SentinelOne CEO On Why AI Agents May Be A ‘Transient Concept’

In an interview with CRN, SentinelOne CEO Tomer Weingarten says that while agentic capabilities are promising, partners need to stay ‘agile and flexible enough to support what comes next’ even beyond the technology.

While the concept of AI agents holds major promise for businesses, it may ultimately be a “transient” phenomenon with other technologies soon superseding it—suggesting that partners and customers should steer clear of agentic capabilities that require a lengthy deployment process, SentinelOne co-founder and CEO Tomer Weingarten told CRN.

“I almost think that in six months, in nine months, you’re going to see [another] shift in how people on-board AI to the enterprise,” Weingarten said in an interview during RSAC 2025 last week.

At RSAC, SentinelOne unveiled a set of new agentic capabilities as part of debuting the next version of its Purple AI security operations technology. The new release, Purple AI Athena, includes agentic AI functionality that aims to mirror security reasoning and orchestration of experienced security analysts.

However, while other vendors require a “long list of prerequisites” in order to deploy AI, “I think what we have done with Athena is to say, ‘Look, security and AI should be everywhere, and it should be easy—and you should just, [with] one click, connect to your entire security ecosystem,’” Weingarten said.

This is crucial, he said, in part because of the rapidly changing nature of the AI landscape. The situation poses a risk that some agentic capabilities, which are more cumbersome to deploy, may already be “irrelevant” by the time they’re up and running for partners and customers, according to Weingarten.

Ultimately, “that might lock them into a place where it might not be agile and flexible enough to support what comes next,” he said.

During the interview last week, Weingarten also discussed the biggest opportunities ahead for partners with SentinelOne right now, including around the company’s AI SIEM (security information and event management) offering.

Here is more of CRN’s interview with Weingarten.

What are the major advancements with the launch of Athena?

This journey started for us five or six years ago in terms of the concept of autonomous cybersecurity. I think for many years, people didn’t believe that autonomous cybersecurity [was] in the realm of possibility. We’ve always believed that what we have been able to deliver for the endpoint—which is a complete, self-healing system [with] no human in the loop—is something that we can apply for all of enterprise security. Athena is really that step forward that shows you the art of the possible. It comes with a few pivotal components for us as a company because it basically detaches and decouples AI from the need of the data platform and the other dependencies. Every [security] vendor is trying to sell you their entire platform—you have to get all of these components to get more secure. And you’re going to have to plug this and plug that and migrate the data and build new pipelines and move away from the incumbent. It’s a whole long list of prerequisites to even get to the point where you say, ‘OK, I’m now enjoying AI.’ I think what we have done with Athena is to say, ‘Look, security and AI should be everywhere, and it should be easy—and you should just, [with] one click, connect to your entire security ecosystem.’ This shouldn’t be a $10 million migration and services project to just get security now. Because by the time you’re going to deploy this agentic capability that your vendor is advertising and make it worth something, it’s going to be probably irrelevant, if I’m honest.

So you feel like organizations are not thinking long term, getting overly focused on the trend of the moment?

I think we’re always at this point in time where we feel, ‘OK, this is it. We’ve solved it.’ [Today] it’s agentic. But I think that agentic is a transient concept. I think that if you look at what Athena is showing, it’s actually showing the orchestration of multiple agentic capabilities. So for you as the user of the system, you want to really think about it in terms of the system—and not the individual agentic capability. Because you’re going to have to imagine that all of these systems are going to be made out of multiple agents and agentic capabilities. You’re not going to manage each and every one of them by itself. You’re going to have a system. You’re going to give permissions to the system. You’re going to control the system. You’re going to supervise the system. You’re not going to go capability by capability and start trying to wrangle all of them into something that’s coherent. There’s an analogy with endpoint [security]. We’re not going about securing every single file, we’re securing the endpoint. That’s very similar [to the] world we live in today—you’re going to have so many agentic capabilities for so many different elements of what a security system needs to be. So for us, Athena is really more about showcasing a system for AI cybersecurity and less about a singular agentic capability. Agentic capabilities are going to be one of the many things that the system is going to be comprised of.

What’s the biggest AI-related issue you’re discussing with people during the conference?

[One issue is], ‘How do I on-board AI usage safely into my enterprise?’ We’ve got our own AI security posture management capability. But I would highly caveat that a lot of the models that we’re seeing today—in terms of how people are thinking about AI security—[are] such an infancy motion right now. A lot of these [approaches] that you’re seeing today that revolve around red teaming and model security—to me, they’re very immature. And also, the ways that people are deploying AI in production today are really in an infancy stage, and they’re very limited in their use cases. I almost think that in six months, in nine months, you’re going to see [another] a shift in how people on-board AI to the enterprise.

If you’d asked that question a year ago, ‘Hey, how are you using AI in your enterprise?’ Then people would say, ‘Oh, the chatbot, the LLM.’ [But] now we’re talking about agentic. In a year it’s going to be something else. So the models of how you secure them, you have to believe, are going to change. [That] is why I think if people are kind of jumping into a bandwagon today where they say, ‘This is what I’m going to do. I’m going to go with this technology or that technology’—I think that might lock them into a place where it might not be agile and flexible enough to support what comes next.

A lot of these [products] seem like they are trying to solve certain problems, but some of these problems are going to be solved by the model companies themselves. And some of these problems are going to take a very different form 12 months from today. Honestly, I think that the focus that is lacking today for everybody across the board when on-boarding generative AI applications is deep guardrails. Nobody can say that the models that are out there are secure. And lo and behold, there is a new technological revolution, and we’re back in that same [situation] where security is an afterthought. And nobody can say differently. This entire conference is trying to run after the genie that’s already outside the bottle. So it’s an afterthought. Nobody can tell you it’s secure by design. It is not.

I think with that notion in mind, you have to really think, ‘How do we design the next iteration of this? Versus, how do we find all the holes and plug them’—which is what we’ve been trying to do for decades now. Look, 2024 wasn’t a great year for security. We saw some of the biggest breaches in history. I don’t think we’ve solved anything in a significant way. Have we protected endpoints better? Maybe—so attackers are now exfiltrating directly from the cloud.

There’s a ton of foundational work that needs to happen. The No. 1 threat vector is still exploits. Nothing has changed. But yet we’re running to protect all these new things, and there’s a lot of shiny new technologies—but we need to deal with vulnerabilities. Let’s all pull back and really remember we’ve got to deal with the infrastructure. Because whatever you deploy on infrastructure that is fragile and broken is never going to be secure.

Do you feel like other vendors are approaching this too narrowly or too inflexibly then?

I think that they’ve, for years, been thinking in silos. ‘I’ve got this product and that product, and this is supposed to solve this, and that is supposed to solve that.’ But the problem is holistic. Cybersecurity is across all of these different surfaces. It’s not enough to secure one surface. You really have to think about, how do I even uncover all of these surfaces? A lot of the problems that we see today are from these surfaces that are unmonitored or that you have a blind spot to or you just don’t have the ability to protect it like some other surfaces.

Even the discovery and the configuration of surfaces needs to be automatic—ideally, autonomous. And I think that’s where, again, with the Athena release, that’s where we’re hinting toward. That’s kind of the future that we envision, one where these systems—and it’s only a question of time—are going to be so autonomous in their operation that we need to start programming them to actually do the entire heavy lifting as well.

We need to start thinking about how we use the systems themselves to deploy securely to begin with because these systems, what they give us, more than anything else, is scale. They give us the ability to see everything. They give us the ability to cover everything and to react to everything in a manner that [even] an unlimited amount of humans almost can’t.

Agentic definitely seems like a technology where, if you don’t do security from the beginning, the consequences are going to be much more serious than postponing security for some other technologies. Would you agree?

One hundred percent. The damage potential and the consequences are even much more dire if you deploy a super-secure agentic workflow—but somebody compromised the admin, and they now control that workflow. And again, this industry is great at marketing, it’s great at pointing to the next shiny thing. But we’ve not solved any of the problems of the past and [now we’re] running toward the future. So let’s at least focus on the present for two seconds and try to solve these issues.

In terms of what you’re hearing from customers and partners, would you say pretty much all of them are interested in using AI for security at this point?

One hundred percent. I think there are three main pillars to what partners are going to be able to do with Athena. One, just expand their practice. Now you can expand much beyond a single surface and start delivering your services on top of [all] the entire enterprise facets that you cover. That’s a major opportunity for them.

The second thing—which is something that they’ve been doing already with Purple AI—is really just to get better, faster, more scalable with AI. AI is not going to replace them. Spoiler alert: It’s not going to replace any security practitioner. They’re all needed. But it gives them scale; it gives them abilities. It gives them more and more advanced capabilities.

And the third piece is it’s going to shift their role to being more of a supervisor of a system than the hands-on controller of the system. And the more they’re able to delegate to the system, the more scalable and the more efficient they become. That will speak to their business model; that will speak to the customers they can serve. We’ve been saying for many, many years, ‘Let’s democratize cybersecurity.’ I think you’re nearing a point where it might actually be a doable thing with AI. If you can create an easy experience—where on-boarding is easy, where it’s a one-click [process], where it’s iOS-level easy to deploy security for your organization—that’s how you democratize cybersecurity. And I think that’s what Athena is starting to show—that security doesn’t need to be just sophisticated and complex. It can be super-advanced and easy at the same time. That’s the biggest unlock that we can all wish for.

In terms of the advancements around reasoning capabilities with Athena, what could you say there?

When we designed Athena we said, ‘Let’s take the collective learnings of our analysts, of our MDR service, of our entire partner ecosystem.’ We have one of the most robust partner ecosystems. These are partners that are leveraging our technology and Purple AI to deliver their own MDR services, their own response services, their own security services. That is gold for how you learn [and develop] these reasoning capabilities. Because now I’m not limited to only how SentinelOne reasons and what we’re seeing in our environments. I’m also privy to all of what my highly scaled partner ecosystem is seeing globally. So when you take all of these learnings, and now you train something like Athena, you end up with a very, very smart reasoning model that can start thinking on all of these different edge cases and possibilities in cybersecurity. It’s really important.

I think one big, overlooked element of how good AI can be is how diverse the dataset [is that] you train it with. In our case, with our strength of serving the MSSP market and small-medium businesses—all the way to the Fortune 10 and governments globally—that level of diversity and visibility is really second to none. We're operating at a scale where the coverage that our products are able to provide for the systems that we cover is best in the market. Just look at MITRE [ATT&CK evaluations]. So we’re seeing more than anybody, and we’re seeing more diversity almost than anybody. That really gives us a compounded advantage when you think about these reasoning models.

How big of a deal is AI SIEM for SentinelOne? And what are you seeing in terms of legacy SIEM players getting displaced?

I think that, first and foremost, it’s all about flexibility. I’m quite sure you’re going to see customers that want to move away from their SIEM, customers that want to augment their SIEM, and customers that want to apply AI to their SIEM. And we’re the only company that can offer that. There’s literally nobody else that can help you, if you can’t really move off your SIEM right now. And moving off a SIEM, it’s not an easy thing to do. So I think for us AI SIEM is for folks that have already experimented with the SentinelOne platform through our endpoint capability or cloud capability—and they’re saying, ‘OK, we’re going to put more data into it.’ And now it becomes a complete data lake for them. It’s an amazing opportunity. One of the fastest-growing parts of our business is that data storage and security event management [related to] it.

But there’s a big realization that I think a lot of folks here either don’t realize or don’t like to talk about. There is almost a zero percent likelihood for a large enterprise to be able to magically wrangle all of their enterprise data into one place. I think it’s virtually impossible. Security data, today, exists in the SIEM—be it Splunk, QRadar, our SIEM. Sometimes you have multiple SIEMs. Then you’ve got some data in AWS, and if you’re a [Google Cloud] customer, maybe in GCP. And maybe you’ve got Snowflake and there’s some data there. To think that you’re now going to find ways to shift all that data into one distinct, sanitized data store—if you even want to do that—I don’t think that’s going to happen. So when you think about something like Athena, it’s designed to be there to tell you, ‘Just connect me to all of those. I will then give you the visibility, the holistic correlations. But don’t waste your time on starting to shift data pipelines and trying to wrangle all your data. Just point me to those places, and any other place that I discover data in, and I’ll be able to give you the insights, and I’ll be able to give you that reasoning. I’ll be able to stitch it together with your other data sources.’

So suddenly, you’re not only dealing in a much more effective way with your data—you’re now getting much better coverage because you’re marrying together all your different data sources. And if you choose to remove one of them—if you want to migrate away from one data store to a different data store, sure, we can offer that for you as well. And it’s very easy for us because we’ve got the SIEM, we’ve got the data lake, we’ve got our proprietary technology. That’s how I see things shaping up, and that’s why I think that people need to think a bit differently about the SIEM opportunity. It’s not like, ‘We’ve got to replace the incumbent, and that’s the opportunity.’ I think that’s part of the opportunity. But if your entire growth is predicated on yanking out SIEMs, I think you’re going to find that’s not as [big of an opportunity] as you thought.

What’s your suggestion to partners about what their AI security priorities should be at this stage?

Start with the basics. It really needs to be a secure environment that you’re on-boarding AI into. Before you think about how you secure AI, make sure you’re on-boarding it into a secure environment and [that you know] what that actually means. I think if you’re able to solve that, it also will become much, much easier to on-board AI.

And then, just be cautious of all these new technologies that have come to solve the problem of today or maybe of yesterday. The pace of change is nothing like we’ve ever seen in the history of mankind. So obviously, if anybody’s offering you a solution today, it might not be exactly what’s going to be needed tomorrow. So the emphasis on flexibility and modularity, I think, should be greater than what people have assigned to it in the past. You’re really thinking about some of these solutions more as building blocks—as [technologies] that can glue things together, that can observe things, that can give you a better vantage point—that from there, you’ll be able to build further.