00:19:47] Ori Keren: the in- the incubants, like
[00:19:49] Andrew Zigler: The incumbents, yes
[00:19:51] Dan Lines: Are, are, are you that?
[00:19:53] Dan Lines: Or were you born, like, in the last, like, three years? I think that totally is a different, like, uh, [00:20:00] ballgame. born now, okay, yeah, you started with agentic development. You're all about how long my agents are running without me having to go and, like, tell them what to do next or your spec- But if you're like, you know, a lot of the, the larger companies that we work with, in a different path.
[00:20:19] Dan Lines: You have like, uh, 4,000, 1,000, 500 developers. You're trying to shift an entire organization. And I think even for those, it's more, like, move beyond just like, okay, how many developers are, are using. I think they've gotten into like, okay, what are the quality gates? What's my AI code review? And now I see them start saying, " Okay, let's say that I'm okay with the code gates and stuff."
[00:20:45] Dan Lines: They're still kind of working there. How am I generating the specs? How am I testing? So it's kind of like a totally depends on when you were born, when this company was born, to be honest with you
[00:20:59] Ben Lloyd Pearson: Yeah.[00:21:00]
[00:21:00] Ori Keren: But even, but
[00:21:00] Andrew Zigler: Yeah, I completely agree
[00:21:02] Ori Keren: even like, um
[00:21:03] Ori Keren: It's very interesting to see within these companies, like these big companies, all of a sudden you have one team that's like, whoa, like ahead, and they're like behaving like, uh... Because we-- I keep thinking about it like three phases. There's like this visibility and, and all you care about is visibility and adoption.
[00:21:22] Ori Keren: Then there's like experimentation, then there's like, okay, transform. Now some companies, like you said, then were born transformed. Now what's, what is super interesting, like you can see these teams that are like taking off within the big companies and yeah, they're like, of course, like my i- my inner loop is, is like a new inner loop, like Dan said.
[00:21:42] Ori Keren: It's like it includes like, uh, idle time for agents and everything is spec driven development. And, uh, it's, it's super interesting because the comp-- the big companies are looking for these teams, and they wanna identify them and kinda, and kinda say, "Okay, how do we replicate that [00:22:00] behavior?"
[00:22:00] Andrew Zigler: Yeah, it's really like an apples and oranges thing. Like, those two different types of teams and what they're gonna get out of AI are, are very different. The, the strategies they have to employ to not only, like, transform but to survive are totally different. And, like, for the ones that have, like, all the preexisting stuff built up, for them to be able to identify those individuals and those teams that are managing to find those new inner loops and are managing to, to unlock new speed and success, like, the way that they, those organizations can mature on their AI pathway is to, uh, uplift those em- those employees and to enable them across the entire organization and really, figure out what it, how, what is working so specifically well for them and how can we replicate across everybody.
[00:22:48] Andrew Zigler: 'Cause, like, I'll say as somebody who has found a lot of success using the tools, like, you don't want to be that person, that 1000X engineer that's, like, doing everything and every- everything just falls into your, like, black [00:23:00] hole of engineering, right? You wanna be able to distribute that ability to work to everybody so the whole organization can up level.
[00:23:07] Andrew Zigler: So, like, it's a really great partnership that you see between, like, these orgs and with these leaders and these teams that do manage to unlock that. You know, like, Ori, what have you, um, seen from, like, teams and what do you think about, like, uh, teams that are, have these kinds of, like, you know, maybe hidden gems within their org and, and how do they unlock them and, and transform everybody else?
[00:23:28] Ori Keren: You mean like, um, um, how do they find ways like to infect like the entire org or something? That's what, that's the kind
[00:23:37] Andrew Zigler: find it and then uplift it and get it to everybody else?
[00:23:40] Ori Keren: So first of all, like, um, um, like we said before, uh, you need to identify them. It's not easy. Think about like what Dan said, like you have like, uh, I don't know, 2,000 developers, 500, I don't know, 4,000. you gotta have like ways to, uh, identify those teams. Uh, what I've seen people doing that [00:24:00] is very successful is like, even internally, uh, I've, I've seen us doing, our engineering organization is doing, is, uh, uh, giving autonomy to teams like to try new things.
[00:24:12] Ori Keren: Uh, for example, there's one team who's like went, okay, everything is spec driven development end to end. And then, okay, how, uh... I was actually a part of like, um, um, you know, um, town hall meeting for a big customer of ours, where the one team that said, "Okay, we cracked the code," was standing in front of everybody and was describing how did they work, et cetera.
[00:24:34] Ori Keren: Um, there's another great thing that, if you learn how to work better, so there's the human element, right? Think about like our industry. Like, uh, if you know that you need small PRs, before that, what you had to do is go and say something like, "Okay, let's run an education program and explain to everybody how important it is," et cetera.
[00:24:56] Ori Keren: Improvement now is, yeah, educating the [00:25:00] teams, but sometimes it's just like prompting better. So all of a sudden, like you don't have to wait to teach everybody how to work. Uh, you sometimes just like prompt better your agents. So then like, uh, it's, it's becoming infectious. Like even the team who didn't like, uh, that still need to learn, uh, are getting like better working agents.
[00:25:20] Ori Keren: Uh, so that's a very interesting topic, how improvement is becoming exponential now, when it's not like just humans. You can just tell the agent, "Hey, this is the right way to work."
[00:25:30] Andrew Zigler: Yeah, I love that insight, the idea of like, "Oh, this is ultimately just a prompt or a context engineering change," and it becomes easier to experiment and find those results.
[00:25:40] Ori Keren: Yeah. Before that it was a six-month, uh, education program. Think about it.
[00:25:44]
[00:25:44] Ben Lloyd Pearson: and, and I wanna make a shameless plug real quick. You, you know, I, what I've, what I've been describing this as is like almost like the agentic halo effect. Like you wanna find the developers that both that have, uh, have a high, like high AI usage, but are also on a team with high AI usage. And [00:26:00] you wanna find the teams where you have individuals that have a high AI usage, but low usage across the rest of the team.
[00:26:06] Ben Lloyd Pearson: And, and I've actually been playing around just with LinearB to like dig into, to our own data, and it's like really surprising how you can just like quickly find like that person who is like clearly at the center of some sort of transformation, and then it's just a matter of like doing, bringing their success to the rest of the team, you know? Yeah, 100%.
[00:26:25] Andrew Zigler: So have a provocative question I wanna put on the table for you, Ori. You know, we've had some discussions in, in this chat about greenfield versus brownfield code bases, and is one of those what you call incumbents. We were existing before AI entered the scene. But if, uh, LinearB was an AI native company and you were launching LinearB this year, you know, what would you do differently? How differently would you build it?
[00:26:49] Ori Keren: Yeah. Uh, thank you for the provocative question. Um, I was talking to Dan the other day, and, you know, trying to, uh, [00:27:00] kind of frame it. W-we understand, like, the en-engineering teams, uh, they need context, right? and by the way, they needed context before, but AI made it, like, mission-critical. If you don't have context, it's like, uh, what do you do?
[00:27:15] Ori Keren: Like, how do you even move? Uh, and it hit me that actually LinearB, what we build is the engineering context layer. Uh, that's what we built. By the way, without knowing at the beginning that that's what we built, uh, in full transparency. Like, uh, like I said, like before that, like improving something was, uh, an education program, and all of a sudden now you have like, uh, oh, you have context.
[00:27:39] Ori Keren: You can... So if I would build LinearB today, and it's not just a theoretical question because my role is also to, to navigate the company to, uh, do this change and to, uh, make sure our product and our vision aligns with that. I would be able to say, "Hey, this is the operational context store," and it has, [00:28:00] like, two legs, almost like one for, um, AI native development.
[00:28:05] Ori Keren: So these agents, like we spoke before, they need like, um, context around repos, et cetera, but they need also super interesting context about teams and u- and then, okay, if I have to find someone with knowledge, how do I do that? And like incidents and services that are up or-or, or down, and that's right. So they need like...
[00:28:27] Ori Keren: So one leg would be like the, what I call the AI native development context, and then the other leg would be the engineering operation context, which is... By the way, excites me sometimes even more than the coding part because remember how we spoke before about how to jump over the ten, fifteen percent, like, uh, productivity level?
[00:28:48] Ori Keren: That will happen not only when, like, you fix, like testing and reviewing and all of that. That will happen once, and we have customers who are doing it today, which is amazing. Once like [00:29:00] road mapping, you won't go into a road mapping session before you have like an, uh, a-an analyst agent to, like, look at all your data, all your gong calls, all your...
[00:29:08] Ori Keren: And say, "Hey, this is probably what should be the roadmap." When you think about even, like, your iteration scope, if you're a team lead, you wouldn't go into, like, this planning session for half a day before you have an analyst agent crunching your data and saying, "Hey, this is what probably you should be working, and that's what's realistic."
[00:29:25] Ori Keren: And even staffing decisions and bigger things. So to sum it up, I would build like a, a context store, and that's what we're actually building, right? That's like the transformation that we're going right now with, for AI native development, but also for the engineering operations side, if it makes sense.
[00:29:42] Andrew Zigler: Yeah, completely. It's like you see the opportunity is that it's teams need to understand not only what's going on in their org, but convert it into this context layer and something that's navigable by humans, but then also act upon-- something that can be acted upon by agents. It becomes this harmonious middle. And I really think that's the differentiator [00:30:00] between LinearB and, and other platforms as well, is because it has the follow-through. It not only is gonna give you that, you know, we're calling it a context engine. It's like you turn it on, you drive it, like it takes you somewhere because it's giving you not only the information, but it's also letting you do follow-through. And like acting on that context I think becomes like a big diff- differentiator. And so like now even in the space that we're moving into in this reality, in the incumbent LinearB world, like how does that capability evolve? How does LinearB go from, you know, it's not just a context layer, it's a context engine.
[00:30:34] Andrew Zigler: You can drive it, you can take action with it
[00:30:36] Ori Keren: Yeah. So, uh, Dan and I were always obsessed even before, like, we understood that we build, like, uh, the engineering context engine on how to help, like, uh, our customers take action. So, LinearB has, like, gitStream, which is like the layer, uh, the orchestrator, if you, if you will, like, when you do-- when you wanna decide, like, [00:31:00] uh, when you w-wanna speed up coding decisions, like, uh, what do I merge?
[00:31:04] Ori Keren: Who do I route this pull request to, et cetera? Well, we're thinking about gitStream now as, okay, orchestrator is great, but what if, like, we give this context also to the existing agents that they can take be-better decisions? That's one direction that we're going. And the other one, um, is on the other side that I spoke to you about, like, the engineering operations side.
[00:31:26] Ori Keren: So think about, like, uh, we built, like a, a library of skills that you can use exactly for those use cases that I just described. Like, when you go into a roadmap session, activate this first, uh, and all the amazing context that we have will help you, like, take smarter decisions. So, we're in the phases of developing it.
[00:31:45] Ori Keren: It's, uh, available for some, uh, design partners. Uh, and, um, we're already, like, um, thinking about the ideas of making it, like, a marketplace and, um, collaborating with our customers on it because, uh, [00:32:00] again, this is the area where I think, like, uh, we'll give the actually the 2X and the 3X like the people want.
[00:32:06] Ben Lloyd Pearson: We have customers that are automating like hundreds to like thousands of developer months or d- developer hours a month through LinearB, you know, and we've always like really thought hard about like making sure that we-- we're enabling organizations to get better, not just like understand their data. so, you know, as AI continues to get rolled out, um, and it matures and the, and the technology changes, Dan, what do-- I'm, I'm curious about what you see, uh, in customers that are getting ahead and what they're doing to prepare, um, for all of these changes versus customers that are maybe, you know, falling behind the curve.
[00:32:40] Dan Lines: Well, listen, the first thing that customers are doing to get ahead, they're getting their end-to-end observability. That's why they come to us. If you wanna go fully autonomous, fully agentic, the first step, this is like the responsible thing to do, you have to know [00:33:00] How does my SDLC look today? What is my efficiency today?
[00:33:04] Dan Lines: What is my quality today? What is my ROI today, and how much is it costing? You have to know that. Now, after you know that, that's when you can then go and start dissecting. This is what I was talking about earlier, dissecting your SDLC and transforming it to your ADLC. So now you can start saying, "Okay, I'm gonna make a change now.
[00:33:27] Dan Lines: I'm gonna start having agents do all the coding work, but I'm gonna put a quality gate in place. I'm gonna make sure that the code going out, the rework isn't gonna spike to something insane. The incidents aren't gonna spike to something insane." Maybe that's the first step. the second step, okay, I understand now we talked about context being super important. Are my specs looking good? Are my specs AI-driven? Ori talked about bringing in Gong call data. So the best, uh... I would say the most mature [00:34:00] organizations, I'll use the word maturity, the most mature organizations that are on the hook to deliver actual business value, not just go like all greenfield and go crazy with agents, but you have to deliver business value at cost, they are mapping their SDLC with data, and then they're using the context, this is what LinearB provides, to make the agents better, make better decisions, get my analyst agent.
[00:34:25] Dan Lines: That what-- That's what they're doing, BLP
[00:34:27] Andrew Zigler: Kind of jumping off of that as we're kind of coming into the end of our, you know, discussion and our time here together, um, you know, we've explored some of how, well, LinearB customers have been using the, the platform to, to transform their SDLC into this ADLC that, that you've identified to find those outer and inner loops.
[00:34:47] Andrew Zigler: And really, I think the common through line of all of these stories and everything that people, uh, engineers have been working on in the last, like, two years for sure, and all of our predictions we've made here, is that there's just a huge amount of [00:35:00] transformation and disruption and new stuff out there all the time.
[00:35:04] Andrew Zigler: And we're reevaluating things that we carried in from our past engineering lives, from the eras before into today. And part of that too is like, uh, maybe like the tooling and the technology and the way that we write and ship code is just not even sufficient for the shape of, of the engineering projects that we do now.
[00:35:21] Andrew Zigler: There's lots of talks about people being like, "Oh, we don't even need programming languages. We need some new language that is closer to the agent and is more natural language." And you have other folks over here that are saying that, you know, um, where the code lives and how we even share code with each other is insufficient.
[00:35:37] Andrew Zigler: And that one really sticks with me. It's like the idea of the Git forge itself maybe being insufficient now for, like, the era of agentic development is really fascinating. I wanna just end on one last thought-provoking question for you, Ori. It's open-ended about, like, where things are gonna go. Do you think the Git forge is sufficient?
[00:35:54] Andrew Zigler: Do you think that even Git itself is gonna change? There's nothing we should take for granted.
[00:35:59] Ori Keren: First of [00:36:00] all, you're right. There's nothing we should take for granted. Like, I think, uh, we don't know where this, uh, will go. I think, uh, I've heard people exactly talking on like the left side, "Hey, you know, you know what? The code is not even the source, uh, the single source of truth of the application anymore.
[00:36:17] Ori Keren: The spec is." So yeah, like you said, maybe a new language, maybe a domain language. Like, okay, agent, take this back and, like, deploy this application. So, and we don't know where this will go. I don't think it will go that fast there. I think spec driven development is amazing if used right. Um, like a, like we, like we said, like, uh, it's allowing you to apply AI also like not only to greenfield, like to refactor.
[00:36:44] Ori Keren: And I think, and I'm seeing some interesting companies who are starting to do things like that, that like the, the commit, the git as a, as the source of truth of like the... Okay, git is the source of truth of the, uh, of the code, but you can also infer about the process if you [00:37:00] look at git, right? You can understand, like, reviews and pull requests and all of that.
[00:37:05] Ori Keren: That's not sufficient anymore. And I'm seeing some cool companies who are s- doing something interesting and saying, no, uh, the, the atom of the, of the development, if you will, is different because it has to capture the interaction with the agent and exactly like we said, what was the agent idle time, and was my prompt good?
[00:37:25] Ori Keren: and did the agent wait for me, like, uh, long? And did I, like, do like three things in parallel? So, uh, the s- the session, uh, the developers have with agent is also becoming like the source of truth of, like, the development process. About where the code lives, I don't know if it will change that fast, but for, to infer deeply into the development process, uh, the commit and git is not enough anymore.
[00:37:50] Ori Keren: Um, and then, and then think about all the amazing stuff you can, like, extract from that, whether you, uh, by humans or by agents. So short answer, yeah, git is not sufficient to [00:38:00] infer, like, the development process anymore. You need, like, uh, something that captures the sessions.
[00:38:05] Ben Lloyd Pearson: Awesome. Well, uh, it's, it's always really great to hear the pr- your, y'all's perspective on the industry because, you know, we are really in a unique position where we're getting to sort of witness firsthand how engineering productivity is changing for the, the AI era. You know, and I, we, I, I really love the conversation around, you know, building context engines.
[00:38:25] Ben Lloyd Pearson: You know, we, we talk about that a lot on this show and, you know, insight compression is a really valuable thing that you can get out of, out of s- uh, platforms like LinearB. So yeah, Ori, Dan, thank you so much for coming on the show today. It's, it's always really great to get updates from you and, uh, we'll drop links to, to some resources in the show notes.
[00:38:45] Ori Keren: Thank you guys. You did a great job. It was great being here
[00:38:49] Dan Lines: Thanks, guys.
