Remote Access for AI Agents and Idle Workstations

Hey everyone, welcome here to back to Techstrong TV. I'm really excited to have this next gentleman on. " Let me introduce you to Benji Boxer.

Benji is the co-founder and CEO of a company called Faze. Hey, Benji. Welcome to Techstrong TV.

It's great to have you on here. It's great to be here. Thanks so much, Alan.

You're welcome. So I'm looking at the art in your background, and clearly the two pieces on top are pieces... Well, they're all art, but those look sort of commercial-ish framed art.

But then behind you, on your, I don't know if it's your right or left because these things get mirrored, it looks like someone's real work of art is there. Yeah, I mean, that's the real art in the room. There's also a Spinosaurus behind me as well, so.

Oh, beautiful. Yeah. I remember.

You know what? So my children are 27 and one's going to be 25. We still have those art like you have there in plastic crates in the storage because my wife will not throw them away.

Yeah. So. We've been accumulating a plastic storage setup as well.

But I also, one of the things that's great about AI is I vibe coded an app for my family where I record my son telling us what he made and then take pictures of it, and then we have a gallery. It's pretty neat. Oh, very cool.

Yeah. Very cool. Yeah.

And I have a daughter as well, but she's only one, so eventually. She hasn't quite started. Who knows what AI will be doing for her when she's this age.

Yeah. Crazy. Good stuff.

Anyway, Benji, you're the co-founder and CEO of a company called Faze. We're going to get into Faze in a second, but beyond having a young son and daughter, tell us more about how you came to co-found and become CEO over here at Faze. Well, thanks.

Yeah. So, most importantly are my son and young daughter, but, in my spare time, I co-founded Faze with a friend that I've been working with for over 10 years, and before that, we've been friends for almost 20 years at this point. We were the co-founders of another company.

We started our first business in 2016. That was called Parsec. Mm-hmm.

Parsec was a remote desktop product for people to play video games. Okay. So we were building a game streaming application for cloud gaming when people wanted, that was a newer technology.

And during COVID, it pivoted into remote desktop for companies, and then it was sold in 2021. So, I have a lot of experience. Faze is a remote desktop product as well.

But I have a lot of experience with that and founding companies, and then in between, I worked for the FYI error, and now we're back. Very cool. Indeed.

Good for you. Thanks. So it sounds like you took what you had done in your prior comp, Parsec was the prior company?

It was, yeah. It sounds like you took Parsec, but then as part of that pivot, you recognized remote desktop streaming was the ticket, right? Yeah.

To doing things and you're doing not similar, but underlying that technology for a new time and a new era, let's call it, right? Yeah. We're bringing that experience from building the game streaming technology and refocusing it on enterprise use cases of remote access.

And we started the company in 2025 when- Mm-hmm ... I had just been working at Andreessen Horowitz, which is a venture capital firm, and we were seeing a lot- I never heard of them. I'm kidding.

Well, just in case the audience doesn't know, but- Mm-hmm ... so it's one of the venture capital firms in San Francisco, and one of the things we kept seeing was the early computer use, AI, models. And so we started thinking that not only do people need a remote access technology for their employees to be able to use workstations, they also will potentially need a remote access technology for computer use agents.

And, I'll be honest, I didn't expect it to happen as quickly as it has. Astra this past weekend demonstrates the computer use models are here and really, really good. But that is one of the reasons we founded the company was in this new era, how do you provide remote access technology to expensive hardware, from anywhere to people and to agents as well?

Sure. So I had a chance to play with Astra this weekend. Same.

And I got to tell you, I immediately noticed the difference because I was mid-project and you know how in the interface down on the bottom, you pick your model and all of a sudden Astra was there. " Boom, I hit it. Yeah.

" Yeah. Right? I mean, it's pretty amazing.

Pretty amazing what they did with the RL environment training, especially with Blender and 3D graphics and computer design work. It's very clear that was a big component of the training because as you saw all weekend on Twitter, people were building games, building 3D scenes, all sorts of really neat things. I got to tell you, I was in the middle of doing a report on the cyber industry post-Mythos and Hugging Face, and I had assembled...

So my creative process is I already know what I want to write. I just need you to go do the research, give me the background, give me the bullets, right? To shoot here.

And, so I had a list of about 30 security companies, some public, some private, and I wanted to match up their stock prices and venture raises to dates of corresponding, right? So was there a big jump after Mythos or Hugging Face or these kinds of things, and look for patterns that we can talk about. And so, what are people betting on in cyber going forward?

Because, and Benji, it blew me away. It created a chart. Yeah.

And then strong bet, weak bet, trouble. I mean, I was amazed, and I work probably much like you, I'm in AI all day, right? Yeah.

I use Claude, I use Chat, I use Perplexity Compute. We have some open core stuff here, all kinds of stuff. And yeah, I'm not saying it's AGI because I don't know enough to tell you, but I will tell you it is definitely a different cat.

Well, Jensen Huang will tell us that it's AGI. Well, he did, and Jensen said it did, and of course, what's his name, the president of OpenAI, Sam Altman said so. But as I said, I'm staying out of that fight.

Yeah. Let's get back to Faze though, Benji. So what you're doing from an enterprise point of view is you're letting everyone in the enterprise sort of have the benefit of the higher level, advanced sort of functioning, even though it's not on their local computer, and it may not necessarily be even up in the cloud.

Yeah. It could be anywhere. Yeah.

So one of the companies we were talking to recently told us that they have 30,000 workstations, and that's one workstation per engineer. And so what they are seeing is that those workstations are not being used all the time because the engineers also have a laptop. So what they are hoping for is that when they need the workstation, they can remote into an on-prem or a one U or 2U racked workstation in a data center or on-prem, and then they can have 10,000 workstations rather than 30,000, and everyone can have a laptop.

So, the number that matters to them is the utilization rate of those workstations because people weren't thinking about those workstations as infrastructure ... a few years ago because they were a typical 3 to $5,000 expense. Now, if you go onto some of the OEM websites, those are looking at, you're looking at like $15,000 per workstation.

That's an expensive asset that cannot sit idle. Absolutely. It's silly too.

So Benji, I'm a little older than you, right? I go back to my early days where they had wise terminals and Wang Computer, and DEC, and all those things, and that's kind of what it was, right? People had relatively dumb terminals they worked on, right?

Literally, we called them dumb terminals, and the brain sat somewhere else, so to speak. And this is, now the brain is an AI brain perhaps, but nevertheless, it's still a very similar kind of model, in terms of that. And as you said, it also makes sense because we have so much computing power sitting idle and idle work is the devil's work as they say, right?

Yeah. Well, we think that that era is back because the constraint is again compute. Back with wise terminals, the constraint was compute, and today we think it is too.

The constraint is not bandwidth anymore. It's none of those things. It's really the, do I have enough compute to get a 3D model built, or does my agent have enough compute to be able to get the tokens it needs?

Sure. Sure. And where does the cloud play in this, Benji?

For us, we believe that a lot of the work will be done on machines that are in your data center on-prem, because there's a lot of value in owning that hardware. But you can also access cloud machines too, that phase works. We don't care what type of hardware you are accessing, and so for our company, it could be in the cloud or it could be a workstation on-prem.

Generally, for all-day use the dollars tend to work out that it makes more sense to have that and host it on-prem, especially if you have some of the tools that we're excited to provide to our customers, like out-of-band management, which then turns the workstation into infrastructure and you can virtually manage it like you would a cloud machine. But yeah, it could be anything. Got it.

But back to this theme of returning back to the future, right, with wise and dumb terminals and what have you. The point is, we collectively all have a lot of unused compute capacity that if we had to... If it's in the cloud, fine.

Shame on us for having idle resources in the cloud. But in the cloud, the nature of the cloud is if you're not using it, you don't necessarily pay for it. You make pay because you have an instance up there.

Yeah. But you're not churning through bandwidth and CPU if it's idle. Because the cloud charges you on bandwidth and CPU.

Where if it's your internal compute or internally owned resources, you don't care if, you don't want to burn it out, but if you could run that thing at 75% all day, well, that's making the most efficient use of your resource. Definitely, and especially if you can move it into a server room where you can have better cooling and power management as well. Sure.

So there's a lot of benefits to it. And we believe that computers, as we think of them, thick clients are going to be optional for most people. And people will still love to have those computers, but because of where technology is going, you're going to use these powerful computers that are generally racked, in our opinion, and then you'll temporarily access them when you need them for different workloads.

Look, for me, it's all about the monitor anyway. I love a good monitor. Yeah.

But that being said, again, a little bit of a throwback back to racks. But what's interesting is, I was telling you, I was just finishing a special report on data centers, is when I was in the data center business 25, 28 years ago, my guys used to walk around with tool belts with screwdrivers because we were racking up, literally racking up servers. Now, you in essence can buy a rack, right?

If AI's your thing, and that's where we're headed to when we think of, well, what's the basic unit for compute? Well, it's not just the chip, then it became the system on the chip, the computer on the chip, then the computer. But now we're talking about a rack at a time, right?

You buy a rack. And maybe that's the future unit. Yeah.

There's also the GB 300, which is a desk side compute or GB 10. Yeah. So there's different models that people are trying out right now.

Look, you haven't mentioned Macs either, right? Those don't get racked so easy. You need like...

So back in the day when I was in hosting, not everyone had rack mounted servers, and I helped put together a roll-up. We did about 30 acquisitions in three years. Some of the data centers I worked in, besides that they had carpeting, which I never could understand why someone would put carpeting in a data center, but anyway they had what I used to call baker's racks, right?

And it was like what baker's would do to put bread on, and they had computers, regular tower computers lined up like that. It was insane. But I don't think we're returning to that per se, but you might have Macs, a closet full of Mac Mini or Mac Studios, and dollar for dollar, it's good computing.

Yeah, especially with the shared memory, for sure. Yeah. Absolutely.

If you can buy enough of them, right? They're kind of hard to come by right now. Everything is at this point.

Yes. But again, that plays right back into Phase's business model, right? Because even what you already have is too valuable not to be put to good use.

Exactly. That's our hypothesis. So where are you now in go-to-market on this?

Yeah, we are working with our initial customers. Mm-hmm. We released the first version last week, and customers are onboarding and getting going.

Great. So it's a GA or it's still sort of a closed release at this point? It's a closed release.

You can't sign up to pay us online right now, but our customers that we're working with are re... We're doing direct sales at the moment. Right.

In a few weeks, you'll be able to just start using it from our website. You can download the app today and use it for free on our website. And then beyond that, if you want to start using our enterprise products, you have to get in contact with us.

Do you think your customers are, or at least initially in this first phase, people who already have compute in their own data centers, in their own closets, or do you think we're about to see a wave of people who are saying, "You know what? I'm tired of token maxing. I'm tired of FinOps with the cloud.

" Yeah. Well, I think that the first customers are actually people who have workstations at every desk. Then it's also a customer, and the customers want to move those workstations from every sitting at your desk to moving it to a server room, primarily for the benefits of utilization rates, as well as power, being able to cool it more efficiently.

Then finally, with the newest GPUs, you're starting to see the need for 220 volt plugs, which in the US at least, is not at every desk. So- Mm ... putting that into a server room, it gives you the capabilities to properly power the workstations.

From there, people are racking workstations, and I think that that's the right strategy for them. And then, in addition, you will see, in my opinion, you'll see more and more people routing their lower intelligence requests to some of the open weight models that are hosted on-prem. And then, when you need a higher intelligence with better RL, sorry, better runtime reinforcement learning, you would then go to a soda model in the cloud kind of thing.

And that's going to be the most efficient way to route your requests, most likely. I love it. Benji, we're at the, supposed to be a 15-minute interview.

We blew through that a while ago. For people who want to get more information on Phase, it's P, it's Phase just the way it is on your shirt, just the way people see it on the underlined your picture. It's P-H-A-Z-E, but what's the website?

app. app. app, yes.

I love it. I haven't seen many of those yet. Thanks.

com is taken. Yeah, I know. coms, my friend.

It's hard. Anyway, Benji, I wish you a lot of success with Phase. Come back on and keep us posted.

I'd love to continue talking about this. Thank you very much for your time. It was great to be here and excited to meet folks in your audience.

Absolutely. Benji Boxer, co-founder, CEO at Phase. Hey, you might have some idle compute sitting at your place that you can make use of.

app. We're going to take a break. We'll be back with more here in a minute.