Latest / Founder Mode / The End of Prompt Engineering with Dennis Pilarinos

Transcript

The End of Prompt Engineering with Dennis Pilarinos · Founder Mode

• 0:00
  I don't think you're setting them up for for success, no
• 0:01
  matter how talented they are, right?
• 0:03
  You can't talk to anyone. Don't look at any documentation,
• 0:05
  don't look at Slack, nothing. Just go look at the source code
• 0:08
  and kind of go and do that. I think that's
• 0:25
  You're listening to Founder mode.
• 0:26
  Where we talked to the people building the tools that builders
• 0:29
  rely on. Today's episodes about AI and
• 0:31
  software development, not the hype version.
• 0:33
  Yeah, we're talking about what actually breaks, what actually
• 0:36
  scales, and why context might matter more than code.
• 0:39
  There's a rush to automate everything in engineering right
• 0:42
  now. But if you automate without
• 0:43
  understanding the system, you just accelerate confusion.
• 0:46
  All right. So Kevin, how much automation
• 0:49
  versus like hands on keyboard coding is happening at pretty
• 0:53
  good AI? It's a good question.
• 0:55
  I mean, all of our code is touched by AI, right?
• 0:57
  I mean, reviews and things like that.
• 0:59
  I would say the amount of code as a percentage that's purely
• 1:03
  written by AI, well over 50%. I mean, I don't know if 80 or 90
• 1:06
  is the right number. I think it would depend on the
• 1:08
  developer. It's interesting though now
• 1:10
  because you see a lot of PRS come in or changes come in and
• 1:14
  it's clear that some AI was used, but there's a bit of like
• 1:18
  this, you kind of did the icing or you reviewed it or you like
• 1:21
  did a little Polish to get it there.
• 1:22
  And so I think some, you know, some engineers would even say
• 1:25
  that like the codes, not 100% AI.
• 1:27
  Personally, all of the code that I've written is AI because I'm
• 1:31
  like just not good at writing code.
• 1:33
  And I spend all my time like prompting it to test it,
• 1:36
  prompting it to break it to sort of like get myself comfortable
• 1:38
  that it's AI. And yeah, so I don't know.
• 1:41
  I'm curious on your side though, like as you've sort of turned it
• 1:44
  to automate your world, what kind of things are you thinking
• 1:47
  about automating? And do you see that automation
• 1:50
  has become a problem for some of the like sort of growth and
• 1:52
  marketing things you guys have run into?
• 1:53
  It's interesting. I just had a conversation right
• 1:55
  before this episode that was the basically the thesis was the
• 1:59
  exact opposite that because everything can be automated now
• 2:02
  and AI is touching everything that services businesses and
• 2:05
  people become even more valuable and important, which is a kind
• 2:09
  of an interesting twist. I was actually talking to one of
• 2:11
  our prior guests, Jonathan Sturgeon, and it was his point.
• 2:13
  He's like, no, fractional CMOS aren't going away because like
• 2:16
  people still want somebody to look at all their marketing and
• 2:19
  all the marketing agents and tell them like, is this working?
• 2:22
  Is this good? Is this bad?
• 2:23
  So I think there's that. And then, you know, I, I
• 2:25
  actually am doing something with my team and, and I think it was
• 2:28
  an idea that came from Hampton was everyone's using AI to do
• 2:30
  everything. It's fine in like the, in the
• 2:33
  context of writing code, it's very binary.
• 2:34
  It's like, Hey, this is the AI kind of output.
• 2:37
  And now we're running, running it through our kind of our own
• 2:39
  system here to do a pull request and kind of go back and forth
• 2:42
  and then push it. But when it comes to like more
• 2:45
  squishy work, like stuff that I do and live and breathe in, it's
• 2:48
  like, did you just ChatGPT that deliverable for the client?
• 2:52
  And so I'm implementing inside of my business like, Hey, if you
• 2:56
  use ChatGPT or any other AI chat bot or AI tool to create a
• 2:59
  deliverable, you need to show your work.
• 3:01
  So basically include the link to your chat along with the final
• 3:06
  product that we're going to review internally and then
• 3:08
  deliver to a client, a partner, whatever.
• 3:10
  There's no more like hiding the work, because I think it's more
• 3:13
  important than ever that people actually take a look under the
• 3:16
  hood and say. Like and so you're saying you
• 3:17
  give the client. The chat should be there
• 3:19
  internally when you're debating like.
• 3:21
  You're like an individual contributor working on something
• 3:24
  and then you have like the engagement leader person that's
• 3:26
  managing you, project manager, whatever they need to see what
• 3:30
  you did before we ship it to someone else that's external and
• 3:34
  basically that validate that it's not just a copy paste AI
• 3:38
  slop. Yeah.
• 3:38
  I like it. That's good.
• 3:40
  Well, it's exciting. Let's let's get to get us out
• 3:42
  here and see what he's up to. I think a lot of the slop and
• 3:46
  hallucination reduction is something he specializes in, so
• 3:48
  I'm sure we'll get into that as we get into the episode.
• 3:51
  Yeah, Should we bring him out? Let's do it.
• 4:03
  Dennis, welcome to founder mode. You built developer tools at
• 4:06
  scale, exited, rebuilt. Now you're tackling one of the
• 4:08
  hardest problems in AI that I personally I've seen.
• 4:11
  Is there really just this notion of AI context?
• 4:13
  And so today we're really going to focus on, you know, where
• 4:16
  does AI fit into like engineering teams?
• 4:18
  And it's kind of like, you know, what are people actually doing?
• 4:20
  And not just the demos we've all seen, you know, too often online
• 4:23
  that. Sounds great.
• 4:23
  Yeah. Thanks for having me.
• 4:24
  I'm excited for this conversation.
• 4:25
  Yeah, Dennis, really, really glad to have you here.
• 4:27
  You, you've been there at the start of major cloud platforms
• 4:31
  and dev ecosystems. What friction did you see over
• 4:34
  and over that pushed you to to build your own company?
• 4:37
  Oh. That's a great question.
• 4:38
  That's fine. No one's ever asked that before.
• 4:40
  Typically we actually use it as part of our hiring practice.
• 4:43
  We have something called ambitious malcontents.
• 4:44
  So people who are really unsatisfied with the world
• 4:47
  that's they, they operate in and kind of want to make it a better
• 4:50
  place. You typically find folks who are
• 4:53
  kind of frustrated with the software that they use or the
• 4:55
  workflows that they have or kind of how that works.
• 4:57
  And they're like, well, instead of just tolerating it and see if
• 5:00
  we can fix it. So that's kind of if our to sum
• 5:02
  my like experience for starting new projects and new companies,
• 5:06
  it's a consequence of of that kind of characteristic just.
• 5:08
  Selling your last company and sort of leading teams inside of
• 5:11
  this like a larger org. Like what made you, what was
• 5:14
  that big thing that made you really jump into sort of founder
• 5:16
  mode again? And like, and how did you sort
• 5:18
  of handle that switch? And now that you've sort of when
• 5:21
  is it and won then back out. And and then I'm always curious
• 5:23
  to sort of, you know, you've sort of navigated that.
• 5:26
  Yeah. I think very early in my career,
• 5:29
  I received this advice that I, it's always been kind of a
• 5:31
  guiding light for me, which is 85% of what you decide to do
• 5:33
  should be based on the people that you work with and then 10%
• 5:37
  on the problem that you work with and then 5% in the
• 5:40
  position. So after, so we sold the company
• 5:43
  to Apple, worked there for about a year and a half, took a year
• 5:45
  and a half off. I'm, I'm a private pilot.
• 5:47
  So just kind of buzzed around for a little bit, wanted nothing
• 5:50
  to do with tech, didn't want to talk to anyone, didn't want to
• 5:52
  nothing about it whatsoever. And then really start to miss
• 5:55
  the people that I, I've been working with.
• 5:56
  And so the folks who started unblock are the same, basically
• 5:59
  the same 1010 people as buddy build and we want an opportunity
• 6:03
  to work together. I thought about the things that
• 6:05
  kind of like irritated me in terms of being a software
• 6:08
  engineer and a software manager and a mediocre developer.
• 6:12
  And I was like, this is crazy that that it's so hard to get
• 6:15
  answers to questions in small and large organizations.
• 6:18
  And so that was the genesis of the product, which is this inner
• 6:21
  frustration and opportunity to work with some great people
• 6:23
  again. So off to the race if you got
• 6:25
  the bang back together and we've been.
• 6:27
  Jamming, I love it. I mean it's funny because Dennis
• 6:29
  and I sort of reconnected most recently in this pilot form that
• 6:32
  we're both in, so both pilots and that's cool.
• 6:35
  But I think when I went into research this, I looked back
• 6:37
  when we first met and it was when you were building buddy
• 6:39
  build, you know, October 15th of 2025, we had just got acquired
• 6:43
  by Microsoft. So we were playing with new dev
• 6:44
  tools and poking around different things and back at the
• 6:46
  time, building on Apple was quite a nightmare, especially
• 6:50
  for teams like, you know, sitting inside of Microsoft that
• 6:52
  weren't Apple and Mac shops. And so it was, it was, it's
• 6:55
  really cool to sort of see that. And, and now to Fast forward,
• 6:58
  you know, 11 years, you know, a lot of things have changed on
• 7:02
  how we build and how the software works.
• 7:04
  And, and I liked your line about being a mediocre engineer
• 7:06
  because it resonates with me two ways. 1 is that I'm, I don't
• 7:09
  think I'd ever even say I'm, I'm below that bar.
• 7:11
  I'm like, I always just say I wrote code, but not well.
• 7:14
  And, and, you know, but was a better talker than a coder.
• 7:17
  And so, you know, I've kind of graduated out of that for now,
• 7:20
  but I'm curious on the airplane stuff before we go back into the
• 7:22
  tech stuff, what sort of lessons, You know, sort of like
• 7:26
  I said, you know, you talk about everybody says, hey, we'll take
• 7:28
  a sabbatical, but you're like, hey, I just have to fly around.
• 7:30
  Like where did you go anywhere cool or anything interesting
• 7:33
  where you're like, man, this is going to be apply later and or,
• 7:36
  or sort of, you know, something you came back to in terms of a
• 7:37
  lesson. For sure.
• 7:38
  Yeah. So I left Apple basically the
• 7:43
  week after the pandemic kind of hit.
• 7:45
  And so the ability for me to like, fly around and do anything
• 7:48
  was pretty limited for a while. So I'm based in Vancouver, in
• 7:51
  Canada, and I explored one of my good buddies and I, we called it
• 7:54
  the trip of a lifetime. We literally got into the
• 7:56
  airplane and we flew wherever we could possibly go.
• 7:59
  With no destination and no time to bring your mind.
• 8:01
  People would ask, when are you getting back?
• 8:02
  And I have no idea. And so, you know, it's, I guess
• 8:05
  kind of reflecting back, like going through these journeys
• 8:08
  with people that you care about and can have a good time with is
• 8:11
  just really, really important. Yeah.
• 8:13
  I love flying. I have a small slow airplane.
• 8:16
  And so it gives you some time to do a bunch of reflective
• 8:18
  thinking. There's nothing fast and cool.
• 8:20
  Big fast airplanes Overrated. I, I feel like we could talk a
• 8:24
  lot about the adventure with no, with no specific destination,
• 8:28
  but I'm going to bring it back to to this this world that
• 8:31
  you're building, building inside of.
• 8:33
  So you've said that AI dev tools are solving the wrong problem.
• 8:37
  What are they missing? I think to date, what we've seen
• 8:41
  is really a focus on solving problems that are limited to
• 8:45
  just the scope of the source code that they're looking at
• 8:47
  right And forget about AI dev tools for a second.
• 8:49
  If you think about hiring someone new on the team and you
• 8:53
  said to them, look, we need you to build this feature, fix this
• 8:55
  bug, but you can only look at the source code, all of the
• 8:58
  information that you needed, just like limited to the source
• 9:00
  code itself. I don't think you're setting
• 9:02
  them up for for success, no matter how talented they are,
• 9:04
  right? You can't talk to anyone.
• 9:05
  Don't look at any documentation, don't look at slack, nothing.
• 9:08
  Just go look at the source code and kind of go and do that.
• 9:10
  I think that's crazy. And I think if you built
• 9:13
  software in any in meaningful capacity, you know that the
• 9:16
  knowledge sits outside of just strictly the source code.
• 9:20
  The source code is what you've done, why you've done.
• 9:23
  It sits in Slack threads and JIRA tickets, Notion documents,
• 9:27
  whatever, whatever your stack might be.
• 9:29
  And so the problem that, you know, we're trying to solve is
• 9:31
  effectively contextualizing or providing context for developers
• 9:36
  to get their work done. And increasingly and more so
• 9:38
  these days, they're tools, right?
• 9:40
  So these agents lack that context and need it arguably
• 9:43
  even more than people because they suffer from not clarifying
• 9:46
  intent or, you know, any number of like deficiencies and
• 9:50
  something, somehow some in the capacity of these things are
• 9:52
  built. So I think context is like is
• 9:55
  something we've been talking about years and something that
• 9:57
  makes a really big difference when when you're thinking about
• 9:59
  what you want to solve for people.
• 10:01
  Yeah, everyone talks about prompts, but you kind of keep
• 10:03
  bringing it back to like the context and when you think about
• 10:06
  how the workflows change, right? And I remember how like, you
• 10:09
  know, we used to hand code everything and then, you know,
• 10:12
  we went to semi automated and now in a lot of cases fully
• 10:14
  automated code Gen. you know, where you're just like prompting
• 10:16
  yourself to a full app. And you know, Jason's more of a
• 10:19
  marketing kind can go prompt lots of stuff.
• 10:21
  He's like a lovable beast. I'm curious what you've seen in
• 10:24
  your company when you sort of compare like, let's just call it
• 10:27
  the sort of state-of-the-art workflow for somebody that's
• 10:30
  like a solo dev or someone that's like, you know, at a much
• 10:33
  larger enterprise, which I'm guessing we're like your product
• 10:36
  makes a lot more sense. We're like, hey, there's just
• 10:38
  the context is a lot more scattered and quite intense.
• 10:41
  Could you walk us through sort of what you guys are seeing from
• 10:44
  where you're you are in terms of the how context is used and just
• 10:47
  how bad and scattered it is? Totally.
• 10:49
  So I think our kind of sweet spot are companies who have
• 10:51
  legacy code bases that have had, you know, 10s or thousands or
• 10:56
  10s of thousands of people who've worked on it over a long
• 10:58
  period of time. You know, a four person startup
• 11:01
  that's been working in a conference room for a year
• 11:03
  doesn't have a ton of context, right?
• 11:04
  Doesn't really have a lot of the lift that you can get from all
• 11:08
  that historical or tribal knowledge.
• 11:10
  One of the things that I, I, there's a whole bunch of like
• 11:12
  interesting human factors that we learn.
• 11:13
  But one of the hardest problems to crack, honestly, is
• 11:16
  recognizing that there isn't a single company that exists in
• 11:19
  the world that I've ever seen that feels that their source,
• 11:22
  that their their documentation is up to date and well
• 11:25
  organized, right. And So what you see, even in a
• 11:28
  small company like ours, we use a handful of different
• 11:31
  documentation platforms because different people have different
• 11:34
  preferences. So pulling all that together and
• 11:37
  then doing conflict resolution, trying to understand what is the
• 11:40
  truth for a given person and a given question, right.
• 11:43
  And so that tends to get hairier as you get larger corpuses of
• 11:48
  data and historical historical materials to kind of like dig
• 11:51
  through. Yeah, bigger kind of larger
• 11:53
  enterprises suffer from this problem more frequently.
• 11:56
  Having worked at Apple, Amazon, Microsoft, big places, lots of
• 12:00
  information, very, very cute pain point.
• 12:03
  When it was three or four of us working at buddy build on month
• 12:05
  three, I knew everything I needed to.
• 12:07
  I had it in my brain. You're like collecting big tech
• 12:10
  Pokémon cards like the mall. It's it's everybody.
• 12:14
  So where does AI generated code actually breakdown in a in a
• 12:19
  real production environment? Yeah, Typically when you have
• 12:23
  these legacy code bases, organizations have architecture
• 12:26
  standards, they have best practices, they have ways of
• 12:29
  doing things within the organization.
• 12:31
  Just yesterday I actually ran this experiment.
• 12:33
  It was super fun. Just within the unblock code
• 12:35
  base, I asked Claude Code to implement a feature for us.
• 12:39
  And you know, just like high quality engineers, these tools
• 12:42
  these days come up with break the problem up into tasks, come
• 12:45
  up with a plan, then go and implement it.
• 12:47
  So I took the plan that clawed code generated and then I
• 12:51
  effectively context enabled clawed code with unblocked and I
• 12:54
  said, hey, use the unblocked context engine as part of your
• 12:56
  planning process. Dump that plan out.
• 12:59
  So now I have two plans, 1 generated with the unblocked
• 13:02
  context engine and one without. And then I was like, let's do
• 13:04
  something very meta. I stuck it into clawed and I
• 13:07
  said, compare these two plans. Tell tell me which one do you
• 13:10
  think is better? So kind of a cousin of Claude
• 13:11
  code, I guess to see if you know it, it could figure it out.
• 13:14
  And we've run this experiment for our code base and I see
• 13:17
  customers doing it with their code as well.
• 13:19
  And unquestionably the code that's the plan that's
• 13:23
  generated. And obviously, subsequently the
• 13:25
  code that gets implemented from that plan is much, much better
• 13:28
  when it's contextualized with all those historical
• 13:30
  conversations, right? So, you know, a very concrete
• 13:34
  example, I had tried to build a feature that would do some data
• 13:37
  enrichment. The way that Claude Code had
• 13:39
  suggested we do secrets management for that API key was
• 13:42
  very, very reasonable, I would say.
• 13:45
  But it's not the way that we do it within the context of our
• 13:47
  system architecture, right? And in fact, if we were to roll
• 13:51
  that deployment out, it would likely have taken down some of
• 13:54
  the services in the way that it had proposed it, not production
• 13:57
  grade software. I ran that same experiment with
• 14:00
  unblock the context engine and it knows, oh, the way that
• 14:03
  you're storing secrets is through Vault and it's like
• 14:05
  retrieved at this part, you know, at this part of the
• 14:07
  runtime and so on and so forth. So knowing how your team
• 14:11
  operates so that you can be informed for what code gets
• 14:13
  generated. It's just super, super
• 14:15
  important. That's how that's how we see it.
• 14:18
  You described kind of this code and documentation is like
• 14:21
  decision grade context, right? And I think that the example you
• 14:25
  just gave highlighted that of like clearly it there's a bar of
• 14:29
  like it works and then there's a bar of like it's actually going
• 14:31
  to work and be scalable and buildable.
• 14:33
  When you think about kind of the the world going forward, how
• 14:36
  much of the way that you would expect sort of teams to build is
• 14:40
  like do we start using less systems of record, right?
• 14:42
  Like I always think about like in the days we had like a wiki
• 14:45
  and a Jira and like a customer docs, and then we had
• 14:49
  presentations and then we had the support system.
• 14:51
  So we'd have like very easily, even in a relatively small
• 14:53
  company, 6-7 places where like pretty good context lives to
• 14:57
  whatever the chat room you're using.
• 14:59
  Do you think there's a world where that starts to like
• 15:01
  compress where we just use less things?
• 15:03
  Or do we think like unblock sort of way of like being able to
• 15:06
  kind of make that look like one thing to code ends up being the
• 15:09
  strategy that most people pick up?
• 15:10
  I think it's both and not to like kind of and I'll tell you
• 15:14
  why. So I think there's a natural
• 15:17
  inclination for people to want to collaborate on problems,
• 15:19
  right? As they kind of riff things back
• 15:21
  and forth. You know, customer support is a
• 15:24
  great scenario where you're collaborating on very a very
• 15:27
  specific problem. But increasingly, I think we're
• 15:29
  having conversations with these agents in a way that also has
• 15:32
  some very interesting context, right?
• 15:34
  If context is the missing layer Dennis, what happens when teams
• 15:38
  just ignore it? It's like any tool I suspect,
• 15:41
  right? Which is like.
• 15:42
  You can do things the hard way if you want, but it's probably
• 15:47
  not in your best interest. For whatever reason.
• 15:49
  I always kind of go back to the idea of someone, a construction
• 15:53
  worker, a construction worker framing a house, right?
• 15:56
  They can use a hammer to frame a house or you can use a nail gun,
• 16:01
  and it still requires that framer to understand the
• 16:05
  fundamentals of how they're building that.
• 16:06
  And if you choose not to use the nail gun, that's fine, but
• 16:09
  you're probably going to be limiting the throughput that you
• 16:11
  can do in terms of your business, right?
• 16:13
  So I think you need to know what you're doing, right?
• 16:16
  You don't want crooked floors and a doors that don't hang
• 16:19
  correctly and all that stuff. And then you want to be using
• 16:21
  tools that help you get that job done very, very quickly.
• 16:25
  Do you think better context actually reduces hallucinations,
• 16:28
  or do you think that those just become easier to flag and spot
• 16:31
  because now you've kind of like, hey, the the true false thing is
• 16:34
  sort of predefined? I mean, we know definitively it
• 16:37
  helps reduce hallucinations. We're close to publishing some
• 16:40
  information that shows that we can actually beat a lot of quad
• 16:44
  code benchmarks by contextualizing those those
• 16:48
  tests. And so clog code raw versus clog
• 16:52
  code with unblocked, you get much better results.
• 16:55
  You know, hallucinations I think was a topic du jour probably 18
• 16:58
  months ago, which feels honestly, probably as long as
• 17:01
  long as 11 years ago when we first crossed paths, honestly.
• 17:03
  But like for our audience, which is predominantly engineers, you
• 17:07
  can't hallucinate answers, right?
• 17:09
  In the same way. Again, we keep talking about it
• 17:11
  within the context of AI, but no one likes working with a guy
• 17:14
  who's like wrong all the time, right?
• 17:16
  They're just going to be like, oh, this guy's an idiot, let's
• 17:18
  not chat with him anymore. You you've tweeted that tools
• 17:21
  amplify the competence of the Craftsman.
• 17:23
  Yeah, the opposite of what we were just talking about.
• 17:25
  So what happens to mediocre teams in our new AI world?
• 17:29
  Yeah, it's a great question. I think there is even if you
• 17:32
  think maybe like let's just go back like a year ago, I think
• 17:35
  there was a lot of consternation around to what degree do new
• 17:38
  grads have roles moving forward in, in building software, right.
• 17:43
  And I think I was at a conference last week, Cloudfare
• 17:46
  said that they were going to be hiring 1000 interns over the
• 17:48
  next year. Interns, right?
• 17:50
  And what they're finding is that those interns know how to use
• 17:53
  these tools very, very well. But they're presumably lacking
• 17:55
  though, is a lot of the life experience, experience that
• 17:58
  knows how to, you know, build actually high quality software.
• 18:01
  On the other end of the spectrum, what you're finding is
• 18:03
  senior engineers who have that life experience are using these
• 18:05
  tools to like enable them very, very quickly.
• 18:08
  I think the problem that you're, you know, quickly becomes is
• 18:11
  like for folks who are in the middle who don't get familiar
• 18:14
  with how the world is moving, that I think becomes a
• 18:17
  challenge. And so I think like in any
• 18:19
  industry, in any profession, in any career, if you're not
• 18:22
  constantly up leveling your skills, you're going to be left
• 18:24
  behind. There's just, there's no
• 18:26
  question in my mind. All we're debating now is like
• 18:28
  how quickly. And so I've gotten away from
• 18:31
  predicting any timelines for this stuff because it's, it's
• 18:33
  just crazy. But it's like also, if you, you
• 18:35
  don't give a tricky or enough test and we'll, maybe we'll talk
• 18:38
  about some of your hiring approaches, you know, in a
• 18:40
  minute. But I'm curious here where, you
• 18:42
  know, you mentioned this, you're like, hey, I'm running plug
• 18:44
  code. I'm you know, I'm running the
• 18:45
  company, but I'm also building and I've got my, you know,
• 18:47
  GitHub sort of green lights have like lit up in the last, you
• 18:50
  know, sort of year, just because it's so much easier to sort of
• 18:53
  get involved and you can jump into something and just like fix
• 18:56
  a bug or ask a question or shoot APR in there and just do
• 18:59
  something. I'm curious, when you coach or
• 19:02
  mentor your own team or sort of, you know, people that are sort
• 19:05
  of coming into the workforce today, what is it that you see
• 19:08
  that really separates kind of the great from like the average
• 19:11
  in terms of engineer, you know, with all the tools we have
• 19:13
  today? Yeah, for engineering.
• 19:15
  Yeah, fair enough. I think the, I don't know that a
• 19:17
  lot has actually changed. I think they almost were talking
• 19:20
  about the skills. I think it's predisposition.
• 19:22
  I think it's people who are curious, people who are humble,
• 19:24
  people who have like user empathy.
• 19:26
  Those are the three kind of like underlying principles that we
• 19:29
  look for. And then it's just a matter of
• 19:31
  like competence to say, can you learn new things very, very
• 19:34
  quickly? Because the space and the
• 19:36
  tooling does change, right? Like, I don't know, go back 1012
• 19:40
  years ago when Visual Studio launched, you know, statement
• 19:44
  complete, you know, you had a year before the next Visual
• 19:46
  Studio. Yeah.
• 19:47
  Yeah. So you're like, oh, this is
• 19:48
  amazing. But it wasn't like 2 weeks later
• 19:50
  where you're like, oh, yeah, statement complete is now
• 19:52
  obsolete. And now what you want to be
• 19:53
  doing is this. Right?
• 19:54
  Like you had a long just station period basically before you had
• 19:58
  to learn stuff. Now, if you're just like, not
• 20:00
  naturally curious. And playing with these things
• 20:01
  like every single day, you're going to get left in the dust, I
• 20:04
  think. I don't, I don't mean this.
• 20:06
  I don't like the sycophant, like I'm a very pragmatic person, but
• 20:09
  it is something that I've never seen in my career.
• 20:12
  And so when we're looking to hire folks, we're looking for
• 20:15
  those predispositions that can be applied to up leveling their
• 20:18
  skills. Yeah, You've actually talked
• 20:20
  about how you hire for success, not just capability.
• 20:23
  How does that apply, you know, in this kind of AI handling it
• 20:27
  every bit of execution now, so the interns at Cloudflare when
• 20:30
  you're building your your team and unblocked like how do you
• 20:32
  think about this? So we, I think it's when you're
• 20:37
  hiring in an early stage and an early company, every hire that
• 20:40
  you make, you know, at size 10 or 12 or 15 or 30, whatever, it
• 20:45
  fundamentally changes the trajectory of the company,
• 20:48
  right? Like it, it can really, you
• 20:49
  know, shift you one way or the other.
• 20:51
  When you're, I mean, I've run teams that are hundreds of
• 20:53
  people big, The impact of that person can have doesn't actually
• 20:57
  fundamentally change the outcome necessarily of the product in
• 20:59
  the same way. That's just math, right?
• 21:01
  So what you try to do, I think is continue to have that kind of
• 21:04
  high bar in terms of those predispositions, those those
• 21:07
  skills. And what we find is that hiring
• 21:10
  people who are owners, right, who care a lot about not just
• 21:14
  the task that they're doing, but thinking about it within a
• 21:16
  broader context of like, why am I doing this?
• 21:18
  What impact is it going to have? Those are the things that are
• 21:21
  just like crucially important to more so when you're early stage,
• 21:25
  but continue to be continue to be important.
• 21:27
  So like they are thinking about success, right?
• 21:30
  Not I. I do the thing that I've been
• 21:32
  told and and I move on with my day.
• 21:33
  How do you when you're measuring or sort of interviewing for
• 21:36
  these skills into that, you know, 2026, it's like the
• 21:39
  interview is very different, right?
• 21:40
  And so how do you interview for that and what are you guys sort
• 21:43
  of what's your process to sort of like, you know, kind of walk
• 21:47
  through bringing somebody new onto the team?
• 21:48
  I love, I've interviewed thousands and thousands of
• 21:51
  people over my career and so I tend to ask the same question.
• 21:55
  So if anyone's listening to this and never chats with me, you
• 21:58
  know what I'm going to ask, which is when you reflect on a
• 22:01
  really great day at the office, what would contribute to that?
• 22:03
  And then the follow up is like when you're frecked on a really
• 22:06
  bad day at the office, what would have contributed to that?
• 22:09
  And I mean, when I say bad, it's like, I don't, I never want to
• 22:11
  go there again, right? It's just so it's been so
• 22:13
  painful. And then the so you can't get a
• 22:16
  sense for what motivates people and doesn't motivate them.
• 22:19
  But the, the trick really is then to say, well, what did you
• 22:21
  do to change so that you didn't have that day again, right?
• 22:25
  Because if like people aren't taking control of their own
• 22:28
  environment, then you know, they're just going to sit there
• 22:30
  and tell you all the reasons why things are broken as opposed to
• 22:33
  like telling you what things are broken and how they're going to
• 22:35
  fix them, right? And so that to me has been like
• 22:37
  a really good kind of clear, well, where is this person on
• 22:40
  the spectrum of like, I'm going to go fix stuff as opposed to
• 22:43
  like, I know I'm a bit of a victim, I guess it's not
• 22:47
  capacity. Passenger.
• 22:48
  Like high agency interview questions, you're testing for
• 22:52
  internal versus external locus of control.
• 22:54
  The world happens to me versus I happen to the world, which I
• 22:57
  like. So we were talking about a
• 22:59
  moment ago, but just the speed at which things are changing.
• 23:01
  What does software development look like from your perspective
• 23:04
  in 3, five years if AI agents keep improving the way they are,
• 23:08
  or maybe even in three to six months?
• 23:10
  But like, what's your what's your take on on where things are
• 23:12
  headed? 3 five years is, I think it's
• 23:15
  just like I, like I said, I'm done predicting time frames.
• 23:18
  Like I don't know what's going to happen next week.
• 23:20
  You know, we, we've been thinking a lot about the, the
• 23:22
  standard software development life cycle or your SDLC.
• 23:25
  Let's take something very concrete like a pull request,
• 23:28
  right? A pull request, it's
• 23:30
  questionable value in the future in my mind, right?
• 23:33
  Because it's like it's meant to communicate the changes in the
• 23:36
  code base on how it evolves and solicit feedback from folks.
• 23:40
  But if you have an agent writing your code and an agent reviewing
• 23:43
  the code, and the agent can review the code and tell the
• 23:46
  agent that wrote the code how to fix it, why do you have this
• 23:49
  construct of a pull request? What utility does it provide?
• 23:53
  And a pull request is arguably one of the fundamental
• 23:55
  constructs of our workflow, right?
• 23:58
  But like, you know, we have a code review agent.
• 24:01
  People really, really like it. But I do wonder if like in a
• 24:04
  year or two years from now, does that thing continue to exist?
• 24:07
  Like, why would it? Why wouldn't I just say to, you
• 24:10
  know, my favorite coding agent, go and do this thing, make sure
• 24:14
  it adheres to the standards. Any issues that you find, go
• 24:17
  ahead and fix them. I might or might not actually
• 24:19
  review that code. And there's lots of
• 24:21
  organizations who aren't at this point, including us.
• 24:23
  Honestly, for some of the code that we generate, depending on
• 24:26
  what it's doing, we don't necessarily even review it.
• 24:29
  We're like, it's an internal tool, go nuts, right?
• 24:32
  If it's a production thing, maybe we want to take a quick
• 24:34
  pass at it, do a sniff test to make sure it doesn't 'cause, you
• 24:38
  know, an outage or a loss of data and these kind of critical
• 24:42
  failures or outcomes. But I, I do think a lot of the
• 24:45
  standard workflow and a lot of the, the tooling that exists
• 24:49
  will get completely flipped on its head.
• 24:52
  Pull request is just like the one that I've been thinking
• 24:54
  about over the last few weeks. Yeah, the pull request,
• 24:55
  interesting. There's actually a couple repos
• 24:57
  we have or we don't we have to pull request turned off and we
• 25:00
  just sort of like let people write code directly to main and
• 25:03
  just check it in and commit it or you have enough CI and stuff.
• 25:05
  So I think that, you know, and again, all these kind of kind of
• 25:08
  like AI native teams that are like one or two engineers, like
• 25:10
  a lot of them are essentially working in that model to an
• 25:12
  extent. But if you talk about like
• 25:14
  context being the Moat, and I think about this, when you are
• 25:17
  that like one or two person team, like all the context is
• 25:19
  like your head, your prompts and your like, there's just not a
• 25:22
  lot of room for that to get messy.
• 25:24
  Like as a company grows or if you're out there talking to
• 25:27
  founders, let's just say somebody who's starting a
• 25:28
  company or, you know, kind of in the early stages, like how would
• 25:31
  you sort of coach or mentor them to think about like how should
• 25:35
  they think about setting up context, right?
• 25:37
  Where, where should they sort of be building?
• 25:40
  You know, should they grab something like what you guys had
• 25:42
  and then say, OK, I'm only going to work with tools that plug in?
• 25:45
  Is there a context you don't want in this system?
• 25:47
  Like are there things you don't want like the software to be
• 25:50
  aware of? Like so like how would you sort
• 25:52
  of manage or design like context for a company if you were to
• 25:55
  sort of like, you know, give that advice to somebody out
• 25:57
  there listening? It's a great question.
• 25:59
  The the one thing I'm very reluctant to do is ask people to
• 26:04
  change their habits or their workflows, right?
• 26:07
  I think, and it's one of the things that people are kind of
• 26:10
  surprised by when they use unblocked.
• 26:12
  We don't want you to change the way that you work.
• 26:14
  Whatever feels natural to you, that's up to the software to
• 26:16
  figure out how to accommodate your workflow or what have you.
• 26:19
  If you feel that you can get more value by changing your
• 26:23
  habits, people will do that and the software that sits around it
• 26:26
  will accommodate it, right? So what documentation system
• 26:30
  should we use? Google Docs or Notion, right?
• 26:34
  You know, different people have different opinions.
• 26:35
  I have one, you know, various members of the team have one.
• 26:38
  Should we just consolidate one that has some benefits?
• 26:41
  But at the end of the day, what your carrot, what you care about
• 26:43
  is the outcome of using those tools, right?
• 26:45
  So if you're thinking about setting yourself up for context,
• 26:48
  use the tools that help you collaborate with your agents and
• 26:52
  with your human Co workers that most naturally give you the
• 26:57
  efficiencies that you need. And the rest is left up to, you
• 26:59
  know, the the agents to like be able to pull that stuff out,
• 27:03
  which is not necessarily hard, but then to reason about it,
• 27:06
  which is actually the harder. Dennis, let's come back to this,
• 27:08
  this traveling without a destination.
• 27:10
  What, what did that teach you about about maybe being in
• 27:14
  founder mode and building companies?
• 27:16
  What it What did you learn from that experience?
• 27:18
  I think the, you know, everyone loves this quote, Everyone has a
• 27:21
  plan until you get a punch in the face, right, Mike Tyson.
• 27:23
  And so I think you just have to be like comfortable with
• 27:26
  constantly with huge amounts of uncertainty.
• 27:29
  And I think today that's more so the case than it was 10 years
• 27:32
  ago. Like I remember when we were
• 27:34
  doing the fundraise for buddy build, the number of venture
• 27:37
  capitalists who told me there's no money in dev tools.
• 27:39
  Like they're just like, no one makes money in dev tools.
• 27:41
  Like repeatedly I was like, I don't, I don't think that's
• 27:44
  actually true. And so we're very fortunate.
• 27:45
  We had folks who we still work with in this second iteration.
• 27:48
  Now there is new dev tools and competition and uncertainty and
• 27:52
  the, and the way that the industry is going to change on a
• 27:54
  daily, weekly, certainly monthly basis.
• 27:57
  So in the same way that you have to like know kind of the
• 27:59
  direction that you're heading, the path, the specific path that
• 28:01
  you take to get you there, you have to be able to adjust,
• 28:04
  right? So Kevin and I share this love
• 28:06
  of aviation. We all have a plan for when we
• 28:08
  want to get there. And there's all sorts of studies
• 28:10
  where get their Itis right, where you literally pilots kill
• 28:13
  themselves in the process of trying to get to a place because
• 28:16
  they rush decisions or they don't factor in the risks or
• 28:20
  what have you. Yeah, the parallels between
• 28:22
  aviation and and building and running a company are it's
• 28:25
  served me quite well. Checklist super important.
• 28:29
  As an aviation nerd, all right, let's talk airplanes.
• 28:31
  Like when did you start flying? And like, how do you balance
• 28:33
  running a company with keeping your certifications and sort of
• 28:36
  like your hours up to date and sort of just keeping, you know,
• 28:39
  on point on both sides of the. Yeah, sounds good.
• 28:42
  I mean, the origin story of my aviation experience is a
• 28:44
  consequence of Buddy Build. At Buddy Build, I was working
• 28:48
  7080 hour work weeks and literally burning out.
• 28:50
  One of our investors who I'm very good friends with at this
• 28:53
  point, basically flew up and had a mini intervention and said,
• 28:56
  like, look, man, you're not going to survive this company.
• 28:58
  And the company's doing quite well, but you're not going to
• 29:01
  survive unless you figure out some kind of other outlet.
• 29:03
  And I've done running and yoga and weightlifting, all the stuff
• 29:05
  that, you know, I didn't pick up a musical instrument or
• 29:07
  whatever, but all the founder stuff that people typically do.
• 29:10
  And as a kid, I think most little boys at some point look
• 29:12
  up in the sky and they see that machine and think that'd be
• 29:14
  cool. And so when you look at my joke
• 29:17
  is when you look at the cost of talking to a psychiatrist and
• 29:20
  getting your pilot's license, about $300.00 an hour or
• 29:22
  something like that, certainly in Canadian dollars, like, you
• 29:25
  know, let's try the aviation thing instead.
• 29:27
  Turns out you have to end up doing both.
• 29:29
  But whatever is so, but that's literally aviation and learning
• 29:32
  to fly with such a unique different experience than
• 29:35
  anything that ever done. It was the only thing that could
• 29:38
  keep me from thinking about the business.
• 29:40
  Like that first time that you take off by yourself with no one
• 29:43
  else in the airplane and you're responsible for landing it.
• 29:46
  That is an all-encompassing all, all your attention is focused on
• 29:51
  basically not killing yourself, right.
• 29:52
  And so that has been that cemented, you know, the the rest
• 29:56
  of my aviation experience, it's just constant learning curve.
• 29:59
  And I use it as a means to balance the stress of building
• 30:02
  and running a company. So it literally changes your
• 30:05
  perspective. I love that and I and I fully
• 30:07
  agree, like the whole 40 chest thing and like not killing
• 30:10
  yourself. And it was funny when Jason's
• 30:11
  with me for the first time, he was like, dude, you're a totally
• 30:13
  different person. Like I can just hear it in your
• 30:15
  voice the way. And it's for sure like all of
• 30:17
  the weird anxieties and weird things you have going on and
• 30:20
  weird tendencies that you have like the instance you sit in the
• 30:23
  cockpit and you're like, all right, you know, thank you for
• 30:26
  take off. And you're like a different game
• 30:28
  now. And like that game is stay alive
• 30:30
  and like, but it it is a really cool sort of way of thinking and
• 30:33
  and playing. And so, yeah, fully agree on
• 30:35
  that one. My favorite good thing is taking
• 30:37
  other people up in the small airplane.
• 30:38
  Like I just absolutely love that.
• 30:40
  And especially like like I said, I live in Vancouver.
• 30:42
  It's arguably one of the most beautiful places living in the
• 30:44
  world. And if you take a small airplane
• 30:46
  and you fly just 1520 minutes north of of Vancouver, the views
• 30:50
  there I've never had a person be like well that sucked.
• 30:53
  Everyone exactly saying like that's amazing get.
• 30:57
  Me back on the ground. This is terrible.
• 30:59
  Yeah, yeah, yeah. Awesome, Dennis, where where can
• 31:01
  folks follow you online? Stay plugged into what you're
• 31:04
  building. Any any offers or anything you
• 31:06
  want to share with the Founder Mode audience?
• 31:07
  Yeah, for sure. You can hit up, you can hit up
• 31:10
  my Twitter bio. It's Dennis Pellerino, super
• 31:13
  long name, but if you type in probably Dennis Unblocked or
• 31:15
  Dennis Buddy build, it'll come up.
• 31:16
  So I'm on Twitter there, I'm on LinkedIn, happy to answer any
• 31:19
  questions or or share any kind of the life experience I've had
• 31:22
  Building 1 company, you called it winning at the very beginning
• 31:25
  of this conversation, but exited one company and then the the
• 31:29
  journey so far on #2 here. So yeah, happy to chat with
• 31:32
  folks anytime. Awesome.
• 31:33
  Well, Dennis, that was sharp, practical and refreshingly
• 31:36
  grounded or up in the air. But hey, thanks for joining us.
• 31:39
  Yeah. Thanks for having me.
• 31:40
  I really appreciate it. That was a really fun one.
• 31:42
  Dev tools flying and a guy who's lived inside Apple, Amazon and
• 31:46
  Microsoft and still chose to go build again.
• 31:48
  Yeah, and he's not just theorizing about context.
• 31:51
  He ran an experiment live with plot code and showed the
• 31:54
  receipts. All right, Kevin, what are your
• 31:56
  top five takeaways from this one?
• 31:57
  Yeah, so #1 AI dev tools focus too narrowly on the source code,
• 32:01
  and that real knowledge lives in Slack tier and dachshund.
• 32:04
  And we've seen this over and over ourselves.
• 32:06
  Contextualized AI produces dramatically better production.
• 32:09
  Safe code #3 hire that curiosity, humility, empathy,
• 32:14
  predisposition over pure skill #4 core work flows like pull
• 32:19
  requests may soon be obsolete. And and #5 you know, flying kept
• 32:23
  him as insane as a founder and taught him to navigate
• 32:25
  uncertainty. And personally, highly recommend
• 32:28
  getting into flying if you're just looking for an opportunity
• 32:31
  to sort of get away from that day-to-day drudgery.
• 32:32
  Founder Mode sponsored by Airplanes.
• 32:35
  All right, if you're building products or leading engineering
• 32:37
  teams, send this episode to someone thinking about AI the
• 32:40
  wrong way. And if you want more
• 32:42
  conversations like this, subscribe, leave a review, and
• 32:45
  share founder mode with your team.
• 32:46
  That's it for today. Because in the AI world, the
• 32:48
  real leverage isn't writing more code, it's understanding the
• 32:51
  system.