PE firms report AI efficiency gains but struggle with EBITDA lifts due to data and QA bottlenecks
Summary
Dave Mangot and hosts discuss why 51% of private equity firms report AI-driven engineering efficiency, yet fewer than 6% achieve a 10% EBITDA lift. Poor data infrastructure is identified as the primary blocker, surpassing talent and model limitations. The conversation highlights the risks of 'democratizing development' by non-engineers, which bypasses quality assurance and increases system instability. Speakers argue that AI shifts bottlenecks from coding to product management and testing, creating 'complexity debt' when debugging time exceeds generation time. Effective AI adoption requires optimizing underlying processes and using metrics like deployment frequency rather than just adopting new tools.
Key takeaways
- 1 Fifty one percent of private equity firms say that their portfolio company engineering teams get more efficient with AI, yet fewer than six percent can show you even a ten percent lift in EBITDA across the portfolio.
Unidentified speaker (speaker_0) · 0:00
- 2 AI shifts the bottleneck from engineering capacity to product management and quality assurance.
Unidentified speaker (speaker_1) · 19:38
- 3 Meta data shows commits increased 220%, releases 36%, incidents 40%, and firefighting 70% with AI.
Unidentified speaker (speaker_0) · 20:13
Episode sections
- 0:00 – 7:52
The AI Efficiency Paradox in Private Equity
Speakers highlight a discrepancy where 51% of PE firms report AI-driven engineering efficiency, yet fewer than 6% achieve a 10% EBITDA lift. This suggests that efficiency gains are not translating to financial results due to underlying operational bottlenecks.
- 7:52 – 15:56
Data Infrastructure as the Primary Bottleneck
Poor data infrastructure is identified as the top blocker to AI success, ranking above talent and models. AI exposes existing flaws in data plumbing, such as reliance on manual spreadsheet exports, which prevents scalable automation.
- 15:56 – 18:34
The Trap of 'Democratizing Development'
Dave Mangot expresses concern over the trend to 'democratize development,' where non-engineers use AI agents to write production code. He argues this approach bypasses necessary engineering rigor and quality assurance, leading to unstable systems.
- 18:34 – 28:27
Stress Testing Product Management and QA
AI is shifting the bottleneck from engineering capacity to product management and quality assurance. While code generation speeds up, the lack of rigorous testing and clear product requirements results in increased incidents and firefighting rather than value creation.
- 28:27 – 31:34
Security Challenges and AI Agents
Speaker_1 discusses recent security vulnerabilities (Linux kernel, PHP) and the inefficacy of certain approaches in addressing them within the context of AI-driven development.
- 31:34 – 32:23
Cost-Efficient AI for Compliance
Speaker_1 highlights a portfolio company using a cheap, open-weight AI model to continuously monitor code for PCI DSS violations, demonstrating that effective AI does not require expensive frontier models.
- 32:23 – 35:18
Engineering Alpha and Tool Selection
Speaker_0 and Speaker_1 discuss how high-performing engineering teams choose the right tool for the job to protect margins, rather than defaulting to expensive solutions or unfamiliar technologies.
- 35:18 – 39:39
Portfolio Ops AI Hiring Trends
Speaker_0 notes that AI-related titles in portfolio operations have grown from 5% to 25% of teams, but questions whether this represents genuine expertise or just fundraising optics.
- 39:39 – 45:32
Build vs. Buy Strategy
Speaker_1 argues that companies should only build custom solutions for tools that are core to their business differentiation, citing examples of a custom storage engine and CRM that significantly reduced costs.
- 45:32 – 49:53
Automating Broken Processes
Speaker_1 identifies the most common mistake in AI adoption: applying agents to broken processes and flaky tests, which results in automated inefficiency.
- 49:53 – 52:06
Deployment Frequency as a Key Metric
Speaker_1 recommends deployment frequency as the primary metric for boards to assess engineering health, arguing that daily deployment indicates the automated infrastructure necessary to leverage AI effectively.
- 52:06 – 55:46
Feature Flagging in Regulated Industries
Speaker_0 and Speaker_1 discuss how implementing feature flagging allows companies in regulated industries to deploy daily while keeping features hidden from customers, improving security and operational readiness.
- 55:46 – 56:46
Podcast Outro and Guest Credits
The host thanks guest Dave Mangot and provides his website (Mangotech.com), book title, and podcast URL (EngineeringApha.fm). The host notes that Dave is easy to find on LinkedIn. The segment concludes with the show's sign-off identifying it as the Private Equity FunCast.
Assets discussed
| Asset | Kind | Stance | Context |
|---|---|---|---|
| AlphabetGOOGL | equity | neutral | Because if I want to go be an engineer, I can go to, you know, Google, Facebook, whatever. They have these incredible engineering organizations that have done all this stuff. |
| Meta PlatformsMETA | equity | neutral | Because if I want to go be an engineer, I can go to, you know, Google, Facebook, whatever. They have these incredible engineering organizations that have done all this stuff. |
| AppleAAPL | equity | neutral | Like Apple, maybe I don't know. But the AI stuff now is really in just like you're saying, it's stress testing product. |
“Stance” is how the speakers talked about the asset, not a recommendation.
Claims worth verifying
51% of private equity firms say their portfolio company engineering teams get more efficient with AI, yet fewer than 6% can show a 10% lift in EBITDA across the portfolio.
Unidentified speaker (speaker_0) ·0:00
How to verify: OneGuide's State of Portfolio Ops Report 2026; survey data from private equity firms. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Poor data infrastructure is the top blocker to AI success, ranking above talent and models.
Unidentified speaker (speaker_0) ·7:52
How to verify: OneGuide's State of Portfolio Ops Report 2026; survey data from private equity firms. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Meta reported that commits were up 220%, releases up 36%, incidents up 40%, and firefighting up 70% in some cases.
Unidentified speaker (speaker_0) ·20:13
How to verify: Meta internal engineering metrics or public blog posts from Meta Engineering regarding AI impact on development velocity. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
CircleCI data shows similar trends to Meta regarding AI's impact on software delivery.
Unidentified speaker (speaker_1) ·22:25
How to verify: CircleCI State of DevOps Report or public blog posts from CircleCI. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Dave Mangot is the founder and CEO of Mangoteque.
Unidentified speaker (UNKNOWN) ·1:51
How to verify: Mangoteque company website or LinkedIn profile of Dave Mangot. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Dave Mangot is the author of DevOps Patterns for Private Equity.
Unidentified speaker (UNKNOWN) ·1:51
How to verify: Book listing on Amazon or publisher website. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Dave Mangot co-hosts the Engineering Alpha in Private Equity podcast with Paul Karner.
Unidentified speaker (UNKNOWN) ·1:51
How to verify: Podcast listing on Apple Podcasts or Spotify. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
AI systems were invented within the same software delivery life cycles used today.
Unidentified speaker (UNKNOWN) ·12:42
How to verify: Historical analysis of AI model training and deployment pipelines. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Resume-driven development is a common issue where developers choose technologies for personal gain rather than system stability.
Unidentified speaker (speaker_1) ·26:03
How to verify: Industry literature on software engineering best practices and developer behavior studies. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
51% of private equity firms say AI is making their portfolio engineering teams more efficient, yet fewer than 6% can show even a 10% EBITDA lift.
Unidentified speaker (speaker_1) ·28:27
How to verify: OneGuide's State of Portfolio Ops Report 2026; verify via OneGuide's public website or direct contact with the firm. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
A Linux kernel vulnerability and PHP security issues were released in the two weeks prior to the recording (late October/early November 2026).
Unidentified speaker (speaker_1) ·28:47
How to verify: CVE databases (e.g., MITRE CVE, NVD) for Linux kernel and PHP vulnerabilities disclosed between October 1, 2026, and November 18, 2026. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
OpenAI published a
Unidentified speaker (UNKNOWN) ·29:48
How to verify: OpenAI's official blog or documentation for For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
A portfolio company uses a PCI DSS agent running on a cheap, open-weight model to monitor code 24/7.
Unidentified speaker (speaker_1) ·31:34
How to verify: Company engineering blog posts, case studies, or direct inquiry to the portfolio company; verify model costs via open-weight model pricing sheets. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Primary sourceAI title roles on Portfolio Ops teams went from 5% to 25% a year.
Unidentified speaker (UNKNOWN) ·35:31
How to verify: Internal HR data of the PE firm or OneGuide's State of Portfolio Ops Report 2026. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
A portfolio company cut COGS by 70% by building its own custom storage engine.
Unidentified speaker (speaker_0) ·39:39
How to verify: Company financial filings (if public) or case studies published by the PE firm or portfolio company. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Primary sourceAnother company replaced a six-figure CRM with a $125/month build.
Unidentified speaker (UNKNOWN) ·39:40
How to verify: Company financial filings or case studies published by the PE firm or portfolio company. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Primary sourceThe CISO of Anthropic advised CISOs to ask boards for money for
Unidentified speaker (speaker_1) ·48:27
How to verify: Twitter/X post by the Anthropic CISO (verified account) around late October/early November 2026. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Primary sourceDORA research shows that releasing more often is safer.
Unidentified speaker (speaker_1) ·55:20
How to verify: DORA (DevOps Research and Assessment) annual reports or State of DevOps Report publications. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
9 out of 10 companies in regulated industries claim customers will not accept more frequent releases.
Unidentified speaker (UNKNOWN) ·53:13
How to verify: Anecdotal claim by Speaker_1; verify via industry surveys or OneGuide's State of Portfolio Ops Report 2026. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Do your own research
- Verify the 51% efficiency and 6% EBITDA lift statistics against the source cited by Dave Mangot or industry surveys.
- Review portfolio company data infrastructure maturity reports to assess data cleanliness and accessibility.
- Analyze engineering metrics (commit volume vs. release frequency vs. incident rate) for AI-adopting companies.
- Check for 'complexity debt' by comparing debugging time savings against AI generation speed gains.
- Evaluate whether 'AI hiring' in portfolio ops represents genuine expertise or fundraising optics.
- Assess security patching protocols for critical vulnerabilities like Linux kernel and PHP in portfolio companies.
- Review deployment frequency and feature flagging adoption rates as indicators of engineering maturity.
- Investigate build-vs-buy decisions to ensure custom AI solutions are tied to core business differentiation.
Full transcriptexpand
- 0:00
- Unidentified speaker (speaker_0): Fifty one percent of private equity firms say that their portfolio company engineering teams get more efficient with AI, yet fewer than six percent can show you even a ten percent lift in EBITDA across the portfolio.
- 0:12
- Unidentified speaker (speaker_1): If your data infrastructure has bottlenecks in it, AI is going to expose those bottlenecks pretty severely. I mean,
- 0:20
- Unidentified speaker (speaker_0): I haven't sort of seen any front page disaster where the receptionist was rewriting the accounting system.
- 0:24
- Unidentified speaker (speaker_1): Claude wrote the thing. So like is Claude's understanding of how all these complex distributed systems fit into each other a perfect model? Or is it imperfect? Is that the reason that we had the problem to begin with? Like, there's a lot of complexity that just shows up there.
- 0:42
- Unidentified speaker (speaker_0): AI should stick to frameworks and rules better than most people, because I've had the the hot shot developers who love to do something slightly different, whatever, right?
- 0:50
- Unidentified speaker (speaker_1): Resume driven development. Yeah,
- 0:52
- Unidentified speaker (speaker_0): yeah, resume driven. That's good.
- 0:54
- Unidentified speaker (speaker_1): But the system that they're doing it in can't absorb what they're doing, and then we're not seeing the results.
- 1:00
- Unidentified speaker (speaker_0): I I think we are leaking in the factory here, right? What's coming in one end of port engineering isn't coming out the other, which means we've got leaking somewhere, right?
- 1:08
- Unidentified speaker (speaker_2): It's not venture capital,
- 1:11
- Unidentified speaker (speaker_2): it's private equity, it's the private equity.
- 1:15
- Unidentified speaker (speaker_2): fund cast. Or yourself a drink and have a seat.
- 1:19
- Unidentified speaker (speaker_0): Here's a number that should bother every operating partner listening. 51% of private equity firms say that their portfolio company engineering teams get more efficient with AI. Yet fewer than 6% percent can show you even a 10% ten percent lift in EBITDA across the portfolio. So either half of us are wrong
- 1:39
- Unidentified speaker (speaker_0): or the money is going somewhere between the code and the profit and loss. Dave Mango has spent the last year figuring out which, and he's our guest today on the private equity fundcast. Dave, welcome back for the third time
- 1:51
- Unidentified speaker (UNKNOWN): to
- 1:51
- Unidentified speaker (speaker_1): the
- 1:51
- Unidentified speaker (UNKNOWN): private
- 1:51
- Unidentified speaker (speaker_0): equity fundcast. It's nice to see you. Yeah,
- 1:53
- Unidentified speaker (speaker_1): it's good to
- 1:54
- Unidentified speaker (UNKNOWN): be here.
- 1:55
- Unidentified speaker (speaker_1): And this time I don't have to, you know, get on a plane or anything. So it's
- 1:59
- Unidentified speaker (UNKNOWN): that was
- 2:00
- Unidentified speaker (speaker_0): good. All right, well so so last time you were with us. uh the book had just come out. So what's changed in your day job since then? Cause I don't know, there's been a few things in the technology business that may have changed in the last year or two.
- 2:11
- Unidentified speaker (speaker_1): Yeah, I think it's weird. There's like some stuff out there that people are really excited about and they really want to take advantage of it and
- 2:21
- Unidentified speaker (speaker_1): somewhat successful, I guess. Uh I think you
- 2:25
- Unidentified speaker (UNKNOWN): must be talking about AI and how you
- 2:26
- Unidentified speaker (speaker_1): look at it.
- 2:27
- Unidentified speaker (UNKNOWN): You
- 2:28
- Unidentified speaker (speaker_0): must be talking about AI. Uh I'm a I'm a mystic here, I've guessed that.
- 2:31
- Unidentified speaker (speaker_1): Yes, that was uh very astute.
- 2:35
- Unidentified speaker (speaker_1): But uh yeah, I think the interesting thing about all those discussions is uh there's we're for those of us like you and I who've been doing this for a while, we're accustomed to the various stages of the hype cycle or the J curve or whatever you want to call it. and like it's playing out rather predictably.
- 2:55
- Unidentified speaker (speaker_0): Shocking. I mean, when you think about it, we've been through a lot of these trends before, right? You know, mi uh mainframe to mini, MIDI to client server, client server to you know, browser, browser to mobile, like you know. Uh and we all had that. Uh well, I didn't, but we had that uh offshoot into cryptocurrency, which we we all should just forget about
- 3:12
- Unidentified speaker (UNKNOWN): here for the moment.
- 3:13
- Unidentified speaker (speaker_0): So so but but in in the the middle of all
- 3:15
- Unidentified speaker (UNKNOWN): cloud was in
- 3:15
- Unidentified speaker (speaker_1): there somewhere. Yes,
- 3:16
- Unidentified speaker (UNKNOWN): cloud
- 3:17
- Unidentified speaker (speaker_0): was in there somewhere. Yes, you're right. So I think it was browser then cloud or some people would say cloud browser. So but in the middle of all this, you started a podcast about PE engineering. So what what made you decide, hey, yeah, now's the time to jump behind the microphone because it's a great show and I think it's it fills the real gap. But what you know Now what was the what was the uh impetus for you to jump into this uh podcasting game
- 3:37
- Unidentified speaker (speaker_0): at this point?
- 3:38
- Unidentified speaker (UNKNOWN): Yeah,
- 3:38
- Unidentified speaker (speaker_1): I think you know, like this is obviously a major transformation. And uh one thing we didn't want to do was be another AI podcast, because I think there's no shortage of those out there. Um but it's really
- 3:56
- Unidentified speaker (UNKNOWN): more
- 3:56
- Unidentified speaker (speaker_0): AI podcasts or leadership podcasts at this point because I think it's neck and neck at this point.
- 4:00
- Unidentified speaker (speaker_1): Oh, really?
- 4:01
- Unidentified speaker (speaker_0): Yeah, oh
- 4:02
- Unidentified speaker (speaker_1): I think the leadership ones are I I don't know if there's any more or less than there used to be, but um
- 4:09
- Unidentified speaker (speaker_1): but there's been so much conversation in private equity, you know, as someone who's gone through stages of of posting prolifer pro prolifically on LinkedIn about oh well we're moving past the stage of financial engineering. Now you're gonna actually have to do real improvements
- 4:29
- Unidentified speaker (speaker_1): inside these companies instead of holding on to these companies for a while and then letting the market do the work. And and this kind of conversation has been going on for years.
- 4:40
- Unidentified speaker (speaker_0): Oh, I should trust me, I know.
- 4:42
- Unidentified speaker (speaker_1): And only in the past, like I feel like a year or two, like now that the economy isn't just buoying everybody. and bringing them along. Do I feel like with that and AI coming on? Like, I think people are actually interested in making a lot of these improvements. So we wanted to create a
- 5:02
- Unidentified speaker (speaker_1): podcast that was about these real operational changes that you can make to actually get change, Chain you know, change, uplift, whatever you want to call it. And that's alpha, because it's these choices that. you're making. So we called it engineering alpha in private equity because as someone who's a longtime
- 5:22
- Unidentified speaker (speaker_1): technologist like yourself, like I actually really do believe that there's things that we can do inside companies and there's choices that we can make that will actually move the needle. And so I felt like, you know, with Paul, we wanted to kind of get that information out there.
- 5:38
- Unidentified speaker (UNKNOWN): So
- 5:38
- Unidentified speaker (speaker_0): so one uh so key to this episode. So I'm just pulling from one of your episodes, yeah, in the one guide state of portfolio ops report that just landed,
- 5:46
- Unidentified speaker (UNKNOWN): you
- 5:46
- Unidentified speaker (speaker_0): and Paul broke it down. Um and the finding that jumped out to me, because obviously I'm in the business: 51% of the firms report real efficiency gains in in portfolio company engineering, and yet under six percent have captured even a 10% ten percent EBIT.lift. EBITDA left. So w what's happening between the spaces and those numbers? Like that's a that's a huge gap, right? That we're seeing it
- 6:06
- Unidentified speaker (speaker_0): we're seeing it on one side, but it's not coming out the other end of the sausage factory, basically.
- 6:10
- Unidentified speaker (speaker_1): Right. I think uh that was a great episode. Uh thanks to Dante and uh Kate for putting that together. I think like we talked about like a couple seconds ago about the the hype cyc cycle what they call it, the Gartner hype cycle. Yeah.
- 6:25
- Unidentified speaker (UNKNOWN): There's the like
- 6:26
- Unidentified speaker (speaker_0): I call it the valley of anguish, but yes.
- 6:29
- Unidentified speaker (speaker_1): Right. There's the peak of inflated expectations, and then there's the trough of disillusionment, and then there's whatever the steady state is after that. And I think
- 6:39
- Unidentified speaker (speaker_1): this finding of the 51% claim efficiency, but less than you know, 6% six percent or whatever it is, like can show it in the EBITDA. I think we're in that. that that that part right there where we had the peak of inflated expectations and now there's this sort of trough of disillusionment because
- 6:59
- Unidentified speaker (speaker_1): people don't really know how to leverage AI yet. All the patterns for how to do this isn't there, but that doesn't stop people from rushing headlong into it. Like
- 7:09
- Unidentified speaker (UNKNOWN): Yeah let's go,
- 7:10
- Unidentified speaker (speaker_1): let's do AI, let's go, let's do it. And then they all go do it. But the system that they're doing it in can't absorb what they're doing, and then we're not seeing the results. And so, like the way I like to talk about it is the difference between efficiency and effectiveness. Like, we're getting
- 7:27
- Unidentified speaker (UNKNOWN): a lot of things. Oh, that's an interesting way to say
- 7:28
- Unidentified speaker (speaker_0): it.
- 7:29
- Unidentified speaker (speaker_1): But the effectiveness doesn't seem to be there in the numbers.
- 7:33
- Unidentified speaker (speaker_0): So and you believe I because I believe this too. You believe this is not a measurement problem because sometimes PE people think it's a measurement problem. I I think we are leaking in the factory here, right? What's coming in one end of port engineering isn't coming out the other, which means we've got leaking somewhere, right? Yeah. Uh and so one of the keys also in that report, I just want to do a quick translation because sometimes we get
- 7:52
- Unidentified speaker (speaker_0): semi-technical people that listen, like PE people. So a quick translation, you know, when we talk about data infrastructure, what we're gonna talk about is is the pipes, where your data lives, whether the data's clean, whether a system can get at it without a human exporting a spreadsheet first, which is one of my favorite lines, because that's so true. You can kill this world, Dave, just by eliminating Excel. You probably
- 8:12
- Unidentified speaker (speaker_0): cripple most you know financial firms in this world, right?
- 8:15
- Unidentified speaker (UNKNOWN): So
- 8:15
- Unidentified speaker (speaker_0): so with so with that in mind, right? That same report uh that you you just you and Paul discussed put poor data infrastructure at the top of the blocker list, above talent, above models. So you y you're you're you're right in this world. Why does this come first? You are you seeing this with everyone you talk to? And why is this why do you think this is the front bumper?
- 8:33
- Unidentified speaker (speaker_1): Because it's a system and you know, especially the uh you know, I guess it depends on what you're using AI for. Uh but in a system, yeah, anyway, yeah, I guess hopefully everybody who has their MBA was forced to read the goal by Goldrat. I I hope. It
- 8:51
- Unidentified speaker (speaker_0): probably depends on that. I mean I didn't I didn't get an MBA because you know I uh I I so I'll save the uh I got a master's in computer science because I wanted a real job.
- 9:01
- Unidentified speaker (UNKNOWN): Well I doubt they
- 9:01
- Unidentified speaker (speaker_1): made you read a your the goal for that but you know the whole idea and the goal is like where's the bottleneck in the system? And you know in the goal it's all about this factory and you know we have to paint the door doors or whatever they had to do. And they, you know, wound up uh bringing old equipment out of
- 9:21
- Unidentified speaker (speaker_1): storage in order to alleviate the bottleneck and like that increase throughput through the system. And if your data infrastructure has bottlenecks in it, AI is going to expose those bottlenecks pretty severely. And maybe in the past, like if you were doing AI. coding, your bottleneck was how fast
- 9:41
- Unidentified speaker (speaker_1): we could write code.
- 9:43
- Unidentified speaker (UNKNOWN): But
- 9:43
- Unidentified speaker (speaker_1): if our bottleneck isn't how fast we can write code anymore, and we start putting, let's say, 10 times as much code through the system, it doesn't mean the rest of the system is gonna be like, oh yeah, we were totally built for you know something 10 ten times the throughput that we were doing before. And you know, from a private equity
- 10:03
- Unidentified speaker (speaker_1): EBITDA perspective, like that probably would have been pretty stupid if you had built all of your infrastructure to handle ten times as much, you know, information flowing through it as it needed. Because why did I spend all that money to handle ten times something I'm not even doing? So, you know, in some respects
- 10:23
- Unidentified speaker (speaker_1): it's it was a great idea. In some respects, okay, but now we have to fix those things.
- 10:29
- Unidentified speaker (speaker_0): I I guess that the so that's interesting because the fix in this case, in my opinion, right? Yeah. AI is moving so fast
- 10:37
- Unidentified speaker (speaker_0): that uh the fix,
- 10:39
- Unidentified speaker (UNKNOWN): you know,
- 10:40
- Unidentified speaker (speaker_0): the the pressure to get that fix in is huge. And it just can't be fixed that quickly. I mean, it's not just as simple of pointing AI at your bad plumbing, because in a lot of cases, you
- 10:50
- Unidentified speaker (UNKNOWN): know. I've owned enough
- 10:51
- Unidentified speaker (speaker_0): I've owned enough old houses, Dave, where you know, I have the guy going, Well, you know, you've got the you know, transverse copper pipe under I don't understand what he's talking about, but I know it's bad, right? Like it's not something that'd be be be hit quickly, right? Yeah. And is you know, so uh you're right in the sense that PE guys like me don't want you spending money on something that you didn't need. You're gonna have ten 10 users and you built that infrastructure for a million.
- 11:11
- Unidentified speaker (speaker_0): Totally get we never but now I think the challenge is I don't know that anybody saw the speed with which AI was was gonna hit this back plumbing because it's very capable in a lot of things that that humans aren't that capable of or didn't want to do. And I think it's exposing it. So I guess are you seeing that for real in the field? Like when
- 11:31
- Unidentified speaker (speaker_0): you
- 11:32
- Unidentified speaker (speaker_1): Oh absolutely I mean I was talking to uh head of portfolio CTOs at a major European firm yesterday, and he was like, I see portfolio companies that take broken stuff and then try to put it put it inside their AI factory, and guess what? It's just as broken as it
- 11:52
- Unidentified speaker (speaker_1): was. I I think the the thing that I like to caution people against is getting too worked up about how we have to fix these things because
- 12:03
- Unidentified speaker (speaker_1): all of the AI systems that you know we're using to generate all this code, they weren't generated in a vacuum. They weren't invented like on another planet with non Newtonian physics and all like they were invented here on this planet. I I took my daughter when she was
- 12:23
- Unidentified speaker (speaker_1): like, you know, this is many years ago, like five days old to the C. supermarket and my mom was freaking out. She said, How could you take her there? There's germs there. I was like, I'm pretty sure she was designed to live on this planet. So I I'm feeling pretty good about it. And I think people forget that with all this AI stuff. Oh my favorite
- 12:42
- Unidentified speaker (UNKNOWN): word is
- 12:42
- Unidentified speaker (speaker_1): reimagined. We have to reimagine everything. It all has to be reimagined. Like, doesn't really have to be reimagined. These AI systems were built inside the same software delivery life cycles that we use now. It's just now this the coding part isn't as slow. And all these companies
- 13:02
- Unidentified speaker (speaker_1): that are out there that are still doing I hope this isn't too technical for like your audience, but like manual QA. Like we write all the code and then we have a bunch of manual testers run the performance test, regression test, security test, the the all this stuff. Like, okay, in the past,
- 13:22
- Unidentified speaker (speaker_1): if it took a long time to write code, maybe we could get away with that. That doesn't mean that people like me didn't come into your company and say, hey, you know what you need to do? Get rid of that right now. But like people, to your point, kind of didn't want to do it.
- 13:37
- Unidentified speaker (UNKNOWN): And
- 13:38
- Unidentified speaker (speaker_1): now either you do it or you get six percent of companies being able to show ten percent EBITDA lift. Like that those are your kind of your two choices. And so a lot of what people are talking about reimagining is simply the regular good old engineering practices that
- 13:58
- Unidentified speaker (speaker_1): people like you and me have been advising people to do forever because that's what these tools were designed to plug into. They were designed to plug into very good engineering practices.
- 14:10
- Unidentified speaker (UNKNOWN): I
- 14:10
- Unidentified speaker (speaker_1): don't have them. Wesh.
- 14:12
- Unidentified speaker (speaker_0): So that's so that's that's interesting because um there is a disconnect, I think, a lot of times between particularly um
- 14:19
- Unidentified speaker (speaker_0): firms that buy lots of P PE firms that buy lots of different kinds of companies, not just software companies, or are buying companies that might be big manufacturers that have a lot of thought. software custom built in that organization, right? Yeah. That they I that's a level of detail that's hard because I don't think the the DevOps infrastructure QA pipeline has always been b
- 14:40
- Unidentified speaker (speaker_0): had the focus that it I think people sort of shrugged. I mean I I
- 14:43
- Unidentified speaker (UNKNOWN): use I that's a good way of
- 14:45
- Unidentified speaker (speaker_1): saying it.
- 14:45
- Unidentified speaker (speaker_0): I use the standard joke of you know um it's like when you go to the dentist and the dentist your dental hygienist says, Do you floss every day? And you say, uh-huh.
- 14:54
- Unidentified speaker (speaker_1): But you
- 14:55
- Unidentified speaker (UNKNOWN): don't.
- 14:55
- Unidentified speaker (speaker_0): You know you don't. She knows you don't, right?
- 14:57
- Unidentified speaker (UNKNOWN): So
- 14:58
- Unidentified speaker (speaker_0): that's kind of the hey, what's your quote? What's your code coverage? 90%. You just wait. Really?
- 15:03
- Unidentified speaker (UNKNOWN): Okay,
- 15:03
- Unidentified speaker (speaker_0): it's more like 85. And you get down somewhere on 30, and it's like starts to be believable, right? So yeah.
- 15:08
- Unidentified speaker (UNKNOWN): So
- 15:08
- Unidentified speaker (speaker_0): for people who don't understand,
- 15:10
- Unidentified speaker (UNKNOWN): kind
- 15:10
- Unidentified speaker (speaker_0): of walk the audience through what actually happens inside a typical mid-sized. engineering organization when code generation gets cheap overnight. What have you seen from like the C-suite pressure, from what happens in the engineering organization? Like, how has this manifested itself live in terms of, you know, oh my God, what do we do now?
- 15:29
- Unidentified speaker (speaker_1): Yeah, I think there's a I mean, I I don't think that people are upset. I think people are super excited. Like I that's what I'm
- 15:37
- Unidentified speaker (UNKNOWN): seeing, and
- 15:37
- Unidentified speaker (speaker_1): that's what I'm hearing. Like, there's a lot of excitement. I think the problem is like kind of mixed in a little bit with this. We're gonna reimagine everything. Everybody wants to be the person who reimagined it and like got the big wins. And like the thing that I'm struggling with the most right now when I talk to
- 15:56
- Unidentified speaker (speaker_1): CTOs is this idea of we're going to democratize development.
- 16:03
- Unidentified speaker (speaker_1): What? Okay. I don't
- 16:04
- Unidentified speaker (UNKNOWN): even know
- 16:04
- Unidentified speaker (speaker_0): what that means.
- 16:05
- Unidentified speaker (speaker_1): Anyone in the company can come in on any day, sit down with some AI agent, talk about the thing that they want in the product, and in 20 minutes, the agents will have
- 16:17
- Unidentified speaker (UNKNOWN): written a whole new implementation
- 16:19
- Unidentified speaker (speaker_1): and it'll be shipped out to production. And we don't want that just to be the engineers because that would be terrible. We want innovation. anywhere. So, like, if the receptionist decides that our accounting software needs some kind of special thing, they he should be able to come in, sit down with an AI agent, and 20 minutes later, that
- 16:39
- Unidentified speaker (speaker_1): thing that the receptionist thought of should be in production. I'm like,
- 16:45
- Unidentified speaker (speaker_1): I hear what you want to do, but I'm not really following why that's a good idea.
- 16:52
- Unidentified speaker (speaker_0): Right. Right.
- 16:54
- Unidentified speaker (UNKNOWN): Right.
- 16:54
- Unidentified speaker (speaker_0): I mean, I so look, I've seen what we've seen success is we have absolutely seen success where I never get the moving left or moving right thing, but let's call it moving right that product management is beginning to move both right and left a little bit, be able to show something with Is fidelity that a customer can react to. Yeah. Closer to just
- 17:14
- Unidentified speaker (speaker_0): you. screen Figma screenshots on one end and communicate more effectively with engineering in terms of what the thing should do. I'm
- 17:23
- Unidentified speaker (UNKNOWN): still
- 17:23
- Unidentified speaker (speaker_0): seeing the gap, however, underneath that. And what does that mean in terms of our development process? Not reimagining, but how often are we releasing features? How are features vetted? Yeah, what are we getting from product management in terms of the performance? numbers or scalability things that we need for these features? Because they may not know, right? This becomes one of
- 17:43
- Unidentified speaker (speaker_0): those customers are now seeing fidelity, you know, high fidelity prototypes that
- 17:48
- Unidentified speaker (UNKNOWN): they
- 17:48
- Unidentified speaker (speaker_0): think can be product quicker, right? That's
- 17:50
- Unidentified speaker (UNKNOWN): my
- 17:50
- Unidentified speaker (speaker_0): carousel marrow for you from the prices, right? Um but I just I just don't see that, you know, Yo, that's not gonna happen. hap Like that's not gonna happen, right? So, you know, I see that as a potential disaster. I don't know if you've seen any, I mean, I haven't sort sort of seen any front page disaster where the receptionist was rewriting the accounting system, but I see a lot of that's not too far off.
- 18:10
- Unidentified speaker (speaker_0): And I I guess you're seeing the same thing when you talk to clients and customers.
- 18:15
- Unidentified speaker (speaker_1): Yeah, I mean, I think for for those of us who have been through this a bunch of times, like I think it's kind of interesting how we're stress testing the system and in in the same way that we've talked about. how private equity needs to move on from financial engineering and do whatever, I've been talking to product people for
- 18:34
- Unidentified speaker (speaker_1): years about where do you go to learn product?
- 18:38
- Unidentified speaker (UNKNOWN): Right.
- 18:39
- Unidentified speaker (speaker_1): Because like if I want to go learn how to be a good product manager, where are those companies? Because if I want to go be an engineer, I can go to, you know, Google, Facebook, whatever. They have these incredible engineering organizations that have done all this stuff. Like I could learn some pretty advanced
- 18:58
- Unidentified speaker (speaker_1): engineering techniques, but where do you go to learn product? Like Apple, maybe I don't know. But the AI stuff now is really in just like you're saying, it's stress testing product. Yes. Because what's product there for? Like, is product there? to build prototypes? Like, I don't really want my product
- 19:18
- Unidentified speaker (speaker_1): people spending a lot of time on prototypes. And that's probably what AI is good for. But what I do want them to do is to understand what my customers want. Because you know, talking about your 10% EBITDA lift, like, that's what I feel like the growth there is what's and that's what I want people to be really good at. And I
- 19:38
- Unidentified speaker (speaker_1): think that's one of the things that people aren't even talking about that much right now is how much stress this is putting on the product organizations because engineering is starting to get to the point where they're like, Okay, we can do this faster, we can do this faster. You know, you've said in product for all these years, oh, we have all these great
- 19:58
- Unidentified speaker (speaker_1): ideas. If only you would hire more engineers,
- 20:01
- Unidentified speaker (UNKNOWN): if only
- 20:02
- Unidentified speaker (speaker_1): you would put in the budget for more engineers, we could do all this stuff. And now it's all starting to come together. And so it's kind of fascinating to watch.
- 20:13
- Unidentified speaker (speaker_0): It is. But I I you know, quoting from the I'll quote from the meta numbers, which you also quoted, you know, commits up 220%, releases are up only 36%, thirty-six percent, yet incidents are up 40%, forty percent, and firefighting is up seven. seventy percent in some cases, right? This is the I think this hits exactly what you're in some ways talking about
- 20:33
- Unidentified speaker (speaker_0): that
- 20:34
- Unidentified speaker (UNKNOWN): I
- 20:34
- Unidentified speaker (speaker_0): think the initial pass has been pro product management just generating code or generating more high fidelity prototypes, not necessarily in all cases doing what you're suggesting, which in turn ends up being more commits, right, that are flowing into the system. It's a little bit of I always quote Little's law, the the mayor. manufacturing law that, you know, that having more whip,
- 20:54
- Unidentified speaker (speaker_0): you know, doesn't get more through the factory. In fact,
- 20:57
- Unidentified speaker (UNKNOWN): having
- 20:57
- Unidentified speaker (speaker_0): less whip gets it through the factory.
- 20:59
- Unidentified speaker (speaker_1): Yeah, and I think we it's important to sort of separate some of that stuff. Like number one, um, you know, the idea of a prototype versus Is like something that we actually spend the engineering dollars on. Yep. Right. Because I think I don't for people who don't understand
- 21:19
- Unidentified speaker (speaker_1): how this works inside software companies that well, like you can create a lot of prototypes, and then only certain prototypes are you actually going to spend the actual time
- 21:30
- Unidentified speaker (UNKNOWN): and
- 21:30
- Unidentified speaker (speaker_1): money on is to make them something that you want to be responsible for for years. Yes.
- 21:37
- Unidentified speaker (UNKNOWN): And it's
- 21:37
- Unidentified speaker (speaker_1): it's kind of a little bit like
- 21:39
- Unidentified speaker (speaker_1): uh we used to do hack days, and I still do this with some clients sometimes. Hackathons,
- 21:45
- Unidentified speaker (UNKNOWN): we used to call them. Yep.
- 21:46
- Unidentified speaker (speaker_1): At the end of a hack day, every hack day project does not become a product. Like that's just not how it works. And there's somebody has to decide, hey, this was a really cool project. Let's spend the engineering time and money to make this into something real. And so I I think that's kind of important. And then the second thing is
- 22:06
- Unidentified speaker (speaker_1): I think with like the Facebook numbers with like a lot more outages, then the outages being longer, we're seeing that those results in uh in all the studies. Like it's not just like the Facebook one. So I just don't want anyone to think they can like dismiss it. Like the circle CI numbers show the same thing. The I think it was
- 22:25
- Unidentified speaker (speaker_1): Ferros Feros or something that they all show these same things and I for me and I wrote a piece about this like I think the problem is that engineering is not spending as much time as is required to understand what they're dealing with. And so like the AIs write the stuff the
- 22:45
- Unidentified speaker (speaker_1): engineers kind kinda of half look at it say yeah that looks okay and then it goes out and it breaks. And then if you want to fix something that's broken, you have to understand it. I mean, you were talking about your plumbers giving you like the thing about all the pipes, whatever, like and you're like, Yeah, okay, whatever, but like they understand what they're doing. Right.
- 23:05
- Unidentified speaker (speaker_1): And we don't, and and I think that a little bit of the problem right now is like you kinda kind of have to understand the software that's out in production if you want to fix it quickly. Otherwise you're like, okay, how does this work? And I and I've I've seen this, I've I've lived this. I worked at a social media company
- 23:23
- Unidentified speaker (UNKNOWN): I
- 23:23
- Unidentified speaker (speaker_1): don't know, let's say fifteen years ago, and they had one guy, one engineer who was on call. So it I was woken up because I was in And operations, and yeah, I'd it'd be something where I was like, uh this isn't an infrastructure problem. This is a software problem. So I would wake up the one person who
- 23:43
- Unidentified speaker (speaker_1): was my counterpart, and they would look at the code and be like, I I don't know. I don't work on this part of the code base. I I work on the front end. This is like a financial transaction system. I don't know how any of this stuff works. And they would spend like a bunch of time at three o'clock in the morning trying to figure this stuff out and try to write a patch. But you know,
- 24:03
- Unidentified speaker (speaker_1): eventually we matured past that and we started splitting up the on-call stuff for the software engineers. So people were on call for an area that they were familiar with, that they, you know, worked on the code. And like that's what I feel like is happening inside a lot of these companies like Facebook and and all these other places,
- 24:23
- Unidentified speaker (speaker_1): is like the engineers are waking up at three o'clock in the morning going, I don't know how this stuff works. And like, can they ask Claude, hey Claude, how does this work? They can,
- 24:34
- Unidentified speaker (speaker_0): but
- 24:34
- Unidentified speaker (speaker_1): also Claude wrote the thing. So, like, is Claude's understanding of how all these complex distributed systems fit into each other a perfect model? Or is it imperfect? Is that the reason that we had have the problem to begin with? Like, there's a lot of complexity. that just shows up there at three o'clock in the morning and
- 24:54
- Unidentified speaker (speaker_1): takes a long time to fix things that you don't understand.
- 24:57
- Unidentified speaker (speaker_0): Well, that and I think I I see that coming from too. I mean, you you had said that uh I think uh CTO only brings the the first bullet of you know how much development is up to the board, right? Uh and yeah,
- 25:09
- Unidentified speaker (UNKNOWN): I
- 25:09
- Unidentified speaker (speaker_0): I don't look at that as dishonesty, I look at that as I I often have trouble with the boards, they don't care about the other numbers. When the other numbers are really important, right? That and it's not their fault, it's because what they're competing on in the industry is how much new feature stuff is coming out, right? So they're all talking about the code coming out, right? They're not talking about the other stuff, but you're exactly right that it's
- 25:29
- Unidentified speaker (speaker_0): it's causing more of these problems. And you you're you're basically looking at stuff that you didn't write. And now I had a really good CTO friend of mine. say, you know, who's a I think a phenomenal coder, right? And he's like, hey, just for the record, humans don't write that great code. We all get lazy. Theoretically, AI can write better code, but that doesn't
- 25:49
- Unidentified speaker (speaker_0): mean we're going to get better code out of it. And he and he's right, right. Like I think it, you know, AI should stick to frameworks and rules better than most people, because I've had the the hotshot developers who love to do something slightly different, whatever. Right. But resume.
- 26:03
- Unidentified speaker (speaker_1): driven development. Yeah,
- 26:04
- Unidentified speaker (speaker_0): yeah. Resume driven. That's good. You should you should publish a book on that. Copyright to Dave Mango, 2026. Um, but but I think that that is a challenge in how do, you know, how
- 26:15
- Unidentified speaker (UNKNOWN): do
- 26:16
- Unidentified speaker (speaker_0): we get you know boards and management teams appreciating that the other pieces of this puzzle, because the the very first thing when I looked at this, so first of all, two things. One,
- 26:26
- Unidentified speaker (UNKNOWN): an
- 26:26
- Unidentified speaker (speaker_0): old guy like me could write code again, because
- 26:29
- Unidentified speaker (UNKNOWN): you
- 26:29
- Unidentified speaker (speaker_0): know, I just can't stay close to any language, right? But now I can, right? Yeah. Provided I'm not hopping into something I know nothing about where I can get in real trouble, like Rust. But anybody who's old like me or you, the very first thing was like, Holy shit, this can solve some real QA problems, because that's the thing that we could really do. And I don't think anybody jumped into that as the very first thing. I think most of
- 26:49
- Unidentified speaker (speaker_0): other organizations that we've dealt with. have jumped right into speeding up development or throwing features in. And it sounds like you're seeing the same thing mostly when you talk to folks.
- 26:58
- Unidentified speaker (speaker_1): Yeah. And I think it's a shame because a lot of the things that you're talking about, like we didn't do QA well or we didn't do this, that, or whatever,
- 27:10
- Unidentified speaker (UNKNOWN): is
- 27:10
- Unidentified speaker (speaker_1): it's because nobody wanted to do that. Sucks. It's boring work. It's tedious. It's whatever. And like, why don't we have the AIs do the things that suck? Like, or the things that we couldn't do. Or or all these other that's that's a great use of AI. Like, I, you know, I have no problems with like
- 27:30
- Unidentified speaker (speaker_1): making the AI work all night on something that I wouldn't want to work on myself because it's a computer. It's not like I don't I don't feel
- 27:40
- Unidentified speaker (speaker_1): I don't really care that the LLMs can talk to us and do good mimicry of what a human sounds like. I don't is I don't mind making a computer work all night. And I think this is something that's gonna come out in the next few months, and it's something that I work with clients on. Um one is
- 27:58
- Unidentified speaker (UNKNOWN): one
- 27:59
- Unidentified speaker (speaker_1): of the things I work with the clients on is we need to have more meetings. which most engineers aren't really excited about, but like
- 28:07
- Unidentified speaker (speaker_1): get used to it because we have to understand what we're building, how we build it, how we're gonna build it, what was built, all all these other things. And in order to do that, we actually have to talk to each other and be like really intentional. And the second thing is this whole idea of like things that we didn't like, and I I feel like there's a tremendous opportunity here It's not
- 28:27
- Unidentified speaker (speaker_1): even just things that we didn't like, but the thing I've kind of been thinking about in the past like week or two is a lot of portfolio companies, you know, especially let's let's say in the last two weeks, like depending on when this is recorded, like all these security things that are coming out now. You have We to patch everything, patch the whole internet,
- 28:47
- Unidentified speaker (speaker_1): patch the there's a Linux kernel vulnerability that came out. I think it was yesterday or something like that. Then there was the PHP stuff last week. There's like just mad security things that are going on. And I've worked with so many portfolio companies where the security people are like, well, what we've decided since we can't nobody's
- 29:07
- Unidentified speaker (speaker_1): gonna fund this the amount of security people that we feel like we need, we're gonna come up with these security champions.
- 29:15
- Unidentified speaker (UNKNOWN): I
- 29:15
- Unidentified speaker (speaker_1): I don't know where this came
- 29:16
- Unidentified speaker (UNKNOWN): from. I don't
- 29:17
- Unidentified speaker (speaker_0): know, I don't know what that is.
- 29:18
- Unidentified speaker (speaker_1): They're developers on the these the regular development teams that went through like a little bit more security training because they're interested in it and they're supposed to represent security inside these different teams. And like the whole model is broken, it never it doesn't work. Like I I I see it in places.
- 29:39
- Unidentified speaker (speaker_1): and I'm like, okay, like it's a kind of lip service, whatever. It's fine. I understand what you're trying to do and why you're trying to do it, it just doesn't work. But
- 29:48
- Unidentified speaker (UNKNOWN): if
- 29:48
- Unidentified speaker (speaker_1): you look at the OpenAI um software factory that got published in August, they have a security agent that reviews the code. And I was like, oh my god, that's the security channel. champion. That's that's what security was going for all this time, all these
- 30:08
- Unidentified speaker (speaker_1): years. They were trying to figure out a way of getting some representation on the development team that represented their interests. And I'm like, we can use AI for this. That's awesome. Like that is instead of all this, oh, we're gonna reimagine the things and ever and we're the div d you know. the receptionist is gonna r
- 30:29
- Unidentified speaker (speaker_1): forget all that. Like that's like that's silly. But like we've always wanted to have security representation on every developer team. And now
- 30:38
- Unidentified speaker (UNKNOWN): we can do
- 30:38
- Unidentified speaker (speaker_1): it. And like that's to me a much more interesting use of these agents and AI than simply like, oh, we're going to get make everybody into a product manager now. Like um, okay. That's
- 30:53
- Unidentified speaker (speaker_0): yeah. I mean, to me, that seems like all you know, uh I mean, following up on that, you know, security, whether it's infrastructure, data, like these parallel agents looking for data issues, quality issues, you know, holes. I mean, even even portfolio companies that have older infrastructure. I mean, I was never a PHP guy, it's just such an ugly language. I mean, if you're a Marco Arment Armit fan,
- 31:13
- Unidentified speaker (speaker_0): uh, you know, he's just his his four. 40 page screed on PHP from 20 years ago, even
- 31:17
- Unidentified speaker (UNKNOWN): though he writes a lot of PHP. Still, you
- 31:19
- Unidentified speaker (speaker_0): know, somebody holes in it, right? But you can now run it against your own code internally. Like
- 31:25
- Unidentified speaker (speaker_0): most of these things can be done internally. Uh, you know, it can be done in a really efficient way, which wasn't true even six months ago, in my opinion.
- 31:34
- Unidentified speaker (speaker_1): Yeah, I was amazed. Uh, somebody has in an one of the port codes I was talking to, I think, um, has a PCI DSS agent. And all it does is go around and look at the code base 247 and looks for PCI DSS violations and tries to fix that stuff before somebody else
- 31:54
- Unidentified speaker (speaker_1): has to do it. And I think my favorite thing about that, because now I remember who it was, is they weren't even using like any kind of fancy.
- 32:03
- Unidentified speaker (speaker_1): like you know, super frontier model, super expensive, whatever. Like it was a very cheap, open weight model that they could just run 24 7 because it was cheap and it was effective. And I was like, this is good engineering. Like, this is what But you know, and the the places I've worked where I
- 32:23
- Unidentified speaker (speaker_1): felt like there's really high performing engineering team. teams, that's what these teams do. We don't just like say we've got a million dollars, so let's spend a million dollars. We're like, what's the right tool for the job? How do we apply it? What you know, what's that's most effective, those kinds of things. And I I thought it was just amazing.
- 32:42
- Unidentified speaker (speaker_0): Well, that's so that's really interesting, though. The right tool for the right job is uh is an aside for a second, because and most of our portfolio companies of a certain size. right? They've got a series of programming skills historically, database skills, app server, whatever, that that team knows, right? If the right tool for the next job is a different tooling, they often don't make that choice because they just don't know it.
- 33:03
- Unidentified speaker (speaker_0): So it's like I gotta go get consultants or I gotta learn some new language. And a lot of cases they just do it in the in the platform they've always done it in, which is not always the best way to do it, right? I think AI, now it still doesn't solve the problem here if you don't know. anything about something that AI is generating, you still have a challenge, but there is, you know, a good architect understands what the pieces should be, I guess, right?
- 33:22
- Unidentified speaker (speaker_1): Yeah, I mean, uh uh the technology, I mean, to your point about architect uh, we've said in the DevOps movement forever, architecture matters, technology doesn't. Because you put great technology into a lousy architecture, you're not gonna get a great response. You put okay technology into a great
- 33:42
- Unidentified speaker (speaker_1): architecture, you can do an awful lot with that. Um, and in this case, you know, I'll I use like uh the example from cloud. Like if I need storage on the cloud, you know, sometimes I need storage where I need to be able to access the data in like fractions of a millisecond.
- 34:02
- Unidentified speaker (speaker_1): Sometimes I need storage where like if it takes three seconds, like it's totally fine. The cost difference between those two different types of storage is dramatically different. And so, you know, I think the good engineers will say, like, hey, what's the right tool for the job here? And like, I'm Yeah, not going to
- 34:22
- Unidentified speaker (speaker_1): buy the super sub fraction of a second storage for something that doesn't need it. I'm just shoveling money. Like that's just ruining my margins. Like how how to ruin margins one oh one. And I think that was the beauty of this, like this them using these open weight models for this kinds of things is like, sure, we
- 34:41
- Unidentified speaker (speaker_1): could go see whatever the latest thing that OpenAI has to offer, whatever the most expensive thing is, because it'll probably work.
- 34:50
- Unidentified speaker (UNKNOWN): But
- 34:50
- Unidentified speaker (speaker_1): that's gonna destroy your margins. And we w we want a ten percent EBITDA lift, which is kinda what we've been taught dancing around the whole episode. Like I'm not gonna get that by using the most expensive thing where I don't need that. And I I think that's I think that's uh that's getting back into engineering, which I think
- 35:10
- Unidentified speaker (speaker_1): is you know, that's where we get engineering alpha from. That's yeah, that's I think that stuff is very important.
- 35:18
- Unidentified speaker (speaker_0): But here's you know, so this gets us to the next level. Like I so uh we see this on our side, the portfolio operations org chart at private equity firms
- 35:27
- Unidentified speaker (UNKNOWN): is changing fast.
- 35:28
- Unidentified speaker (speaker_0): Now, of course, that's been happening for the last five or ten years. I
- 35:31
- Unidentified speaker (UNKNOWN): mean,
- 35:31
- Unidentified speaker (speaker_0): I've been I've
- 35:31
- Unidentified speaker (UNKNOWN): been
- 35:31
- Unidentified speaker (speaker_0): in uh portfolio operations since 2002, which is there wasn't nobody did that way back then, right? But what I'm seeing like AI title roles on Portfolio Ops team went from 5% to like 25% a year, right? Um, you know
- 35:48
- Unidentified speaker (speaker_0): in your interactions with some of these folks, are you seeing that as real capability, or is that you know more of a fundraising slide at the moment? Because to my argument, getting down to On the next level, which you just talked about in terms of you know really engineering alpha and understanding how these things work, you have to be slightly more technical because you can't just throw it at Claude. You have to decide does this task work
- 36:08
- Unidentified speaker (speaker_0): really well with this open weight model? What kind of task is it? Is it QA different open weight model for QA versus you know QA on a document, for example? I'm just not seeing the depth yet on these teams. I think it's for me, it's it seems like a fundraising slide more than real uh expertise, I guess, on these on these private equity firms. I don't know if you're seeing that, you know,
- 36:28
- Unidentified speaker (speaker_0): uh with the firms you talk to.
- 36:30
- Unidentified speaker (speaker_1): I think the interesting thing is it's it's not as I would have expected. So
- 36:38
- Unidentified speaker (UNKNOWN): okay.
- 36:38
- Unidentified speaker (speaker_1): I am seeing a lot of these hires and I completely understand the pressure to well But look we have an AI person too and look we're we're you know we're like a uh we also do that which
- 36:51
- Unidentified speaker (UNKNOWN): I
- 36:51
- Unidentified speaker (speaker_1): think in some respects is funny because I don't remember that with cloud. I don't remember people having
- 36:57
- Unidentified speaker (speaker_1): cloud or anything like that. But the thing I've
- 37:01
- Unidentified speaker (UNKNOWN): I
- 37:01
- Unidentified speaker (speaker_1): guess I'm pleasantly surprised maybe is the right thing is the people that I've talked to at a lot of these different firms are actually very knowledgeable and very good at what they do. And like they're doing the best they can with what. they have. And like sometimes they're not they don't have what they want.
- 37:21
- Unidentified speaker (speaker_1): Um but they are knowledgeable and they do recognize like, hey, you know, this company is trying to take some broken thing and make it AI and like we're just gonna get a broken thing with AI. Like and they're honest about it. And you know they're they're they're because and I think this was in
- 37:41
- Unidentified speaker (speaker_1): the um in the report, like I think they tend to be technical and because of that they're kind of grounded and they're kind of like, look, yeah, I'd we'd love to do this, but like this is just basic blocking and tackling that you just have never wanted to do. I s there was um
- 38:01
- Unidentified speaker (speaker_1): there was a great thing that was going around, I think it was late last week, where somebody wrote this long blog post. I I yeah. Anyways, uh they wrote this long blog post about the security people, and the security people uh are in this terrible situation because all of this security stuff is happening.
- 38:20
- Unidentified speaker (speaker_1): And it said one of the lines that really jumped out at me was it said, sure, if you have the ability
- 38:27
- Unidentified speaker (speaker_1): to ingest these patches and then update all your images and then deploy them out globally in a safe manner, then this isn't a big deal. But no company in the past decade has been willing to do that work. And I'm like, Well,
- 38:43
- Unidentified speaker (UNKNOWN): I
- 38:43
- Unidentified speaker (speaker_1): guess the chickens are coming home to roost now.
- 38:48
- Unidentified speaker (speaker_0): I love
- 38:49
- Unidentified speaker (UNKNOWN): I love that
- 38:49
- Unidentified speaker (speaker_0): I'm a city person and know that expression. but yeah, I you know, chickens I didn't see my first tree till I was 20, probably, Dave. So yeah, yeah, it's a good farm metaphor, right?
- 38:57
- Unidentified speaker (speaker_1): Yeah. So and I, you know, I think that these security folks understand that. They un they're not they're not like fresh out of college. They they they've seen these things and they understand like, yes, my my portfolio companies should have invested in that kind of stuff, and they didn't. and now I have to kind of deal with
- 39:17
- Unidentified speaker (UNKNOWN): it.
- 39:18
- Unidentified speaker (UNKNOWN): Yeah,
- 39:19
- Unidentified speaker (speaker_0): and I I I I do think the tooling helps us deal it, which is actually a good thing. I mean, if we were trying to deal with this without AI,
- 39:25
- Unidentified speaker (speaker_1): Oh, yeah.
- 39:26
- Unidentified speaker (UNKNOWN): Oh
- 39:26
- Unidentified speaker (speaker_1): my god.
- 39:27
- Unidentified speaker (UNKNOWN): A lot more work, a lot more dollars.
- 39:29
- Unidentified speaker (speaker_0): It's a great opportunity, which is I think what people miss, even if it's not as sexy, right? So so to that point, um, you had mentioned um uh on buy versus build, right? Because again, we're talking about
- 39:39
- Unidentified speaker (speaker_0): how much do
- 39:40
- Unidentified speaker (UNKNOWN): you buy,
- 39:40
- Unidentified speaker (speaker_0): how much do you build. Uh you worked with a company that cut cogs by like 70% by building their own custom storage engine. Yeah. And then uh another company replacing a six-figure CRM with 125 bucks a month build. What what separated those custom builds from stuff that just becomes a millstone? And some of that is the organization, some of that is the technology. Obviously, AI makes
- 40:00
- Unidentified speaker (speaker_0): this easier, but those are pretty startling numbers. for somebody to get from an alpha standpoint.
- 40:05
- Unidentified speaker (speaker_1): Yeah, and um, and they were earned, you know, especially uh
- 40:11
- Unidentified speaker (speaker_1): especially the storage one. But I think the the thing that, you know, and I I've worked at SaaS companies for a long time. I was at Salesforce, people might consider that a SaaS company. Um, but
- 40:25
- Unidentified speaker (UNKNOWN): I
- 40:26
- Unidentified speaker (speaker_1): think the thing that I always told people was, you know, oh well, you should buy from us because we're a SaaS company, don't build it yourself, blah blah blah, because that's you know what you say. But
- 40:37
- Unidentified speaker (UNKNOWN): I
- 40:38
- Unidentified speaker (speaker_1): think the important thing that I kind of stress to people is like it
- 40:42
- Unidentified speaker (speaker_1): it's not there's nothing wrong with building your own stuff, it just has to be something that's completely core to the way that you do your business.
- 40:51
- Unidentified speaker (speaker_1): And so, like, if I'm going to I don't know if anyone's going to get this reference, I don't want to write my own JIRA because
- 41:00
- Unidentified speaker (speaker_1): I hope everyone
- 41:01
- Unidentified speaker (UNKNOWN): like JIRA is a
- 41:02
- Unidentified speaker (speaker_1): ticket tracking system.
- 41:04
- Unidentified speaker (speaker_0): Legally, you're required in the United States to have at
- 41:06
- Unidentified speaker (UNKNOWN): least
- 41:06
- Unidentified speaker (speaker_0): one Jira license if you're
- 41:08
- Unidentified speaker (UNKNOWN): an
- 41:08
- Unidentified speaker (speaker_0): American citizen.
- 41:09
- Unidentified speaker (UNKNOWN): I
- 41:09
- Unidentified speaker (speaker_0): think that's the a rule of date.
- 41:11
- Unidentified speaker (speaker_1): So apologies to all our European listeners who don't have to have. a Jira license. Um but you know if Jira is the core thing that makes your business run and that's what you're selling to your customers, then yeah, maybe you should rewrite your you rewrite Jira in your own thing and take responsibility for and
- 41:31
- Unidentified speaker (speaker_1): own it. But if it's not, then I kind of think doing the SaaS thing is is more important. But in these cases, like this firm that like their their relationships, like where they go out and find companies to acquire, like their CRM is central to that. That's a huge differentiator for
- 41:51
- Unidentified speaker (speaker_1): them for how they
- 41:52
- Unidentified speaker (UNKNOWN): work.
- 41:53
- Unidentified speaker (speaker_1): And then for this company that was a uh SaaS metrics company, they take their customer data in and they store it, and that's a central completely. component of how they make money. And so, you know, that company, that portfolio company wasn't going and rewriting Jira just because, like, and
- 42:12
- Unidentified speaker (speaker_1): especially nowadays, right? Well, we have agents, so the agents will rewrite it. And like, that's not a thing. And so they were able to do these things and create custom solutions that were tailor-made to their exact specifications.
- 42:27
- Unidentified speaker (speaker_1): because they know those things super well because that's what they live in, like day in and day out. And I can build a custom version. I mean I've built custom versions of all kinds of stuff this year using Cloud
- 42:40
- Unidentified speaker (UNKNOWN): for my
- 42:40
- Unidentified speaker (speaker_1): own personal use. Yeah,
- 42:41
- Unidentified speaker (speaker_0): me too.
- 42:42
- Unidentified speaker (UNKNOWN): Um
- 42:42
- Unidentified speaker (speaker_1): and it stuff works really well for me because I don't not tied to what someone else's vision of the thing could, you know, should be, but that's because I know exactly what I want to do for my specific use case and a lot of the things that I've written. I can't be like, hey Jim, you know, you should use this thing. You'd be like, Well, that's not how I work.
- 43:01
- Unidentified speaker (speaker_0): Right.
- 43:02
- Unidentified speaker (speaker_1): I'd be like, Right.
- 43:03
- Unidentified speaker (UNKNOWN): Yeah, but you should
- 43:03
- Unidentified speaker (speaker_1): just change the way you work to the way that I work. I mean, that's it's kind of crazy. And so it has
- 43:09
- Unidentified speaker (UNKNOWN): to be
- 43:09
- Unidentified speaker (speaker_1): for me like it has to be something that's fundamental to the way that I do business because now. it becomes part of my core value proposition.
- 43:17
- Unidentified speaker (speaker_0): Yes. And this this this Dave was the so in the 80s transition when relational databases started to get useful enough for production. I mean, initially they weren't. They were basically just light data analytics, but as it started to get good enough for real production systems that you know the that was the exact same time as you're
- 43:37
- Unidentified speaker (speaker_0): heading into like 1999. you
- 43:39
- Unidentified speaker (UNKNOWN): know,
- 43:39
- Unidentified speaker (speaker_0): 80s to 1999, that everybody said, Oh, no, no, we need to buy off the shelf systems because we're not going to be ready by 1999. And every off the shelf ERP, CRM vendor in the world said, We have best practices. Modify your business to use our practice, best practices, which I think is basically proven to be not true. To your point, if it's core to your business, you now have the ability, even
- 43:59
- Unidentified speaker (speaker_0): if you weren't necessarily low level engineers, to custom. customize that thing so it fits exactly your business purpose and my argument is I think that's a thing we should consider doing which which nobody would have done even three years ago uh unless they had a a real team under them because it's such a complex uh you know thing to to try and do.
- 44:17
- Unidentified speaker (UNKNOWN): Yeah
- 44:17
- Unidentified speaker (speaker_1): well the tension there is automating bad processes also. Yes,
- 44:23
- Unidentified speaker (UNKNOWN): yes, yes.
- 44:23
- Unidentified speaker (speaker_1): Right. Like I I think one of my favorite things I was listening to one of the AWS uh what do they call them? Um I don't know what are they systems advisors or whatever they're
- 44:35
- Unidentified speaker (UNKNOWN): called. Yep.
- 44:36
- Unidentified speaker (speaker_1): And they were like we
- 44:38
- Unidentified speaker (UNKNOWN): don't have the
- 44:38
- Unidentified speaker (speaker_1): ability to come into your organization and change the way that you do stuff. All we can do is identify everything. Meaning if in order to you know make this transaction I have to go to Jim and say, Jim, do you authorize this transaction? Then they will have an agent contact Jim
- 44:58
- Unidentified speaker (speaker_1): and say, Hey, Jim, do you authorize this transaction? Which is really fun when that's happening 10,000 times a day. And you're like, well, yes, we're automating our business processes, but also
- 45:12
- Unidentified speaker (speaker_0): poor Jim uh does not want them automated. Well, that's when I start my agent. All right. So to that point, Dave. So fictionally, set the stage here for a second, right? So you've had an operating partner is listening to this podcast on a treadmill with a portfolio CTO call at 10, right? And so I know you've been you lean on deming, which I agree with, right? And my quote here,
- 45:32
- Unidentified speaker (speaker_0): which you use, we translate aims to goals, and then we do stupid things in the name of the goal. So to that point, what's the most common stupid thing you're seeing right now?
- 45:41
- Unidentified speaker (speaker_1): Oh, it clearly the most common stupid thing is taking our broken stuff and then trying to put agents on it. And then
- 45:49
- Unidentified speaker (UNKNOWN): now
- 45:49
- Unidentified speaker (speaker_1): I've got it's broken with an agent instead of before. Like, you know, just like it was before. And so, like, simplest example, I've got a bunch of flaky tests,
- 46:00
- Unidentified speaker (UNKNOWN): and
- 46:00
- Unidentified speaker (speaker_1): like they, you know, and then I'm like, oh. well, we're just gonna have an agent write the code, and
- 46:06
- Unidentified speaker (UNKNOWN): we
- 46:06
- Unidentified speaker (speaker_1): get these really inconsistent results. Well, yeah, because your tests are flaky, and like the agent, you know, I for people who don't know in the audience, like when an agent's writing code, it's very much this we used to call it TDD. I guess we still call it TDD, T D
- 46:24
- Unidentified speaker (UNKNOWN): uh,
- 46:24
- Unidentified speaker (speaker_1): test-driven Test Driven development idea. Like I've got a test the way you do with test driven development, I write a test that you know says w that these conditions were met, which is what basically all the agents trying to do is solve for these conditions being met.
- 46:45
- Unidentified speaker (speaker_1): Then it's red. And if the number is one 150, fifty, it's red. If it's negative fourteen, 14, it's red.
- 46:51
- Unidentified speaker (speaker_1): If it's 14, then we're green. And all the agents are trying to do all day long. This is how they work for anyone who doesn't understand this. Trying to make the test go green. Yes. They actually don't even care how they do it, as long as it goes green. That's
- 47:06
- Unidentified speaker (UNKNOWN): what they're
- 47:07
- Unidentified speaker (speaker_1): trying to do.
- 47:08
- Unidentified speaker (speaker_0): It's like my kids doing the trash or dishes. They don't care how they do it, Dave, just because I'm
- 47:12
- Unidentified speaker (UNKNOWN): going gonna to get
- 47:12
- Unidentified speaker (speaker_0): credit rate for getting and it does. done.
- 47:13
- Unidentified speaker (speaker_1): What do you want?
- 47:15
- Unidentified speaker (speaker_1): Like, there's still spaghetti on this. Yeah, I know, but I put it in the dishwasher with the spaghetti, so it's in the dishwasher. Right.
- 47:22
- Unidentified speaker (speaker_0): Well, you didn't say, you know.
- 47:24
- Unidentified speaker (UNKNOWN): Yeah,
- 47:24
- Unidentified speaker (speaker_1): but so um, but you get into these situations with these flaky tests where like sometimes 14 shows up as red, sometimes it shows up as green.
- 47:34
- Unidentified speaker (UNKNOWN): And
- 47:34
- Unidentified speaker (speaker_1): then you want these agents to figure out how to write the code and it's a moving target. And it keeps changing on them. And and so that's by far the thing that I see people not doing is just the basics that are the things like I said earlier, like just they're just good engineering practices.
- 47:54
- Unidentified speaker (speaker_1): And like to your point, we can have the agent go in and fix the flaky test. Like obviously we want to make sure that it didn't just set everything. to true
- 48:03
- Unidentified speaker (UNKNOWN): and
- 48:03
- Unidentified speaker (speaker_1): then
- 48:04
- Unidentified speaker (UNKNOWN): we're done.
- 48:05
- Unidentified speaker (UNKNOWN): But
- 48:05
- Unidentified speaker (speaker_1): you know, that's one way of fixing the flaky test, but that's also our spaghetti example. Isn't
- 48:10
- Unidentified speaker (UNKNOWN): that
- 48:10
- Unidentified speaker (speaker_0): United isn't that United Airlines uh you know uh
- 48:13
- Unidentified speaker (UNKNOWN): Bowen?
- 48:16
- Unidentified speaker (UNKNOWN): But
- 48:16
- Unidentified speaker (speaker_1): yeah, I mean I think that's by far the thing I I see. And there was a great tweet last week from the CISO of Anthropic.
- 48:27
- Unidentified speaker (speaker_1): Go figure. And he basically said something the effect of like, it's gonna get real bad for the next 18 months. And you know, we just have to hold on and and this and that. But like I thought one of the fascinating things he said was like basically like CISOs, go to your board and ask for
- 48:47
- Unidentified speaker (speaker_1): money. And he's like, not for more agents or anything, for the basics. It's like that is the quote, the basics. And I was like, oh my God, it's so refreshing to like hear the CISO of Anthropic. It's like a frontier model lab saying, like, focus on the basics. And that's
- 49:07
- Unidentified speaker (speaker_1): exactly what I'm seeing inside these portfolio companies. It's is like, you release once a week if you're lucky. and you do it in this like weird thing, but you want agents to crank out six thousand changes a week,
- 49:22
- Unidentified speaker (UNKNOWN): you
- 49:22
- Unidentified speaker (speaker_1): don't even have the ability to deploy it. Like fix the basics, like and then you can take advantage of these. So I see that all the time. It's so common.
- 49:31
- Unidentified speaker (speaker_0): So it would you say would you say this then? If a firm can fund exactly one thing across the portfolio next quarter to close that eBay EBITDA diagram. gap, what would it be? Would it be something along the basics? Would they have to analyze the basics? Like what would be your you got one thing you can do to sort of throw some positive uh EBITDA based on all this wonderful AI activity,
- 49:51
- Unidentified speaker (speaker_0): what would it
- 49:52
- Unidentified speaker (UNKNOWN): be?
- 49:53
- Unidentified speaker (speaker_1): I I'm I'm trying to figure out if this is like too too much for my podcast audience and less for yours. Like I th I mean you you and I have talked about the Dora metrics uh way, way back when. Um, I think deployment frequency is a tremendous proxy
- 50:13
- Unidentified speaker (speaker_1): for stuff. And I'm not trying to dismiss the other Dora metrics that are about to
- 50:17
- Unidentified speaker (UNKNOWN): be. Yeah, no, I agree with you.
- 50:18
- Unidentified speaker (speaker_0): But I but I but I agree with
- 50:19
- Unidentified speaker (UNKNOWN): most
- 50:21
- Unidentified speaker (UNKNOWN): of that.
- 50:21
- Unidentified speaker (speaker_1): But deployment frequency to me says like if I can deploy every day, let's just use as our example. That says to me that it's very likely, and it's not a guarantee. I understand all that. It's very likely that I have my processes set up in an automated way that
- 50:41
- Unidentified speaker (speaker_1): allow me to take advantage of AI. That's that's a it's a simple, I get it, it's a simple first pass proxy, but like that means I've done the work to figure things out. It doesn't mean that you know there's not a thing that Jim is like the Simpsons bird who just keeps pressing
- 51:00
- Unidentified speaker (speaker_1): approve. You know, it doesn't mean they don't have that. I mean, I hope they don't, but but just as a simple first pass proxy, the thing that I would be asking if I'm the board, my CTOs, how often do we deploy?
- 51:14
- Unidentified speaker (UNKNOWN): And if
- 51:14
- Unidentified speaker (speaker_1): it's not daily, I think there's that's where you start digging in. And obviously, people on the board probably don't have the ability to dig into that. But like really, like that kind of discipline gets your uh
- 51:30
- Unidentified speaker (speaker_1): gets all of your systems in place to be able to do that. And like if you want to think about it, since you mentioned deming, like in terms of a manufacturing thing, like do I really like if I have a factory and I'm producing brooms, do I wanna produce a broom once every five days? Like, no, I wanna be producing
- 51:49
- Unidentified speaker (speaker_1): brooms every day all the time. And that means the system is just flowing. And if it takes me a week to produce a broom, that means there's probably an awful lot of things going on to get that broom out the door on Friday. And
- 52:03
- Unidentified speaker (UNKNOWN): like
- 52:03
- Unidentified speaker (speaker_1): that's why I think it's just a pretty good proxy.
- 52:06
- Unidentified speaker (speaker_0): So so you know the the um
- 52:09
- Unidentified speaker (UNKNOWN): at
- 52:09
- Unidentified speaker (speaker_0): the smaller size organizations, like typically lower middle market, that is always the it's been two problems, right? That that yes, we have not been able to to deploy every day. And part of that was no feature flagging, right? So you know what
- 52:23
- Unidentified speaker (UNKNOWN): I mean? So it
- 52:23
- Unidentified speaker (speaker_0): gets because I think everybody's initially looked at deploying every day was is oh you just you're just deploying a website that's different we're well past that mentality but I think that's the thing we're seeing pushing our portfolio companies to and that has worked it's like look let's get feature flagging in and let's be sure we can push without problems every day
- 52:39
- Unidentified speaker (UNKNOWN): because that makes
- 52:40
- Unidentified speaker (UNKNOWN): it for people who
- 52:41
- Unidentified speaker (speaker_1): don't who
- 52:41
- Unidentified speaker (UNKNOWN): yes
- 52:42
- Unidentified speaker (speaker_1): understand
- 52:42
- Unidentified speaker (speaker_0): 100%
- 52:43
- Unidentified speaker (UNKNOWN): yeah
- 52:43
- Unidentified speaker (speaker_1): all
- 52:43
- Unidentified speaker (speaker_0): right so oh
- 52:44
- Unidentified speaker (speaker_1): but
- 52:45
- Unidentified speaker (UNKNOWN): well
- 52:45
- Unidentified speaker (speaker_1): you got you got you got me right on this perfect thought that I've I've been enjoying so much for the past whatever is I work with all you know, private equity portfolio companies.
- 52:56
- Unidentified speaker (speaker_0): Yep.
- 52:57
- Unidentified speaker (speaker_1): So many of them are like in regulated industries and this and that and and I'm saying like no we we have to deploy more often, we have to deploy behind feature flags, make it safe. But and I I mean I I'd say nine out of ten companies that I say this to go
- 53:13
- Unidentified speaker (UNKNOWN): our
- 53:14
- Unidentified speaker (speaker_1): customers will won't accept more changes. They our customers won't allow us to release more often. Our customers won't that did it all like I I mean I hear this like is just a r it's just normal, like nine out of ten. But these same companies want to democratize coding so that anybody can
- 53:33
- Unidentified speaker (speaker_1): release anything at any time from anywhere. And I'm like, how do you hold these two ideas in your head simultaneously? It's like
- 53:44
- Unidentified speaker (UNKNOWN): it's insane. But
- 53:45
- Unidentified speaker (speaker_0): you know, it's I think
- 53:46
- Unidentified speaker (UNKNOWN): I
- 53:47
- Unidentified speaker (speaker_0): have seen it. I mean, just you know, to wrap up here on the positive side, Dave, I have seen it. So I we we one of our portfolio companies got a new head of engineering in, you know, nine months ago. And that was this very first thing was that it's not a product that gets released every day. It never would be. It's more like an ERP type product. He's like, we're releasing it every day. And the customers won't take it because every time they did a release, it was a disaster.
- 54:07
- Unidentified speaker (speaker_0): Last board meeting is like, oh yeah, we released like every day for the past month and we've had no calls on it. Yeah.
- 54:12
- Unidentified speaker (UNKNOWN): Like,
- 54:13
- Unidentified speaker (speaker_0): like literally, like if you do it right,
- 54:15
- Unidentified speaker (UNKNOWN): he says,
- 54:15
- Unidentified speaker (speaker_0): now we're prepared if we have to do something quickly, put up new features, we
- 54:18
- Unidentified speaker (UNKNOWN): are ready to
- 54:19
- Unidentified speaker (speaker_0): go. And the customers will, you know, they'd to be able to say, we don't want you to install it. And you're like, we've already installed it, we just have to turn it on. Like, you know, is we way easier than this, you know, this install side of thing. So
- 54:29
- Unidentified speaker (speaker_1): Oh yeah, I would always ask them in response, like big security vulnerability comes out, customer data is exposed. You tell the customer, we're gonna patch this, and they're go, No, no, no, we don't we don't
- 54:41
- Unidentified speaker (UNKNOWN): like to
- 54:41
- Unidentified speaker (speaker_1): take changes. We're not we're you know, we're we're not that kind of industry. Why don't
- 54:47
- Unidentified speaker (UNKNOWN): you patch the
- 54:48
- Unidentified speaker (speaker_1): customer data in a month or maybe in a quarter? We'd be much more comfortable with our having all of our data out on the internet for that period of time. And they're like, Oh no.
- 55:00
- Unidentified speaker (UNKNOWN): No,
- 55:00
- Unidentified speaker (speaker_0): but I think that's that is when we're getting that mindset, people understand it. There, they are seeing the results. Like we did that first in this particular case, ahead of you know AIing the engineering organization to your point, and that that worked out great. That we're we're able to basically you know, turn this thing into a bit of a factory, which has been super helpful as we plan on a go forward basis.
- 55:20
- Unidentified speaker (speaker_1): Yeah, because as we know from the Dora research, releasing more often is safer.
- 55:25
- Unidentified speaker (UNKNOWN): For
- 55:25
- Unidentified speaker (speaker_0): sure. 100%. Yes.
- 55:27
- Unidentified speaker (speaker_1): Go figure.
- 55:27
- Unidentified speaker (speaker_0): Yep.
- 55:28
- Unidentified speaker (speaker_1): Like completely
- 55:29
- Unidentified speaker (UNKNOWN): the opposite of
- 55:30
- Unidentified speaker (speaker_1): what we thought in the early 2000s. Just flip it on its head. And I love, I put that out on LinkedIn a while ago and I was like, you know what? We were wrong.
- 55:41
- Unidentified speaker (speaker_1): Like, I'm
- 55:42
- Unidentified speaker (UNKNOWN): sorry. We were.
- 55:43
- Unidentified speaker (speaker_1): We were wrong. Like, that was wrong. Like,
- 55:46
- Unidentified speaker (speaker_0): right. It's it's uh it's a paraphrasing the line of Animal House. You know, you messed up, you trusted us, you know. Oh well, yeah. That's our mulligan. That's our mulligan, Dave. Right. Dave, always a pleasure. It's always great to see you. Um, for those of you who would like to find Dave, Mangotech.com, uh, the
- 56:02
- Unidentified speaker (UNKNOWN): book.
- 56:03
- Unidentified speaker (speaker_0): Engineering Alpha and Private Equity. Uh
- 56:06
- Unidentified speaker (speaker_0): EngineeringApha.fm, uh, monthly DevOps for Portfolio Companies Newsletter, LinkedIn. Dave's everywhere. He is a great read. I'm
- 56:14
- Unidentified speaker (UNKNOWN): hard to find every
- 56:15
- Unidentified speaker (speaker_0): minute you read it. Not hard to find, which is
- 56:17
- Unidentified speaker (UNKNOWN): good. As
- 56:18
- Unidentified speaker (speaker_0): always, Dave, a pleasure.
- 56:20
- Unidentified speaker (speaker_1): Thanks so much for having me. It's always good talking to you, Jim.
- 56:23
- Unidentified speaker (speaker_0): Ciao for now.
- 56:26
- Unidentified speaker (UNKNOWN): Check
- 56:26
- Unidentified speaker (speaker_2): Technology now. issues, middle market PE back companies. They bicker with each other and they don't take themselves too seriously. It's not venture capital,
- 56:39
- Unidentified speaker (speaker_2): it's private equity. It's the private equity funcast. Pour yourself a drink and callacy.
More from Private Equity FunCast
Parker Gals | There's No One Path to Private Equity (w/ Margaret Poisson, Clearhaven Partners)
Episode discussion: Margaret Poisson discusses non-traditional paths into private equity and the role of advocacy in career advancement.
Speakers’ claims: Margaret Poisson, a partner at Clearhaven Partners, outlines her unconventional entry into private equity, moving from audit at RSM to transaction advisory before pivoting to the buy side via a Booth MBA. She emphasizes the importance of securing live deal experience during internships to compete with traditional banking recruits. The discussion highlights how Clearhaven’s founder-led culture and focus on the lower middle market attracted her away from traditional metrics like AUM. Poisson also challenges the notion that there is only one path to PE, noting the rise of direct analyst programs and the disruption of traditional recruiting cycles. Both speakers stress the critical role of having advocates in closed-door conversations for career progression.
Amend and Pretend: Private Credit's Hidden Defaults (w/ Tuck Hardie, Houlihan Lokey)
Episode discussion: Private credit defaults masked by extensions and PIK as sponsors delay losses amid rate hikes
Speakers’ claims: Tuck Hardie and Speaker_1 argue that private credit default rates are artificially low because creditors and sponsors are using 'amend and pretend' strategies to avoid recognizing losses. They explain that Liability Management Exercises and Payment in Kind (PIK) clauses allow overleveraged borrowers to extend debt maturities while maintaining appearances of solvency. The speakers note that sponsors have incentives to keep equity alive to continue charging management fees, while lenders avoid bankruptcy to protect reputations. However, limited partners are increasingly demanding capital returns, which may begin to break this cycle of deferred distress. The episode concludes that many maturities due in 2028-2030 will likely be extended further, with a potential concentration of refinancing activity after the U.S. presidential election.