Engineering Alpha in Private Equity

Speakers argue that quarterly patching is obsolete and harms valuation due to AI-driven code volume and cyber insurance requirements.

14 min Listen to the original Episode page
Private EquityCybersecuritySoftware EngineeringAI InfrastructureDue DiligenceCyber InsuranceVenture CapitalTech Operations

Summary

The speakers contrast traditional software deployment with modern automated pipelines, arguing that quarterly patching schedules are obsolete. They cite Jason Clinton of Anthropic, who warned of an impending 'vulnerability tsunami' due to insufficient automated patching capabilities. The discussion highlights that AI-generated code volume will outpace infrequent deployment cycles, leading to outages and customer attrition. Speakers note that cyber insurance underwriting has tightened, requiring proof of robust security practices. They suggest that poor patching infrastructure will negatively impact due diligence and exit multiples for private equity firms.

Key takeaways

  1. 1
    right so Jason Clinton who's uh CISO or deputy CISO at Anthropic I guess depending on when you're looking at the stuff uh he put out a tweet uh in August that said the vulnerability tsunami is coming in just a few months and there won't be enough time to patch critical infrastructure.

    Unidentified speaker (speaker_2) · 2:40

  2. 2
    A standard patch process involves a security engineer editing a config file (e.g., changing version 23 to 24), checking it in, and having the system automatically build a new container image and deploy it.

    Unidentified speaker (speaker_2) · 5:33

  3. 3
    The primary fear preventing frequent patching is not the lack of tools, but the fear that a fix will break existing functionality, which automated testing aims to eliminate.

    Unidentified speaker (speaker_2) · 9:06

Episode sections

  1. 0:11 – 2:36

    Modern Software Deployment and AI Context

    Speakers introduce the episode topic, contrasting traditional software engineering with modern automated deployment pipelines involving AI, containerization, and Kubernetes orchestration.

  2. 2:36 – 6:47

    The Vulnerability Tsunami and Industry Readiness

    Speaker_2 cites Jason Clinton of Anthropic regarding an impending 'vulnerability tsunami' and recent Java/PHP vulnerabilities, arguing that the industry lacks basic automated patching capabilities.

  3. 6:47 – 10:43

    The AI Delivery Paradox and Quarterly Patching

    The discussion highlights the inefficiency of quarterly patch schedules, noting that AI-generated code volume will outpace infrequent deployment cycles, leading to outages and customer attrition.

  4. 10:43 – 13:38

    Cyber Insurance and Exit Valuation Implications

    Speakers discuss how cyber insurance has become harder to obtain, requiring proof of robust security practices, and suggest that poor patching infrastructure will negatively impact due diligence and exit multiples.

Assets discussed

AssetKindStanceContext
Anthropicprivate companyneutralJason Clinton who's uh CISO or deputy CISO at Anthropic I guess depending on when you're looking at the stuff uh he put out a tweet uh in August that said the vulnerability tsunami is coming in just a few months and there won't be enough time to patch critical infrastructure.

“Stance” is how the speakers talked about the asset, not a recommendation.

Claims worth verifying

Jason Clinton is the CISO or deputy CISO at Anthropic.

Unidentified speaker (speaker_1) ·2:36

How to verify: Check Anthropic's official leadership page or LinkedIn profile for Jason Clinton's title. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Primary source

Jason Clinton posted a tweet in August warning of a 'vulnerability tsunami' arriving in a few months.

Unidentified speaker (speaker_1) ·2:36

How to verify: Search Twitter/X for posts by Jason Clinton from August 2026 containing the phrase 'vulnerability tsunami'. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

A PHP vulnerability emerged a few weeks prior to the podcast recording.

Unidentified speaker (speaker_2) ·3:00

How to verify: Check CVE databases (e.g., NVD) or security news archives for high-severity PHP vulnerabilities disclosed between late September and early November 2026. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

A Java vulnerability emerged in the week prior to the podcast recording.

Unidentified speaker (speaker_2) ·3:00

How to verify: Check CVE databases or Oracle/Sun security bulletins for high-severity Java vulnerabilities disclosed in the week before November 20, 2026. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Anthropic's CISO advised asking boards for help for 'basics' rather than AI.

Unidentified speaker (speaker_2) ·3:44

How to verify: Review the transcript of Jason Clinton's public statements or blog posts from August 2026 for this specific phrasing. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Primary source

Many portfolio companies operate on a quarterly patch schedule.

Unidentified speaker (UNKNOWN) ·6:52

How to verify: This is an anecdotal claim by the speaker; verify via industry surveys on software maintenance practices (e.g., State of DevOps Report) or private equity portfolio company audits. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

AI agents can write 10,000 lines of code a day.

Unidentified speaker (UNKNOWN) ·7:46

How to verify: This is a hypothetical figure used for illustration; verify typical AI coding assistant output metrics via vendor documentation (e.g., GitHub Copilot, Cursor) or independent studies. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Cyber insurance has become significantly harder to obtain over the years.

Unidentified speaker (speaker_2) ·10:43

How to verify: Review annual reports from major cyber insurance carriers (e.g., Chubb, AIG, Zurich) or industry reports (e.g., Marsh, Willis Towers Watson) on cyber insurance market trends and capacity. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Microsoft is known for 'Patch Tuesday' or 'Patch Wednesday' releases.

Unidentified speaker (UNKNOWN) ·10:01

How to verify: Check Microsoft's official security update documentation or historical release calendars. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Primary source

Do your own research

  • Verify Jason Clinton's role and tenure at Anthropic via LinkedIn or Anthropic's official leadership page.
  • Check Anthropic's official blog or Twitter/X account for the specific 'vulnerability tsunami' tweet referenced.
  • Review CircleCI's public blog or case studies for data linking deployment frequency to outage reduction.
  • Examine recent cyber insurance market reports from brokers like Aon, Willis Towers Watson, or Marsh for underwriting trend data.
  • Audit portfolio company tech stacks for evidence of automated CI/CD pipelines versus manual quarterly patching.
  • Check for any public disclosures or SEC filings from portfolio companies regarding cyber incidents or insurance coverage gaps.
  • Look for counter-arguments from security vendors or industry groups on the feasibility of daily patching for legacy systems.
  • Verify the specific Java/PHP vulnerabilities mentioned by Speaker_2 via CVE databases or vendor security advisories.
Full transcriptexpand

0:11Unidentified speaker (speaker_0): Coming to you through your headphones. This is Engineering Alpha and Private Equity. The news this week with Paul Carner and Dave Mango.

0:20Unidentified speaker (speaker_1): Thanks, Jay. Welcome everyone to another episode of Engineering Alpha and Private Equity. Dave, how are you today?

0:27Unidentified speaker (speaker_2): Oh, I'm all fired up today.

0:30Unidentified speaker (speaker_1): Been been reading the news lately or

0:33Unidentified speaker (UNKNOWN): anything you want to talk about?

0:35Unidentified speaker (speaker_2): And uh I guess like the thing I'd like to explain for people who maybe don't understand how some of this stuff works in modern software companies and engineering organizations is uh especially now with AI. right? Uh I decide I want to build some kind of a feature or I want to fix something. Let's say I want to fix something. So I go and I look at the code base and I fix whatever needs to be fixed. Maybe it's a dependency, maybe it's a bug. I don't know. Something like that. Sure.

1:10Unidentified speaker (UNKNOWN): And

1:10Unidentified speaker (speaker_2): then because I have AI, I check the code in, it runs a bunch of automated tests like security and performance and all these other kinds of things. And if it patches pass passes, thank you, all the tests, then the uh system will build a most likely a container image, which basically just is like a thin operating system with the code laid on top of it. I don't want to get it more technical than that. And then that will go and go into some deployment queue. And then, you know, we're probably running Kubernetes, maybe, which

1:46Unidentified speaker (UNKNOWN): is

1:46Unidentified speaker (speaker_2): just this orchestration system that Google built. And then it goes out, and then you know, the old traffic or the traffic that was going to the old version of this container uh goes away, and the traffic that goes to a new version of this container. uh starts flowing, and then all my customers get like this new version. And that happens a hundred percent automated if we're doing our fancy AI dark factory stuff. Sounds pretty cool, huh?

2:17Unidentified speaker (speaker_1): Yeah, I mean it it sounds like you're

2:20Unidentified speaker (UNKNOWN): kind

2:20Unidentified speaker (speaker_1): of laying the groundwork for walking our listeners through an another sort of episode around AI and you know and how to how to deal with uh with that right

2:29Unidentified speaker (speaker_2): right but this was actually an explanation about cybersecurity.

2:33Unidentified speaker (speaker_1): Oh really

2:34Unidentified speaker (speaker_2): surprise

2:36Unidentified speaker (speaker_1): right

2:37Unidentified speaker (UNKNOWN): cybersecurity okay

2:38Unidentified speaker (speaker_1): now that has been in the news

2:40Unidentified speaker (speaker_2): right so Jason Clinton who's uh CISO or deputy CISO at Anthropic I guess depending on when you're looking at the stuff uh he put out a tweet uh in August that said the vulnerability tsunami is coming in just a few months and there won't be enough time to patch critical infrastructure. Well, that's freaking scary. And like, you know, if you've been reading the news lately, like you're seeing like all this stuff, like PHP vulnerability came out a few weeks ago, Java vulnerability came out this week, and everybody's pulling their heads. hair out. Like, oh my God, all the everybody has to spend all night patching all this stuff. It's a it's a nightmare. It's a disaster. And Clinton followed up from his August tweet and said, as bad as things have gotten in the past four days, this is going to get a lot worse before it gets things are going gonna to get a lot worse. Well, they are going gonna to get a lot worse before they get better. Security leaders never let a crisis go to waste. Please go to your boards and ask for help right now. Not

3:44for AI, for the basics.

3:46Unidentified speaker (UNKNOWN): We

3:46Unidentified speaker (speaker_2): need to hold on for about 18 months.

3:49Unidentified speaker (UNKNOWN): And

3:49Unidentified speaker (speaker_2): like I think the super interesting thing about what he said is especially coming from anthropic,

3:56Unidentified speaker (UNKNOWN): not

3:56Unidentified speaker (speaker_2): for AI, for the basics. And all the things that I sort of just described to you about like

4:02Unidentified speaker (UNKNOWN): I

4:03Unidentified speaker (speaker_2): fix something and then a bunch of automated tests run, and then when those automated tests say that everything's okay, the stuff gets deployed out to the production servers. That's the basics. And it's stuff that we haven't done as an industry. And I don't think you and I need to get into why we haven't done

4:21Unidentified speaker (UNKNOWN): that. It's sort of

4:22Unidentified speaker (speaker_2): irrelevant at this point. But like those are the things that need to happen. So that all this stuff about, oh my gosh, everyone's gonna be up all night and we're gonna have to do all these emergency break fixes and all this other stuff,

4:36Unidentified speaker (UNKNOWN): all

4:37Unidentified speaker (speaker_2): that stuff is supposed to be like a nothing burger. Like, yes, a new Java version comes

4:43Unidentified speaker (UNKNOWN): out.

4:43Unidentified speaker (speaker_2): Okay.

4:44Unidentified speaker (speaker_1): Right.

4:46Unidentified speaker (speaker_2): So what?

4:48Unidentified speaker (speaker_1): And and here's anthropic, literally. telling you that the you know the benefits of AI rest on the foundations of, you know, all these other things that, you know, as you're saying, yes, maybe should have been in place, but need to be in place even more so now, and that they're not automatically in place and you need to go and ask for money for them.

5:08Unidentified speaker (UNKNOWN): Yeah,

5:09Unidentified speaker (speaker_2): I don't even think these guys are talking about AI at all.

5:12Unidentified speaker (speaker_1): Yeah.

5:13Unidentified speaker (UNKNOWN): Like

5:13Unidentified speaker (speaker_2): in fact, I think they kind of got it wrong. right? He said, We need to hold on for about 18 months. Like, what do we need to hold on for? Like, the way this is supposed to work, if you're doing it right, is vulnerability comes out that says Java needs to be go from you know I don't remember what the numbers are, so I'm just gonna make them up right now. From twenty-three 23 to twenty-four. 24. Some security engineer goes in. to some config file and edits it and moves it from 23 to 24. Like, yes, we can build all kinds of fancy tools around that. I understand. Whatever. Just trying to keep it simple. They check that thing in. The system sees that the stuff that's in production is at 23 twenty three and it needs to be at 24. twenty four. So it goes and builds a new image and it lays the software down on top of that and then it systematically goes around to all the different places where this image is needed and it upgrades them. Like that's what it's supposed to be. It's supposed to be a nothing burger.

6:16Unidentified speaker (UNKNOWN): And

6:16Unidentified speaker (speaker_2): like I don't even know why he's saying hold on. We shouldn't have to hold on. We it's just

6:20Unidentified speaker (UNKNOWN): right, well I think

6:21Unidentified speaker (speaker_1): part part part of it is at least the way I read it is in the AI angle is a lot

6:26Unidentified speaker (UNKNOWN): of

6:27Unidentified speaker (speaker_1): this stuff is coming to light because of AI, right? And which which is a good thing in some sense, right? But unfortunately, like the discovery of it is out in front of people's ability to fix it for some reasons that are um, you know, maybe reasons that shouldn't be there but nevertheless are and and uh should be addressed.

6:47Unidentified speaker (speaker_2): Yeah. A lot of people like are like, oh, we patch once a quarter.

6:52Unidentified speaker (UNKNOWN): And

6:52Unidentified speaker (speaker_2): I mean I've worked with I don't know how many portfolio companies that that's like a thing. And like

6:57Unidentified speaker (UNKNOWN): it

6:58Unidentified speaker (speaker_2): doesn't work anymore. You know, the world has changed. Like whether AI, this, that, or the other thing, like, yes, AI is finding some vulnerabilities. Some of these vulnerabilities are things that are really exploitable, and some of these vulnerabilities are like, well,

7:14Unidentified speaker (UNKNOWN): if

7:15Unidentified speaker (speaker_2): six other things happened, then it would be exploitable. Which,

7:19Unidentified speaker (UNKNOWN): you

7:19Unidentified speaker (speaker_2): know, that that's a whole r, you know, security is about risk. So like you have to decide like how much risk you're willing to take on. But like as things are getting found faster, we're gonna have to patch more often, which is something we should have been able to do all the time. And

7:38Unidentified speaker (UNKNOWN): if

7:38Unidentified speaker (speaker_2): you're not able to do that, how are you gonna do AI?

7:43Unidentified speaker (UNKNOWN): Like

7:43Unidentified speaker (speaker_2): how is that gonna work? Uh AI's gonna write

7:46Unidentified speaker (UNKNOWN): ten

7:46Unidentified speaker (speaker_2): thousand lines of code a day and we're gonna p release that once a quarter. Like, you know, you and I have talked about the circle CI report. Like that's a fast path to more outages, longer outages, more customer attrition. Like I mean

8:02Unidentified speaker (UNKNOWN): it's

8:02Unidentified speaker (speaker_2): uh we already know how that how that story turns out. We know the ending of that story.

8:08Unidentified speaker (UNKNOWN): Right.

8:09Unidentified speaker (speaker_1): Yeah. And so part of the benefits of AI is obviously you know AI-driven development, but you know, but then I would think also like the ability to discover vulnerabilities in your own network, right? And in your own infrastructure. Um, and in it sort of makes me wonder. how many CTOs are out there who are not kind of leveraging those tools out of fear, right? For what might be found and and the ability to to deal with it. I don't know what your

8:40Unidentified speaker (UNKNOWN): take on

8:41Unidentified speaker (speaker_1): that is, but I I feel like if I were a CTO, which is a maybe a scary thought, right? Like I would I would want um that that sweep through um my code base to be done and those those vulnerabilities found and then you know yeah we should have the ability to fix them. If we don't, we should we should get that ability quickly.

9:00Unidentified speaker (speaker_2): They already have those tools. Those tools are not those don't need to be invented, and there's nothing new about that. That's

9:06Unidentified speaker (UNKNOWN): something

9:06Unidentified speaker (speaker_2): we've been doing in the industry for decades. Like that's none of that's new. It's the fear is that if I do fix something, I do patch something, is it gonna break something? That's the fear. And that's what we have to drive out by having automated testing.

9:25Unidentified speaker (speaker_1): Right.

9:26Unidentified speaker (speaker_2): And the problem is that's a lot of work. Like it's a lot of work to make sure that all my stuff passes. And you know, maybe we're used to testing just like X because that's where the thing that changes the most and now Y is changing. Yeah,

9:41Unidentified speaker (UNKNOWN): I don't have confidence that

9:43Unidentified speaker (speaker_1): a passing test is enough, right?

9:45Unidentified speaker (speaker_2): Yeah. And uh, you know, that's that's something we all have to deal with on an individual basis for depending on my infrastructure, my code base, my all all these other kinds of things. And sometimes that stuff works out great and sometimes it doesn't. It just kind of depends on the situation. But like,

10:01Unidentified speaker (UNKNOWN): you know,

10:02Unidentified speaker (speaker_2): Microsoft is famous for releasing patches. I don't know what they call it, patch Tuesday or patch Wednesday or whatever, and then like everybody jumps around and and do does these things, and like now we have to have the ability to patch things on demand, which again, same problem with AI. I need to be

10:23Unidentified speaker (UNKNOWN): able

10:23Unidentified speaker (speaker_2): to deploy on demand, otherwise, I build up inventory. and then that creates outages and whatever. And I I guess for me, like the question I have for you is like, what does this look like for private equity people that want to have like a great exit for their portfolio company? Like, I can't imagine, you know, in the past, cybersecurity companies used to give out uh or cyber insurance companies used to give out insurance, and it wasn't that hard to.

10:51Unidentified speaker (UNKNOWN): get.

10:52Unidentified speaker (speaker_2): And over the years, it's become much more difficult because you actually have to prove to them that you're doing things that make it less risky for them to insure you, which is something in the beginning that wasn't a thing, and that's

11:05Unidentified speaker (UNKNOWN): just

11:05Unidentified speaker (speaker_2): become more of a thing over time. What does that look like now? What does that look like? You know, I want to have a big exit for my company.

11:13Unidentified speaker (UNKNOWN): Do you

11:14Unidentified speaker (speaker_2): think that's really going to get left out of due diligence?

11:18Unidentified speaker (speaker_1): Yeah, I mean it's it's obviously a risk. Um, you know, one way to look at it is a risk and and a and a growing risk, um, you know, surface. But the other one I think is just again what we've been talking about here, the uh

11:34Unidentified speaker (UNKNOWN): the

11:35Unidentified speaker (speaker_1): the additional potential benefits of having all that stuff in the right shape that are getting you know compounded now with ai and and and so it's not a simple matter of you know saying all right well you know your your stuff is is not in as great a shape as it should be and so um you know we'll we'll we'll fix it. Um it's like, well no actually like that is going to be gonna a key blocker for all these major compounding benefits

12:06Unidentified speaker (UNKNOWN): that

12:07Unidentified speaker (speaker_1): we would attach, you know, pretty big multiple to, but we can't do that because you don't have the right foundations in place. Or, you know, or that's likely to lead to vulnerabilities down the road that haven't even been surfaced yet. And like we can't necessarily put our put our fingers on the exact vulnerability, but like we know this process isn't the right process and that is going to have a whole lot of of um sort of um you know opportunity costs in a way attached to it um that wasn't there before but but is now

12:34Unidentified speaker (UNKNOWN): yeah

12:35Unidentified speaker (speaker_2): I think kind of the point you're making is if you make this investment

12:39Unidentified speaker (UNKNOWN): you

12:39Unidentified speaker (speaker_2): get it you get the benefits of it from multiple angles.

12:42Unidentified speaker (speaker_1): Right.

12:43Unidentified speaker (speaker_2): Not just

12:43Unidentified speaker (UNKNOWN): one.

12:44Unidentified speaker (speaker_2): Not it's not just well we did this because of X. Like we we did this and we're

12:48Unidentified speaker (UNKNOWN): that. going to do it. It's a

12:49Unidentified speaker (speaker_1): more valuable investment now which Hopefully it'll

12:53Unidentified speaker (UNKNOWN): get done

12:53Unidentified speaker (speaker_1): now when

12:53Unidentified speaker (UNKNOWN): it hadn't gotten

12:54Unidentified speaker (speaker_1): done before. Um and and let's just hope that happens and and you know the uh the the sort of bright, shiny uh objects aren't aren't necessarily the ones that are getting all the all the attention. But you know, these foundations are are actually part of the uh the the the brightness there and and uh and and should be addressed as well.

13:16Unidentified speaker (speaker_2): cool well i hope everybody got head faked by the cybersecurity aspect though i'm sure it'll be in the show notes or the or the show title but um and

13:26Unidentified speaker (speaker_1): somehow i doubt this will be our last cybersecurity

13:28Unidentified speaker (UNKNOWN): episode Dave

13:29Unidentified speaker (speaker_1): um we've we'll probably talk to some guests about this at some point but this was this was a good a good kickoff thanks everyone for listening

This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights