63: Why AI Coding Agents Keep Guessing | Context Gap & Code Intelligence with Dennis Pilarinos

Bethany (00:00) Welcome to the Overcommitted Podcast, your weekly dose of real engineering conversations. I’m your host Bethany, and I’m joined by…

Brittany Ellich (00:07) Hey, I’m Brittany.

Erika (00:07) I’m Erika

Bethany (00:09) We met while working on a team at GitHub and quickly realized we were all obsessed with getting better at what we do. So we decided to start this podcast to share what we’ve learned. We’ll be talking about everything from leveling up your technical skills to navigating your professional development, all with the goal of creating a community where engineers can learn and connect. Today on Overcommitted, I’m so excited. We are joined by Dennis Pilarinos.

Founder and CEO of Unblocked, a contextual code intelligence platform that helps developers understand the why behind their code. Dennis has spent nearly two decades building developer tools at some of the biggest names in tech. He helped build Azure at Microsoft, worked on AWS, and then founded BundyBuild, a mobile CI-CD platform that Apple acquired in 2018 and turned into Xcode Cloud. Now he’s tackling what he calls the real bottleneck in software development, not writing code, but understanding it.

Welcome Dennis!

Dennis Pilarinos (01:02) Thanks so much for having me. I’m excited to have this conversation.

Bethany (01:04) So excited! To kick us off, what’s something you’re currently building or learning that has you excited right now?

Dennis Pilarinos (01:10) I mean, I think it’s…

handful of different things, they’re all pretty much some form of automation of some internal tooling. I found in the past we would look for, we’d have to assign engineering resources to accommodate or accomplish, I would say like non-product tasks. And what I’m excited about now is the fact that I can basically just do that very, very quickly myself off the side of my desk. so endless number of examples. It’s amazing to see other engineering leaders kind of get back to their kind of core roots.

and build these little, mostly internal tools that help the company.

Bethany (01:43) That makes a lot of sense. I mean, you’ve said previously that writing code was never the main challenge, but it’s finding that context. So what really does that context gap look like day to day? Is it different for these leaders compared to engineers, or is it quite similar?

Dennis Pilarinos (02:01) I think, so when I think about, historically how software has been built, like you mentioned, I’ve been building developer tools for the better part of 20 years as an engineer, as an engineering manager, as a business leader or what have you. ⁓ one of the things that I found personally frustrating was the inability to get the information I need to get my job done. Right. And as an engineer, your job is ultimately to like write code, right? You’re either fixing bugs, implementing features, whatever it might be. And the way that you would get that information historically was typically interrupting your coworkers.

for spending a bunch of time digging through information. But you spend a ton of time doing that, especially if you’re working in a legacy code base with a lot of other people. And the actual coding part of it is relatively insignificant. think DX released a survey that said, the 2026 developer survey just a couple of weeks ago, where they said 14 % of an engineer’s time is actually spent writing actual code, which I thought was an interesting stat. And so that context gap is quite painful.

It’s actually funny, when we started Unblocked, we were gonna call the company Bother, which is like, you don’t bother me, I don’t bother you, right? It’s just like, we can use Bother, get the answers for the things that we need, and move on with our days very, very quickly. So, you know, it’s an interesting time. I think there’s an interesting transition where, while it’s been historically true that engineers are the folks writing code, increasingly, it’s their agents, right? And so, we’re seeing that

same

problem arise with agents. If you use any of these tools, a lot of them struggle with the context. And what do they do? Well, the same thing a person does. They kind of guess and play kind of whack-a-mole trying to figure out what things need to go get done, or what context they need in order to do their things. And so we’re finding that context gap is even more acute, I would say, today, as increasingly agents are writing even more code than the…

developers that they kind of work for.

Bethany (03:51) Absolutely, Erica, ⁓

Erika (03:51) how do you handle

the situations where you don’t have enough information? The answer is still to ask somebody. Because I find that the overconfidence in sort of AI assisted workflows can be really misleading of it says a really competent answer, it’s found all the solutions, and then I start asking it, well, how do you know that? And it’s like, well, actually, I guessed.

Dennis Pilarinos (04:16) Yeah, it reminds me, I think like in my career, I think I’ve gone through these stages in my career as well. What at some point I was like overly confident and probably didn’t have the skill to measure it with the confidence. Right. And so that’s, know, you’re not quite necessarily right out of school, but probably shortly thereafter, you’re like, I’ve been there. I’ve done that. I can like absolutely nail this thing. And you go down your merry way and then you’re like, ⁓ I didn’t, I haven’t experienced this problem or this scale issue or this like security vulnerability or any number of things.

And so that’s, feel like a lot of people have this impression of these agents where they’re like actually very, very, I would argue expert level at writing code, but like strangers to the system. They don’t have the tribal knowledge. And so they’re like, I know based on like what I’ve been trained, this is the right way to do things. And you’re like, actually not here, it’s not. So yeah, and you end up.

Yeah, this kind of almost like uncanny valley where it’s like, we should do it this way and here it is. And then you realize, nope, not even close.

Bethany (05:15) Yeah, I definitely feel that. And you’re kind of restarting over whenever you start a new agent session. Basically, it’s not even that it builds this context as it goes. Most typically, it just kind of you have to restart that process every time you you pick up one of these tools. ⁓ Okay. Yeah. Yeah.

Dennis Pilarinos (05:33) Yeah, it’s.

I was

like, it’s one of those funny things we’re just like, as people, we are a context engine, right? Like we sit in Scrum meetings, we learn and like evolve our thinking and our awareness, right? That’s why folks who’ve been around codebases for a long time are really, really valuable. imagine, I wrote this blog post, a difference between like 50 first dates in Memento. Imagine that like, you know, the way the technology works right now, certainly is that you, every time an agent gets spun up, it needs

to basically recreate that context, right? It is, and you’re babysitting it, you’re constantly spoon feeding it like this information. In, you know, 51st dates, that was Drew Barrymore, right? Waking up every morning and like there’s a system around her for her to like boot up her effectively memory, right?

But in reality, most of these systems are environments that look more like Memento, if you remember that movie where it’s like the guy had like tattooed stuff and like he has notes and like stuff all over the place. That’s actually the real world, right? And so these agents have no working set basically. They’re like, let’s see what’s around there. And they face a Memento world, which is horrifying. Great movie, but.

Erika (06:42) Yeah,

you’re right though. mean, and that also can happen to me when I like, even if I don’t look at a PR or something, or like an area of the code for a couple of weeks, then I come back, I have to like similarly like, oh wait, what does this mean? And like, how does this connect to that? So yeah, it’s a people and an agent problem. And yeah.

Dennis Pilarinos (07:06) Yeah, I feel like it’s like…

We have limited context windows, right? Like I typically don’t remember what I did even a week ago these days. I have a newborn, so my context window is progressively getting smaller because I don’t sleep much. But there’s all these external factors that play into it. But yeah, it’s exactly right. I often don’t remember the decisions that we made that influenced why we built the system a certain way or any number of these types of things that is actually codified, right?

It’s in the system and records. It’s in our pull requests. It’s in Slack. It’s in our documentation. It’s in the source code. But we don’t, for whatever reason, we don’t want to give agents a holistic view of that typically right now. So that’s kind of the pain point. And they’re really good at it once they have it.

Erika (07:48) You also don’t want it to be like deterministic. Like you don’t want it to be like, because we did it this way, now we always have to do it this way. Like there is some nuance there of like, this is a decision that we made, but we’re going to recognize that that

Bethany (07:48) here.

Dennis Pilarinos (08:00) Yeah, absolutely, absolutely.

Bethany (08:02) think too, Brittany and I went to the Pragmatic Summit, which there was a really good conversation there about AI ⁓ adoption and how companies with good systems tend to adopt AI better than companies with maybe like systems that need work still. So I think if you’re solving that human issue with context, it often helps the AI also solve the context issue for those sessions.

How exactly does Unblocked combat this or does it also help the human side of things as well as just the AI agent things side of things?

Dennis Pilarinos (08:36) Yeah, no, it does both actually. help engineers save time and…

agents kind of stay on track is the way that we think about it, right? Because they just like, they go all over the place. But yeah, we started by helping engineers, right? Like I don’t want to have to be interrupted or interrupt my coworkers. So we have a handful of surfaces that give people a very natural Q &A experience. They can ask questions in a web experience. They, I mean, I think one the hardest parts about building developer tools is breaking into the habit loop of an engineer, right? Like having someone…

have to remember to do something in order to get some value, it takes a while. It’s actually a really, really hard problem to solve. And so we created an experience where in Slack, and this is super popular in like help channels, right? If you, have customers who have five, seven, 10,000 people in a help channel. And what do people do? Actually, I can think of one. It’s like help dash GitHub, ironically. These, this organization has a help dash GitHub and people come in and they just ask their question, right? Like it’s the first time it’s ever been asked.

it hasn’t been, but Unblocked, you know, that’s their workflow. They go in, they type their question, they’re waiting for someone on the support team to answer it. Unblocked helps alleviate that support load because it sees that it can answer that question. And people love it, especially if they’re geographically distributed, time zone shifted, any of those types of things where they’re getting a high quality answer very, very quickly. And so breaking into a person’s like workflow where they’re getting AI providing value to them, especially at a time where they’re trying to like move on with their work, right?

they’re stuck on something. We had a customer say that they’ve reduced their support load by like 50 % because Unblocked is helping them, right? That’s just people. And so we started with that, right? Making it part of the workflow where you are, you if you can ask questions in Slack, Teams, in this web dashboard, a standalone Mac app, any number of ways for people to ask questions. And I would say, I don’t know, maybe the last six to eight months, it’s been very clear that agents suffer from that same problem. And so we

we

plug into that agent workflow as well. It’s kind of crazy, right? Because quickly you can help solve that problem where people and agents get this information. The next thing, and I think a lot of teams are experiencing this, is that you’re now running into this bottleneck where you’re stuck on pull requests, right? All of this code is getting generated. Now you have a ton of pull requests that have been, and we see this in the open source community a whole bunch of places, right?

Yeah, we had kind of felt that ourselves and we built a code review product that’s based on the context, right? Like, I want code reviews as if they came from like the best engineer on the team. And so we built it internally just for ourselves because we weren’t happy with kind of what existed in the market.

we gave it to a handful of folks that were customers to be like, what do you think of this? And people absolutely lost their mind. And so we’re kind of helping people not only write more code, but actually make sure that it’s the right code for their organization. And we offer this code review, like context-powered code review product that helps as well. So it’s a crazy time, yeah, for sure.

Brittany Ellich (11:30) I think that’s super interesting. And I feel like it’s solving one of the biggest problems with AI, is like making sure that it has the right context. But I’m curious your thoughts on like how people maintain relationships with their coworkers now when we’re moving towards this world where we can use AI to get all the answers. Previously, I used to reach out to my coworkers and that was like our interaction now. And now I’m finding myself like, have to like deliberately try to build those relationships and be like, Hey, how are you? Like, instead of just relying on.

⁓ you know the the touch points i’m like hey can you can you review this pr for me ⁓ do you have any thoughts on that or how that’s changing

Dennis Pilarinos (12:02) Yeah.

I mean, I think with every technological shift, I think there’s some unfortunate casualties that we have to be mindful of and figure out how do we maintain. People are inherently social creatures. I’m actually somewhat of an introvert. And so I have a certain amount of social tendencies, and after a while, I’m tired of it. But I think a lot of the technology that we see today

It’s very easy for people to just become siloed, especially if you’re not working in an office with people, right? If you’re working from home, you have to make conscious efforts to like continue to build that social fabric. And because at the end of the day, that is in my opinion, how like the best teams operate. So I don’t know that I have like a concrete solution for that, but it’s definitely something we’re seeing, right? We see people who say that they are kind of AI fatigued. It’s not, it’s not like working with a person in many ways.

So yeah, don’t, I’m curious to see how that plays out over the next, I’ve actually have a moratorium on stating any kind of timelines now because it’s so incredibly difficult to predict over the next little while. I’m gonna say three months, six months, a year, who knows.

Erika (13:06) I think of it a little like, there’s a whole discussion of what’s valuable to talk about in person versus remote. And you’re talking about shifts in technology, video call technology, change the way we work, And a lot of what I’ve heard of, what’s valuable to have in person is,

these brainstorming sessions, really like building those personal connections, sort of like the higher level thinking questions, like the procedural stuff, like everyone get on a PowerPoint and look at these slides. You know, nobody’s got time for that. But in a video call, you can probably kind of put it on the background. It’s probably fine to have that in some kind of like a remote meeting or something like that.

And I think of it a little bit similarly with sort of conversations with my coworkers in AI assisted working environments where like I might kind of do the investigation with the agent. I might, you know, like do some spelunking, look at some of the code, like build sort of that mental model. And then when I do have that conversation, like, hey, what do you think about this? Like it’s still valuable to have that human input of like.

checking that, know, sanity checking, like does this align with your design values? Does this align with our like system values? Like is there anything that I’m missing here? That like that kind of stuff ⁓ I think is still valuable conversations to have with another human.

Dennis Pilarinos (14:30) I think, yeah, I really like this idea of honing in on the creative type workflows, anything that requires some level of creativity. I think it would also add, I think there’s a lot of value in friction, right? And so actually evolving your idea by talking to someone who might actually have a different perspective. I think a lot of the tools today, to your point, it’s like, absolutely, you’re right. This constant positive affirmation and sometimes maybe not the friction or pushback that you need to really evolve the idea.

you know, for now, I think people are a great way to combine those two.

Bethany (15:01) I completely agree. I’ve found as a fellow intrepid that I have felt more more siloed lately just because we’re removing a lot of those touch points now with agentic development. And so recently I’ve found it’s helpful to even push beyond, to like force yourself to actually pair with people even though it’s not as much of a need anymore because you’ve got your permanent pair

programming buddy, but you learn so many cool things about the code base or things that you might not be touching or even tricks on how to use agents when you’re programming. So it’s been really helpful to push past that and try to force myself to talk to people for sure.

Dennis Pilarinos (15:42) Yeah, it’s, find the, the, I think of like, there’s kind of almost like an AI adoption maturity model where people go from like very, very exploratory kind of like, I’m using AI to help me do kind of tab complete to like people in the middle who are like, we’ve got MCPs and skills and like, you know, I’m running agents locally on my machine, maybe a handful of them all the way at the very end of the spectrum, which is like, have fully autonomous orchestrated agents that are checking in code into production. don’t even look at the code, right? That kind of that whole spectrum.

and I don’t know

maybe with the exception of one or two companies that like everyone is on like on the same end of that spectrum. And so to your point around even sharing how people are using these things and like helping navigate, well, this is why we do it this way or we’ve adopted this part of that spectrum has been really eye opening even for us, right? We did this exercise a month and a half ago, two months ago. We’re our most prolific engineer in terms of lines of code, which is, whatever. He didn’t use any AI coding tools.

none. All the code that he’s written is all handwritten. He used AI as a means to rob or duck basically, to talk to someone and get feedback and input. And then we had other folks who were like, we have this internal full orchestration system that will kick off these long running jobs, will implement features for us. And so we’re like, you should try this. He’s like, oh, this is amazing. We’re like, yeah, yeah.

Even sharing that is like this kind of very eye-opening experience, I think, for people within organizations.

Bethany (17:12) Absolutely. kind of diving into that future of being a software engineer. I know you recently had a post on Twitter about devs who rely primarily on vibe coding eventually hit a wall if they don’t understand the fundamentals. And I’m curious, what is the line between leveraging AI to move fast or even just autonomously committing code and checking

code versus ⁓ understanding and needing to understand what’s under the hood.

Dennis Pilarinos (17:43) Yeah, I think there’s a lot of, there’s a lot of.

noise in the ecosystem around like you can vibe code anything, right? Like I could, you know, one of the folks that we work with used to work at GitHub and it’s like, yeah, I mean, GitHub is just like a layer over top of Git. You can go and try and build that and see what that looks like, right? There’s, it’s far more complicated than that. And so I think even in the macro economic climate, there’s a lot of SaaS businesses who are taking massive hits in terms of their stock price because people think they can be replaced or supplanted by

people who are just building their own internal apps. I think these are like extreme views of the world, right? I think, you know, building…

Certainly in existing code bases like legacy teams and trying to build something without knowing how that system works You can probably do a handful of things up to a certain point, but then when it goes wrong How do you get yourself out of that corner? And so I think it’s very exciting for a lot of people to be like look what I’m able to do from scratch Most you know, I’ve been building software for 20 some odd years. It’s very rare that you have to go file new project. Let’s start something

from scratch. It’s always within an existing system. And so if you use these tools without understanding how that system works, when something goes wrong, how do you figure out how to resolve that, right? When the AI is like, I don’t know, and it just goes into this AI doom loop where it just keeps going back and forth between solutions that don’t resolve the issue. Then what happens? It’s probably good for you to understand the system. That’s kind of my perspective. Do you folks feel differently?

Erika (19:13) Yeah, I also think of like the familiar kind of adage or I don’t know if this is how widespread this is, but I feel like I learned pretty early on that like the last 10 to 20 % of a project is always the hardest part. And yeah, that first, initial, like the initial commits getting like the bulk of the work.

usually goes pretty fast and then you try to finish it up and there’s always like all these last little things that are kind of tricky and you have to figure out how to fit them back in and have to figure out how to deploy it like and yeah I mean it kind of reminds me of the same sort of percentage logic I’ve heard with

like LLM assisted coding where they’ll get you 80 % of the way and then the rest of the 20 % you have to do yourself. yeah, whenever I hear people say like, SaaS is dead, I’m like, I don’t think you’ve ever built software. Like I don’t think you’ve ever like actually deployed anything to production because that’s just not how.

Like, sure, maybe like running something locally, but like something that’s at scale that other people are going to use. Like, what do you… I don’t think you know what you’re talking about.

Dennis Pilarinos (20:19) Yeah, yeah,

I mean even simple little apps that I build like for myself, know, I’m hacking away on something over the holidays like in Christmas, right, to like automate some home automation stuff that I’ve been doing. It gets deployed to like some AWS infrastructure, calls a handful of APIs, so on and so forth. It broke.

And I was like, I don’t know exactly how that all works. I was like, I’ve long paged that out and like, who knows what API it is. And like, and now I’m sitting there having to go and maintain it and like update it. And I was like, that’s one project. if I had 50 or a hundred of these things, right? You know, if my wife walks in and presses a button and the lights don’t turn on, that’s not a good place to be. Like it has to work. Right. And like, I can’t, I need to make sure that it’s, it is reliable. Right. So, um, these production kind of scenarios.

are

reality checks, I think, for a lot of folks.

Bethany (21:08) absolutely. think one of the best parts of these systems is that you basically have a super knowledgeable, like we were talking about before, a super knowledgeable coworker. And so it’s been really interesting to be able to poke its brain, like really dive in and understand how things work. It would be easy to fall into the trap of just pushing it up. And sometimes, sometimes for small things, I am like, ⁓ it’s probably fine.

But then I’ll read the code and I’ll be like, no, actually, wait, wait, wait, nope, it’s not fine.

So it is just so important to still have an understanding ⁓ beyond it. And I think it’s just the same as people who said, you can just copy and paste from Stack Overflow and it’ll just work. The thing is, you have to know what to copy and paste from Stack Overflow for it to work. I think at least now it’s hard to imagine a future where that won’t be the case anymore. But who knows? I mean, I’m curious what your thoughts are if the role of a developer is going to

change meaningfully in the next three to five years with these tools or if it likely, like we’re saying, still rely on having the knowledge, the fundamentals to diagnose these issues.

Dennis Pilarinos (22:18) For sure, I think.

Like again with every technological shift, think people in whatever trades that they are need to like up level their skills, right? So I use this example. I think about this all the time, which is you drive by construction sites and you see people who are framing houses or putting up buildings, right? It’s very, very unusual to see them using a hammer to do framing, right? They’re using nail guns because they can get their job done faster. But like you really want to

understand

how you line up the two by fours and make sure that everything is plumb because at the end of it if you get something done very very quickly but like it’s slanted that’s not an acceptable outcome right and so like while these tools help you up level the skill I think you still need to have that underlying skill you know I remember probably a year ago people felt like

that there was going to be like this, that people who were coming out of university or colleges didn’t have jobs, right? They’re like, if you don’t, you know, and so there was like this fear, I think, in the economy. And I had a friend who had, who whose kids were having difficulty getting internships or entry level positions. And I think that’s, you know, there was some concern around, well, AI is just going to take these jobs away.

but we’ve actually seen play out, and I’ve seen this at a number of companies. There’s a company that I know that plans to hire 1,000 interns this year, right? Because those folks are getting exposure to those tools and learning how to like update their, like use their, develop their skills materially with the assistance of those tools. In that organization, they also have these very senior folks who recognize that these tools can be helpful and are quickly up-leveling their skills as well.

The observation from the CTO was it was the folks that in the middle that actually were kind of reluctant or weren’t picking up these skills. He’s like, those folks are the ones who are kind of stuck, right? So they have to like figure out what they’re going to do. Otherwise they won’t be able to keep up with the direction in which the industry is moving. I thought it was an interesting like shift between like you’re new, no jobs because AI replaces you to actually, know how to use this tool really well. We’re going to teach you the fundamentals while you show us how to

use this tool. I thought it was super, super cool outcome.

Brittany Ellich (24:32) Yeah, I think that also we go through these periods of time every few years too where we’re like, there’s no entry level jobs. Like I remember that when I graduated from college too, and it was, you know, it’s like feast or famine. I think it’s just a cyclical nature of everybody saying like, we’re going to offshore everything to, no, wait, it’s better to have things here. And then, AI is going to do everything. And like, wait, no, we still need humans to do, to do this work. So, but I do talk to a lot of,

you know, folks in college right now studying computer science that are like, do I need to go do something else? Like, do I need to pick something else? I’m no, I don’t, I think it’s going to be fine. You’re going to be well needed. Yeah, exactly. Yeah.

Dennis Pilarinos (25:07) Yeah, like you should probably learn how to use a nail gun. You know what mean? Like, as you probably

feel, it’s like, don’t come out of there with like a rock or a hammer, right? Like, learn how to use a nail gun. But like, those fundamentals still need to exist.

Bethany (25:20) Okay, before we wrap up, I actually wanted to pivot to ask a bit about your career. You’ve had such a cool path in DevTools with building Azure, working on AWS, and starting BuddyBuild. Did you have any particular moment where you knew you wanted to leave Big Tech and go build something yourself? Was there anything that inspired you or gave you that itch to go off and do that?

Dennis Pilarinos (25:45) That’s a great question. think I’ve been very fortunate in my career, even working in like big companies where I’ve had the flexibility and opportunity to like start early stage projects. like kind of like quote unquote a startup within a large company. And so that’s a lot how the Azure project actually started, which is like, you know, I had a direct internet connection in my office and I was like, oh, it’d be really cool if we had these APIs that developers could like copy. And I remember like campaigning the Microsoft IT folks to be like, can I just get a static IP address and I’m going to register this domain name.

and

they’re like, what the hell are you doing? And I was like, it’ll be fine, it’ll be fine. And so that rolled into this crazy kind of cloud platform. But same thing with AWS. I started the Vancouver office here, was the first AWS employee. And I think there’s like 3,500 folks that are here now. And so just working for AWS, let alone Amazon. And so the next kind of opportunity was BuddyBuild.

If I look back at my career, like I think that…

I’m just a very mediocre developer and like a very impatient person. That’s what it comes down to. I was like, why is this so fricking hard? Right? So I was like, I’m going to call these APIs or like, you know what I mean? I’m like, I want to do get push and a build happens and I get it deployed to my device. Like this shouldn’t be rocket science. Um, and so when I run into these problems or like even now I was like, I don’t want to interrupt the coworkers. I just want to get this stuff done or like, I don’t want to have to dig through this to just get the agent to do it, but like do it right. You know what mean? Do it the way that we do it here. Um, and so ultimately.

those two things. I’m like again not a prolific engineer by any stretch of the imagination. I’m like okay and I’m very impatient so I try to find these. I think the intersection of those two characteristics are what resulted in the career that I’ve had today.

Bethany (27:24) That definitely makes sense. think that’s such a cool experience and so cool to see, to build a lot of the things that we use day to day that are just a given from the ground up. What a cool experience. I’m curious for all your experience building these dev tools and a lot you, it sounds like you were building with yourself in mind as a developer, but do you have any lessons that you took away from what developers

actually want versus what they say they want. As we know, developers are an opinionated bunch, so I’m curious how you sort through those in your head.

Dennis Pilarinos (27:59) Yeah, think it’s always, think, you know, I have lots of friends who started different companies. think if like, if you can build a product for developers that they love, I think that’s a pretty remarkable thing because I think human nature in general is not to say nice things. Like positive feedback is pretty hard to get. And I think developers are a special breed where, you know, we still argue with like tabs and spaces, like these most arcane things, right? And so like, if you can get a developer to say, man, I really love this thing, that’s like an extra compliment. Cause you’re like, you’re not going to be breed.

as a person, let alone in this profession. And so some of the lessons, I think, you what people say versus what they want, there’s always this tension between being able to obfuscate a lot of the complexity, but still give people the option to double click and drill into the details.

My favorite example of this is when we launched Buddy Build, the signup experience was like you point it to a repo and at the end of like a couple of minutes, whatever it was, you get a green build. And people thought it was bullshit. They’re like, no, no, there’s no way you could have possibly done that. And we’re like, no, no, it actually is like a legit build you can deploy to your device. And people were highly, highly skeptical. And so we’re like, okay, well, we’ll just show them like the build output. So like we literally had to figure out a way to stream build logs

off of some remote machine to a person’s browser so they could actually believe that it was actually happening. And so once they saw that, they’re like, OK, this is actually real. So there’s always this tension between make it just work, but some folks want to be able to like, I want to change the prompt, or convince me, or show me, prove to me that it’s actually doing the thing that you say that you’re doing. So finding that on the spectrum, like,

I find folks who work in non-engineering roles are actually very happy not knowing how it works. Whereas engineers who are inherently tinkerers and be like, which things people want to know about is hard to figure out sometimes.

Bethany (29:53) Absolutely. I feel like these tools almost have to be like an iceberg, like simple enough that you can just run with it and then complex enough under the surface that you can really tinker and configure it the way you like it.

Dennis Pilarinos (30:07) That’s

exactly right. Yeah, that’s a great way of thinking about it for sure. Actually, I think in one of the talks, that’s actually the slide that I use, which is like what you see versus what exists underneath. yeah, it’s perfect.

Bethany (30:18) Amazing. Maybe that’s where I got it from.

right. Well, this has been such a cool conversation. It has been great to pick your brain about your experiences and how you’re thinking about software. As we wrap up, we always like doing a little fun segment. And so for this episode, ⁓ we’re doing a little game called Merge Fork Depricade. So Merge is the one you’d integrate into your life permanently. Fork is the one you’d take the best

parts from and then go build your own thing, and then Depricate is the one you’d sunset and move on from. So, this first one, it’s not at all inspired by any other game, it’s fine.

Dennis Pilarinos (30:56) I don’t know what you could possibly be referring to.

Bethany (30:59) Exactly, Okay,

so round one can be interpreted however, but we’ve got Microsoft, AWS, and Apple. I’m happy to go first if you want to noodle on it a bit and then we can go in little circle.

Dennis Pilarinos (31:13) That’d be great. Yeah.

Bethany (31:15) So I think I would I would merge Apple. I was a lifelong Android user and switched to Apple a few years ago and honestly you have not looked back. You just, you kind of have to to actually use Apple products. You just have to say, yep, I’m integrating this into my life permanently. Microsoft I would fork and AWS I would deprecate.

All right, Erica, what you got?

Erika (31:38) you

Yeah, I think I’m also merging Apple. It kills my soul a little bit, but I guess I’m with you. I’ve made the choice, so I guess I have to live with it now. ⁓

Dennis Pilarinos (31:50) Seriously.

Why does your soul?

⁓

Erika (31:56) Wait,

⁓

I don’t know. mean, I think just like the… Like the… There’s just a lot, a lot of like… I think mostly like the production line and…

a lot of the like business ethics around like this is why it kind of feels like a like a devil’s a devil’s choice here. I’m like, there are parts of these that I don’t like about all of these. you know, yeah. But as far as integrating into my life, yeah, I am an Apple user. I think I would fork.

Brittany Ellich (32:22) Yeah, they’re all not the best decisions in the world. So you can ignore the, yeah.

Dennis Pilarinos (32:27) yeah.

Erika (32:33) AWS and deprecate Microsoft.

Brittany Ellich (32:36) Yeah.

Dennis Pilarinos (32:40) You go next.

Brittany Ellich (32:41) okay. I, ⁓ let’s see here. I think I would merge AWS actually. I, I think I wish I could fork too. I don’t think I really want to merge either of them, but I would merge AWS, ⁓ fork Microsoft and then deprecate Apple because I don’t eat Apple in my life. There’s nothing. I mean, I guess I’m working on a Mac, but that’s about it. I could easily move to Windows computer.

Dennis Pilarinos (32:43) you

Sure. having worked at all those three companies, there’s so many different ways of looking at them, right? And I have friends who work still at all three of these companies. So I’m kind of in this precarious position. One of the things that…

Like I have a sense for the culture and it’s not something you can extrapolate for the entire organization, but certainly for broadly speaking. So I’ll kind of speak to like culture and or ⁓ focus on customers. So we care a lot about an end user experience with our software. So when I think about the AWS console, that’s something I would like absolutely deprecate. That is just like brain damage upon brain damage, right? Like I don’t know how to use that thing every single time.

like something honestly AWS in general doesn’t really care about what it looks like it just needs to kind of work. When I think about culture though I would absolutely merge aspects of AWS’s culture. Like they have these leadership principles that are very very strong, highly referenced in their day-to-day like operating and I think that’s I really really respect that and kind of admire it. But from a user experience

Not a single AWS product has been like, this is amazing. that. Apple, I would…

probably deprecate because they have very strong opinions for how they think the world should work. And having seen this play out in the past at a place like Microsoft, I think their lack of customer focus, certainly for the developer ecosystem, I find a little bit, not something that I’d wanna like merge or fork. Microsoft…

I have a soft spot in my heart for Microsoft. It’s when I was probably ⁓ most naive in my career. It’s probably some of the most fun. I’ve had the longest relationships with folks that I’ve met there. I think Microsoft would be probably fork, maybe merge. I don’t know what Microsoft is like today. It’s kind of on the order of 20 years later.

there seems like a lot of people working on the same thing that hurts my head from the outside, but I’m sure it’s understood from the inside. So I don’t know that I gave a super straight answer, like, you know, on those couple of dimensions, that’s how I think about those companies. You can learn from all of them, some things you want to like take with you, some things you alter, some things you’re like, I just think that’s like fundamentally broken. I don’t know, does that answer the question?

Bethany (35:19) I think that does. It adds a lot of nuance to your decisions and that effectively it’s very complex what you see on the surface versus what you experience internally. So that does make sense. All right. For round two, we’ve got vibe coding, pair programming with AI, and agentic AI.

Dennis Pilarinos (35:29) Yeah, yeah, yeah, for sure.

Bethany (35:41) I can kick us off again and we can go in a circle, but I would say probably…

I would merge agentic AI because that’s just how it seems like we’re going and that’s what I use most day to day. I would fork Vibe coding and deprecate pair programming with AI. I think just because Vibe coding has some interesting things to take just by kind of not worrying as much what’s happening ⁓ in the code and letting AI do what

what you do, but also validating ⁓ what the output is. And I think that’s something that will become more important as we go along and seems to be some, what some people are thinking in the industry, but yeah. All right, Erica.

Erika (36:23) I would at this point merge pair programming a lot of this like rubber ducking that we’re talking about. It would fork agentic AI and deprecate five coding.

Brittany Ellich (36:34) Interesting. I feel like I would need a more nuanced definition of both of them on vibe coding. Maybe that’s I’m going to fork vibe coding just for that reason. I would merge agentic AI just like Bethany. That seems like the direction everybody’s going. So I would love to figure out what that is. And yeah, deprecate pair programming with AI. I think that there’s still like a lot of value from it. But I find that I

I’m giving more and more of the reins over to AI anyway, so feel like I’m going more towards the vibe coding side anyway. So, Dennis?

Dennis Pilarinos (37:03) I think I’m

basically exactly aligned with Bethany. I would like merge agentic, deprecate pair, and like fork vibe for the same reasons actually. We see it almost identically.

Bethany (37:14) Yeah, absolutely. I do think there’s value in power programming with AI, but this is about having strong takes, no nuance. All right, finally a very relevant one. Reading the docs, grepping through source code, or asking the person who wrote it. So I think I would say I would merge.

Dennis Pilarinos (37:29) Ugh.

Bethany (37:36) grepping through the source code because that’s always going to be the most the source of truth. I would fork asking the person who wrote it because I do like social interaction and they tend to have the most context as we were talking about today, but maybe an agent will replace that soon enough. And then I would deprecate reading the docs because they get out of date a lot. Yeah. Erica, what you think?

Erika (37:58) Same, same reasons. Yeah, I mean, I love docs. I love well-written docs, but they’re so often out of date and yeah, it’s unfortunate.

Brittany Ellich (38:08) I think I would merge, grepping through the source code, but I think that that’s probably a new answer for me because I would use an AI tool to grep through the source code. I would never do that myself. Maybe read through it and glance through it, but if I had to type something in a terminal, it’s a bad day for me. So I wouldn’t use that. But I would fork reading the docs, I think. I think that there’s more parts of documenting.

that I hope are, you know, I feel like it’s taking off more and more now as for like, yeah, context is important. So I feel like docs hopefully are getting better. And then I would deprecate asking the person who wrote it because that’s often not even an option. You know, and I don’t remember what I wrote three months ago and why I wrote it. So if it’s not written down somewhere, then then I’m going to have some problems. Dennis.

Dennis Pilarinos (38:53) I think, so it’s interesting, as users and customers come on to Unblock, one of the first things they ask is like how…

Like what do we need to do in order to get our docs to like be AI ready or what have you? And I was like, I’ve never met a single team ever that says our docs are up to date and well organized. Like it just does not exist, right? And so I think just that’s the kind of state of the world. Minutes after you’ve written the doc, it’s kind of out of date. So I think I would probably like probably deprecate that. I kind of also want to deprecate.

merging like the source code or sorry, ⁓ gripping through the source code. but I would probably merge that because I think that is like the source of truth. and then I would fork asking the person who wrote it. Cause I do think there is still value in human interaction, especially to the points that we talking about earlier around like creativity, right? Understanding why was, why did we build it this way? Have you thought about doing this? Should we try that? You know, that thing might’ve been written two years ago and there’s opportunities.

for improvement. So that’s I think where I would end it.

Bethany (39:59) Yeah, that makes a lot of sense. I think that’s a really great point that hopefully asking questions to coworkers will not be an interrupt thing anymore, but actually a really fruitful discussion of tools like the ones that you’re building with Unblocked. All right, Dennis, this was such a fantastic conversation. It was so great to chat with you. Where can folks find you if they want to hear more?

Dennis Pilarinos (40:21) Sure. So yeah, first, thank you so much for taking the time. I really enjoyed this conversation as well. I love thinking about both like the technical and human aspects of like how this technology is impacting our ecosystem. You can find me Dennis at getonblock.com. You can go to getonblock.com. I’m on Twitter, but I have a giant long Greek last name. So good luck trying to spell it, but it’s at Dennis Pilarinos. So you can try that as well. Yeah. And obviously you can hit me up on LinkedIn or email.

I’m definitely findable. It’s not quite a unique, like kind of a GUID type identifier, Dennis Pilarinos, but there’s only like two or three of us, so you’ll find me.

Bethany (40:56) Awesome. And we’ll make sure to link all of those in the show notes. ⁓ Well, thank you again for joining and thank you so much for tuning into Overcommitted. If you like what you hear, please do follow, subscribe, or do whatever it is you like to do on the podcast app of your choice. Check us out on Blue Sky and share with your friends. Till next week. Bye.

Dennis Pilarinos (40:59) Awesome.