What this is: Nearer-verbatim operator notes from a public X interview segment. Not a court transcript. Gaps labeled. Hosting ≠ endorsement.
Not Code Solutions LLC viewpoints or advice. Speakers' claims stay speakers' claims (Intent / [unverified] / host-said).
Hedge legend: [inaudible] · [unclear: …] · [gap: …] · Intent/[unverified] for timing, capability, security-incident, and company claims · host-said for Sabrina Halper framing.
Speaker claims: Yinon Costica's statements about his own company (its products, the Google relationship, customer results) and claims about third parties (DeepSeek, the Andreessen post, Instinct) are his or the host's account [unverified] — not checked, not restated as fact. The 43:30–47:15 company segment is out of scope and not included.
Source URL: https://x.com/SabrinaHalper/status/2107893808344272933
Full episode duration: 1:01:44 (native X) · Scoped for Digest More: player 22:45–43:30 (~20:45) + 47:15–58:30 (~11:15) · 43:30–47:15 skipped
Host: Sabrina Halper · Guest: Yinon Costica (Wiz co-founder, per post text)
Posted: Wed 2026-10-07 ~1:00 PM CT · Notes desk: 2026-10-08
Contest OUT · SI framing = our desk wording only; third-party names (AI, Wiz, Google, DeepSeek, Instinct, etc.) kept verbatim when spoken.
Disclaimer: https://legal.codesolutionsllc.com/si-disclaimer — Notes of a public interview — not Code Solutions LLC views or advice.
Lane: Digest More feed only. Not a Daily X release from this pack alone.
[22:45] Yinon Costica
...behind it, so we can only imagine where this would go from here.
[22:49] Sabrina Halper
And you just led me to the next question I had for you, which is, as AI capabilities progress rapidly, putting on your future goggles, what do you imagine like Wiz Atlas 5.0 looks like?
[23:03] Yinon Costica
So, one of the things, again we talked about it in the difference between a model and then a system that composes different models for different tasks, and I think that one of the things we need to unlock is how do we take these expert skills, and how do we bake it into, again, people, process.
[23:25] Sabrina Halper
Yeah.
[23:26] Yinon Costica
How do we make it really accessible to any organization out there, whether because it's... it needs to be affordable, whether it needs to be scalable, whether it needs to be applicable in their own environment, and how can we turn this expertise that is now available to us defenders, how can we turn it into an ongoing process that any enterprise can implement in order to secure their codebase. And the silver lining here for cyber is that historically we've as defenders we never had the resources to secure everything all of the time, which is what we need to do as defenders, right? There is a attacker, defender asymmetry that the attacker needs to find one thing, defenders have to protect everything all of the time.
[24:14] Sabrina Halper
Yeah.
[24:15] Yinon Costica
And I think that AI in a way unlocks that ability to really look at, you know, my entire codebase and start securing it all of the time. So now we need to bridge over the gap of, okay, here is a technology that we know that can solve this problem, and how do we actually scale it so that any customer will have it as a toggle in their Wiz tenant, right? So that's the journey, I don't know if it's five years, maybe it's like more like weeks, months...
[24:42] Sabrina Halper
Two years.
[24:43] Yinon Costica
Yeah exactly, because we are heading there, it's uh, in different phases of previews, we see it already working, but I think that this is... this is what we will need to focus on in the cybersecurity industry is how do we harness these capabilities for the benefit of any customer out there, not just the top one percentile of the market, but anybody.
[25:09] Sabrina Halper
It's so nice to hear that you think that the asymmetry... because there is this deep asymmetry in cyber as you just mentioned, where an attacker has to find one vulnerability, while the defender has to protect everything. So, I feel like there is this very negative sentiment right now. There's just this thing that went viral on X from Andreessen, that basically was like cyber risk has skyrocketed. And it's a belief that the asymmetry is going in the wrong direction. And just to go there for a second, AI obviously dramatically lowers the cost of a cyber attack and also reduces the skills that a hacker might need. So, what is that version of things going on that you're seeing?
[25:54] Yinon Costica
I think that we have... now is the, in a way, now is a very critical time for cyber, because it's easier for attackers to leverage AI to perform one attack, than what it is for defenders to secure their entire estate using the most novel AI security. In the long run, defenders can actually benefit from it because of the reasons we've mentioned earlier, because now they can actually do security across the board all of the time.
[26:28] Yinon Costica
I think that now we're in this twilight zone, right? That we need to accelerate the adoption by defenders before attackers take advantage of it. And I think that this is a bit, you know, our conviction, and when I wake up in the morning and you know... we've sold to Google, [unclear: brief aside], but for me it doesn't change anything because I feel that our responsibility is now to take the best technology, now [unclear: "aided" / "added"] by Google as well, Google models, Google threat intel, and make sure that we bring it to market as soon as possible so we can secure the digital estate we all leverage every day, right?
[27:06] Yinon Costica
So that's like the mission I wake up with every day and we feel you know, we feel the rate of change and the need from customers and we know that this is our mission, to provide them with these capabilities to secure themselves. And us.
[27:28] Sabrina Halper
Yes. And it's cool because everything you build for yourself ultimately every company needs as well. We are not only in the AI era, I think in the last year the use of agents has rapidly increased. Both for enterprises in terms of how much of the most important companies' codebases are now being written by agents, but also for consumers, like, and I want to ask you about both of those things specifically. Maybe we'll start with consumer because it's fun. Have you heard of Instinct?
[27:56] Yinon Costica
No.
[27:57] Sabrina Halper
Okay, so Instinct is an American company, it really was just released like a week ago, it went into kind of like public beta, and it's a personal assistant agent...
[28:06] Yinon Costica
I did hear it.
[28:07] Sabrina Halper
Okay, so I... a friend sent it to me and it's amazing how much I've used it in the last week. I've been in Israel, like, it has booked my appointments for me, booked a hotel. I find myself being so reliant. I haven't given it access to my email or credit card yet, but most of my friends have. This is really like a new brave world that we're entering. I don't pay for any personal cyber security right now. How do you see consumer cyber security changing? And is there anything like you see other people doing that you wouldn't feel comfortable to do? Like what's your message to consumers?
[28:47] Yinon Costica
Absolutely. I think one, it's very in a smaller scale of an individual it's very similar to what we see in the enterprises, but here, I think the first thing that we basically realize is that the opportunity is immense, right? The transformational ability to do things differently. And Instinct, I think also the technology is not new in agents and what it can do, Instinct has made it easy. And that's what actually triggered many to just, okay, I don't need to be a nerd to do this.
[29:20] Sabrina Halper
I don't need to set up an OpenAI API, all these things, I don't need a Mac mini, it's just a text message, exactly.
[29:24] Yinon Costica
Exactly. So I think that this ability shows us that ease of use can unlock tremendous potential for really, you know, just better life, right? Better use of technology. And maybe that's the most important thing to remember before we talk about the other aspects, because...
[29:45] Yinon Costica
The opportunity is huge, and it's positive. Uh, but the second thing is that we also need to put guardrails and think about what is the risk that we might incur, right? So, like you've said, maybe, maybe we need to give it some time before we give it our credit card, maybe we need to think about how do we secure it, maybe we need to look at the audit logs, maybe we need to look at the permissions. And there is a whole sequence of questions that we need to be asking ourselves: Where, where does our data sit? Where is it sent for? Who is this company? Uh, what, what, you know, what regulatory compliance they actually adjust for, like, do they, you know... So there are many, many questions that over time we'll need to ask.
[30:28] Yinon Costica
I think, uh, the, you know, there is an ecosystem that needs to be created in terms of the, uh, regulatory, uh, responsibility. Like we had GDPR for personal data, like we need to think, okay, so what does it mean on companies that provide these services to individuals? What responsibility they have? Uh, so it's all, it's an evolving ecosystem. I think it's not going to be, uh, you know, we're not going to solve, solve it overnight, but I think that we all have a great incentive to build an ecosystem that we can all trust eventually, because we understand the benefit in it.
[31:07] Yinon Costica
And the one thing I will say that I do think is very different than in the past, I think that the, in the past, we really had to explain, uh, the risks of cyber, and why it's important, and, you know, uh, even what does it mean. And I think today the level of awareness is much, much, much higher. Not, not across everybody, of course, but I think in general, uh, the average person understands more, and it's an opportunity also. It's a good thing to look at.
[31:37] Sabrina Halper
Definitely. I think in ways the AI companies themselves are doing the messaging. Like, they're being very open with, you know, incidents when the models do things that the researchers didn't intend them to do. And, um, all the discussion about CISOs, it feels like anyone who's the head of security at any company, this, this is top of mind. So, um, I'm sure that's great for all cybersecurity businesses as it should be.
[32:02] Yinon Costica
Maybe I'll just add one more thing. In the end, everything in security in the end, uh, starts and ends in, uh, trust, right? So who do I trust, uh, to do this? And I think that over time, uh, we will need to, everybody, consumers, enterprises, we will need to define who do we trust and why, uh, in this era. So do I trust any model that is now released, or any, uh, agent provider that I need, or do I actually trust those that I know they are, they have the capacity to actually secure what I'm doing?
[32:39] Yinon Costica
And I think, you know, uh, trends will go, uh, you know, trendy software will go up and go down, but, uh, in the end, I think that whoever will remain is the one that will be able to win trust by the majority of the consumers, enterprises, because in the end that's the most precious thing we have, especially for new technologies that have super access and can do a lot.
[33:03] Sabrina Halper
By the way, I think that's a really interesting point. Um, where maybe a couple of incidents have to occur, but I think the way people think about even adopting new consumer technology might be different. Like, we might all look back on this moment where we're just, like, downloading this brand new startup and giving it access to everything as, as questionable. I mean, it's a great company and it'll, you know, grow, but it's, it's a lot of trust that maybe it hasn't, hasn't earned in ways. So I think that's a very interesting... and I, I think that's a way where it's like, where big tech and these companies that really have the ability to hire the best security, you know, acquire the best security company in the world, like, the reason to, to still be reliant on them, versus as software gets so cheap to make, just adopting any software that comes your way, you know?
[33:51] Yinon Costica
Exactly, that was, uh, also where I was pointing to, because in the end it's easy to create the first batch of, like, wave of software, but then what's the model that actually secures it and continues to maintain it? And, uh, again, I don't want to go too technical, but in the past year, uh, we have had a huge, uh, sequence of, uh, incidents, uh, around supply chain, where, uh, you basically, you're not attacked directly, but you are attacked through the supply chain that you use, the open source supply chain you use as a developer. And I think that this, on the other end, also created a lot of questions of how do we build trust in the software that we consume as developers.
[34:35] Yinon Costica
But the same thing, in the end, will propagate to consumers and machines and the what we install on our phones, because the ease of creating software makes it easier for, maybe, not all the best secure software.
[34:49] Sabrina Halper
On that point, I just want to ask you, um, there is a rise in popularity of, like, open-weight models, and this means that anyone can download a model and, uh, fine-tune it and remove the guardrails and use it for any purpose. So it makes hackers stronger, hypothetically. Um, how do you think, like, this proliferation of open-weight models changes the cyber landscape?
[35:15] Yinon Costica
So, it's a great question. I think that first, uh, there will be a need in the future, uh, to create more and more task-specific models for, for cost reasons, for quality reasons, for, uh, many different reasons, for just building, owning your own stack in a way. I think that, uh, on the other side, we need to also be aware that, uh, it's a supply chain issue. We are getting, like, a model that we don't know exactly how it was formed, we don't know exactly, uh, what what it includes and how it behaves in certain, uh, inputs, outputs. And that's where I think that we, if we, if people choose to use this, we need also to scrutinize, uh, the source of it.
[36:00] Yinon Costica
And are we able to actually test for all of these, uh, potential impacts in a way that gives us the confidence and the trust that we need? Now, uh, I do think again, there is a need that, uh, over time will require these open-weight models, uh, but also we will need to do it in a way that is responsible and testable at scale. So, this is one of these things that is still very early and we need to grow the, um, the framework around adopting it.
[36:33] Sabrina Halper
Alongside.
[36:34] Yinon Costica
Alongside, alongside the, the rapid use and adoption and innovation.
[36:38] Sabrina Halper
Um, going specifically on DeepSeek, I know you guys found some very interesting vulnerabilities when you looked into it. Um, can you, can you share a bit about that?
[36:50] Yinon Costica
Yeah. Actually, DeepSeek, I love that example because, uh, we tend to think about, uh, AI and models and agents as something that is completely detached from the, in the end, from the underlying infrastructure that, uh, runs it.
[37:04] Sabrina Halper
Yeah.
[37:05] Yinon Costica
And in DeepSeek, the interesting thing is that, uh, when, uh, DeepSeek went out and we, you know, we wanted to basically understand, uh, you know, basic, uh, who is this, uh, company, how is this, uh, product, we ran a very basic scan and we were able to find this, uh, vulnerability that is not in the model or the agent, it's actually on the underlying infrastructure that the model, uh, leverages for inputs and training sets. And we were able to basically, uh, access, uh, very large sets of data that, you know, should, should have not been exposed, and they were exposed.
[37:46] Yinon Costica
And I think that this is a good reminder that in the end, uh, agents and models, they in- they, they add additional layers of risks, uh, more attack surface that now we can, we need to look and secure. But everything in the end runs on infrastructure that still has the same vulnerabilities and misconfigurations and identity-related issues that, uh, needs to be taken care of and needs to be, uh, scanned for and fixed all of the time. So in a way, we just need to remember that it's a stack. And on top of the, uh, cloud and the infrastructure, now you have AI, but you can always target the infrastructure below. And it's a good reminder to secure the foun- secure the foundation, secure the basics. The fact that you are doing AI doesn't, doesn't give you a pass.
[38:37] Sabrina Halper
Yeah.
[38:39] Sabrina Halper
Okay, while we're on the topic of models, the model incidents more generally, like, where companies have come out and say, um, you know, with a very transparent front, basically being like, we were evaluating this model and it, it acted in ways that the researchers didn't intend to during the evaluation. How serious of a cyber risk do you perceive that to be?
[39:01] Yinon Costica
I think that it gives us the evidence... us, I mean the general industry, it gives us evidence that we need to prepare for a world where AI is attacking our estate. And it gives us, uh, indications of how this attack could look like, or how should we be thinking about these attacks. And, uh, one of the things that are the most important when I think about, uh, we call it, uh, AI threat readiness. So how do we prepare for these AI-powered threats? And I think that, uh, there is no other way to think about it than we have to continuously... we, we, we can leverage the benefits of AI in attacking ourselves before others try to do that on our estate, right. So, uh, we always say best, uh, defense is offense.
[39:57] Sabrina Halper
Yeah.
[40:00] Yinon Costica
So it's a paraphrase, because actually it's not really going on the offense, it's actually using offense against yourself to test your defenses, and actually identify these, uh, weaknesses or attack paths before someone else finds it. And I think that, um, on this front, attackers in the end have very little knowledge of the environment before they try... before they get in, right? So maybe they know the very... the external surface and they can try to, uh, pen-test it and try to hack inside.
[40:34] Yinon Costica
But they don't know the internals yet. And I think that the defenders have a huge advantage in the sense that they do know, and they can actually analyze it to, uh, really have deep context as they are looking for these attack paths. And I think that this is maybe the, the biggest advantage that we have as defenders is this level of context that we can put in play before, you know, uh, anybody else.
[40:59] Yinon Costica
The other benefit that we have, uh, as, uh, defenders is actually the, uh, ease of, uh, generating the, uh, threat intelligence globally. And I want to explain because we've seen these, uh, attacks, and we all read about it, but the reality that security companies behind the scenes, we are automating the process of fingerprinting these patterns. And because we're using now AI to do that, we can accelerate that process as well.
[41:36] Yinon Costica
So, let me give you an example. Let's say we've seen a supply chain attack in the past, it would have taken many, not many, but many minutes or let's say a few hours to actually get, uh, the fingerprint, the attack, and the indicator of compromise, and then to propagate it across the environment would take additional time.
[42:00] Yinon Costica
Today, we can basically identify that the attack is happening using AI in record speed, in like very few minutes, seconds. Because we see a pattern that doesn't add up, AI can actually identify it, propa- create the indicator of compromise, and propagate it to the rest of the environment. So the way we are thinking about attacks actually changes in two dimensions. One, we can... now we have resources to test our, uh, surface all of the time and try to attack before others do. And second, when something happens, we can actually learn faster than ever in order to make sure we can send immunity to the rest of the, uh, environments. So it's very significant, uh, advantages that we have as defenders.
[42:46] Yinon Costica
And the third thing that we need to think on is a model advantage. Uh, over time, I, uh, believe that we will be able to train models in a way that will be very specific for the defensive tasks that we need. And this will give us an additional advantage. Especially now looking at the, uh, you know, uh, [unclear: Wiz in the Google context], the ability to actually create very specific models for that is another advantage that we get on top of it. Uh, so it's a compounding advantage that we hope that, uh, defenders will use. And the ability to see attacks in the, in the lab, uh, gives us good intelligence on how should we think about defense in real time.
[43:29] Sabrina Halper
Yep.
[43:30–47:15 not in scope — skipped.]
[47:15] Yinon Costica
...everything within them, AI became yet another thing that we just have to secure. So that came organically. The transformational thing was around how do we use AI to secure.
[47:25] Sabrina Halper
Yes. It's both sides of it. What do you think is the biggest unsolved problem in cybersecurity today? Especially in this rapidly changing landscape.
[47:35] Yinon Costica
So I think first, as we discussed, the technology today exists to help defenders. Which is great, great news. I think the biggest unsolved problem is how do we create processes to implement it at scale in organizations, and by doing this we have to solve many problems that today slow us down. In the sense that we need to have many human approvals, so how do we build trust in agentic workflows to do the job for us? We simply have to get through this if we want to benefit from the AI technologies that can help us defend.
[48:16] Yinon Costica
Um, we need maybe to think about many silos. When an attacker looks at an environment, they don't care about organizations or teams, they look at the estate, they go in. And on the defense side, we have many silos. Someone takes care of the application security, someone takes care of the cloud, another person takes care of the network, of the identity. It's all very fragmented. And if we really do want to secure it, we have to think about how do we create a single view or a single process that secures across all of these teams. So we have organizational, um, fragmentation that we need to get through.
[48:53] Yinon Costica
So if we get through the ability to automate and then do it across the entire environment, then I think we are in a really, really good position not only to, you know, not only to match the threats that AI may pose, but I think that it gives us a very different horizon for how we think about security a year from today. Security has always been a very bottlenecked area. We could only do this much to secure our environment because it was a people-heavy process.
[49:30] Yinon Costica
And if we remove that bottleneck, maybe we can create a security environment that is self-healing and secures itself. And it's not something that I cannot imagine that is happening. So the future is bright, we need to remove the adoption blockers to get there.
[49:47] Sabrina Halper
Yes, 'cause we want companies to be able to, um, like leverage how quickly and how capable these models are getting and these agents are... um, without risking everything.
[50:00] Yinon Costica
Exactly.
[50:01] Sabrina Halper
When you look at the risk, do you feel like it's equally risky for companies leveraging agents across, you know, code and now in all these other ways? Do you think the bigger risk is agents will act in unintentional ways and they won't have like visibility into everything? Or is it really just that like hackers can hack in, but also maybe get people's agents to to do unintended things and get in that way?
[50:28] Yinon Costica
It's an interesting question, they are orthogonal. And in a way, um, it's funny because you can take an agent with all of the good intentions and, um, you can give it access to sensitive data and if you ask it the, you know, the right question, it will happily harvest through all of this data and give you the answer, even if it shouldn't have, right?
[50:52] Yinon Costica
So I think that the notion of, when you build agents, we have to think about both risks. Of users, accidentally, maliciously, the agent itself going rogue, and we have to create these layers of guardrails into the agents ahead of time. And guardrails can be at the 'what do we even get... give it access to?'. Guardrails are the audit logs. Guardrails are 'who do we expose it to?'. Guardrails are 'how do we inspect the, uh, prompts and the responses so we know that it's not out of line?'.
[51:26] Yinon Costica
And I think that we have to think about all of the use cases because the behavior becomes... it's not unpredictable, but it doesn't really depend on who's triggering the agent, it depends on what the agent can achieve. And that's where we need to put most of our attention on. So securing the agents, guardrailing it.
[51:49] Sabrina Halper
Yes. I bet there is a lot of, um, founders and just people at companies right now that are torn between 'we need to move so quickly to compete with everyone' and 'how do we simultaneously secure it?'. 'Cause some of the founders I've talked to are like 'everything changes every week and you wanna, you wanna leverage it.' Um, so it's definitely a, it's a crazy time.
[52:11] Yinon Costica
Crazy times, I agree.
[52:12] Sabrina Halper
What is the longest autonomous cyber attack that you have seen an agent successfully complete? Like how capable are these agentic cyber attackers? Or like the ones that hackers are using?
[52:29] Yinon Costica
Listen, I'm looking at the Red Agent that we produced. We, the Red Agent by Wiz is a defensive attacker, right? So it's like an agent that takes the context from Wiz and tries to hack from the outside in. And I can tell you that the... first, almost every customer that enables the Red Agent finds things that we couldn't have found, us humans, until that point. And it basically...
[53:00] Yinon Costica
Basically the crazy thing is that it happens, it operates from the outside. So it just walks into your environment in very creative ways, and when you look at it you say, "Oh, wow, yeah, it makes sense." But I think the ability to enumerate all of these attack paths would have been impossible before. So the level of creativity and the stubbornness, right, of the agent in actually finding these uh combinations is really something that uh every time I get surprised by the things it can find, and I think also our customers uh give us the same feedback.
[53:36] Yinon Costica
And I think that when I look at this level of creativity and um I really, I really understand the unlock of the speed, scale, expertise that AI enables and why it's so important for defenders to use it first.
[53:57] Sabrina Halper
And, and my last question before we end on a more positive note is just um, beyond security of an enterprise, what do you think the internet looks like when a meaningful percentage of the players and users are not human?
[54:17] Yinon Costica
Yeah, I think it's a great question. I can even, you know, even as a product leader I'm looking at Wiz and I realize that we have our user interface and it's beautiful and everything, but I realize that over time uh more and more of it will be consumed through the eyes of an agent, right? And uh the users will never see uh Wiz. And that's okay. Uh, because when I look at Wiz I think, okay, we've built a, let's call it a system of record that really understands uh environments in ways that are valuable for users and agents to consume from Wiz.
[54:56] Yinon Costica
Uh, I look at the uh largest SaaS companies in the world and I know that there has been this talk about SaaS-pocalypse, but the reality is that SaaS just evolved, and Salesforce now becoming the system of record for the business, and Jira becomes the system, Atlassian becomes the system of record for, so each one actually leverages the uh deep data model they've built and making it available for agents as well as humans.
[55:26] Yinon Costica
And through these examples we can maybe look at the internet as a whole and we can say, okay, so what's the future of the internet, right? We, we, we're very used to consume data through specific websites and specific applications, uh and this is what made the web what it is, but maybe tomorrow we will need to think about data repositories that will be available through the web and will be consumed by agents. And maybe a website shouldn't be as much of a, a visual website as much as a...
[55:57] Sabrina Halper
Like, who cares about user interface or how, yeah.
[56:00] Yinon Costica
Yeah, maybe I just want to make sure that the agents get the information they need and I need to attach an MCP uh server to every website so it will be uh very intelligible to agents, better than the UI for humans. So I think that um, I think that we already have hints to where the internet will go to.
[56:20] Sabrina Halper
It's very interesting. Um, totally. And how advertising changes, it is, it's, it's all changing a lot. Um, if you look forward, five years might be too long, so we're gonna, we're gonna cut that to a year and a half. Um, what aspects of cybersecurity do humans still need to be primarily a part of?
[56:45] Yinon Costica
Wow, it's an amazing question. I think that, again, uh, it's, it's hard to make predictions, but...
[56:51] Sabrina Halper
We don't know the future. It's impossible. So we're just trying our best, but...
[56:54] Yinon Costica
I do think that uh humans remain the uh moral compass of whether we are doing the right thing or not. Uh, they remain the human sense of whether this is what we wanted to achieve or not, and becoming the guiding, uh the guiding factor in the end of, you know, whether we achieved it or not. There is something very tricky in cyber, because cyber, unlike anything else or unlike, you know, many other domains, cyber is a very yes or no type of thing. Whether I was able to hack in or not, it's a yes or no question, right? Binary. It's like, that's it.
[57:38] Yinon Costica
And I don't think there are many things that are binary in the world. Is my security program good enough? Um, I don't know. We need to look at it. Maybe I can have a hint, but in the end there is a human that needs uh to give, uh to be the judge and say whether we are heading the right way or not. And we can build as many layers uh of agents and machines to do this for us, but in the end the overall direction uh will be, I hope by a human, that will steer the program. Uh, so this is where I think that, you know, the moral compass and the uh trajectory, the goals, the outcomes that we've achieved, we still need the humans to guide. And it's a lot. It's a lot to do, and I hope that we will spend more time on doing this as we uh automate more and more by the agents.
[58:27] Sabrina Halper
Um, I think there are... hopefully...
[Scope ends 58:30.]