Benjy Boxer, CEO and Co-Founder of Phaze – Interview Series

Benjy Boxer, CEO and Co-Founder of Phaze, is a technology entrepreneur building a high-performance remote desktop platform for engineers, designers, and other professionals working with demanding 3D and CAD applications. Phaze delivers low-latency remote access to workstations with enterprise-grade management and security controls, enabling teams to collaborate seamlessly from anywhere. He previously co-founded the low-latency remote desktop company Parsec in 2016 and led it through its $320 million acquisition by Unity in 2021. Following the acquisition, he held several senior leadership positions at Unity, including Senior Vice President and General Manager of Unity Cloud, where he oversaw products supporting remote collaboration, multiplayer development, asset management, cloud infrastructure, and real-time 3D content creation. Boxer previously worked in product strategy at NewsCred, corporate development at AOL, and asset and liability management at the World Bank.

Phaze is an enterprise remote desktop platform designed for professionals working with demanding 3D, computer-aided design, engineering, and visualization applications. It provides secure access to high-performance workstations from virtually anywhere, with low-latency controls, high frame rates, accurate color reproduction, multi-monitor functionality, USB passthrough, and real-time session collaboration. For IT teams, Phaze offers centralized management of devices, users, access permissions, on-premises relays, connection activity, and enterprise authentication, supported by SOC 2 Type II certification, independent penetration testing, role-based access controls, and comprehensive audit logs.

You founded Parsec around the idea that if remote desktop could meet the demands of gaming, it could eventually work for almost anything. After selling Parsec to Unity, what made you decide to return to this problem with Phaze, and what did you learn from Parsec that shaped Phaze from day one?

At Parsec, we learned that people don’t forgive bad performance. Gamers notice every dropped frame and every millisecond of latency, so we had to build a platform that disappeared into the background. If users are thinking about the remote desktop, you’ve already lost. Parsec became popular in game studios because that performance matters to creators too.

Now, we believe that access to high-performance compute is a requirement for every professional. That’s because every piece of software relies on GPU acceleration to deliver features to customers. 3D graphics performance has always mattered to professional engineers, creators, developers, architects, animators, and editors. Phaze is for those workstation users and the enterprises they work for. Because of the power demands of the GPUs, and frankly the cost, powerful hardware increasingly lives somewhere other than under a desk.

We built Phaze because we saw the same problem emerging in a much larger market. Companies are centralizing compute, but employees still expect their tools to feel local. Serving these enterprise customers requires us to focus not only on the remote desktop, but also the enterprise administration and products that support IT teams at large companies that are adjusting their strategies for our new world.

Phaze is built around the idea that workstations are becoming infrastructure. What does that shift look like inside enterprises today?

The workstation is following the same path as the server. Companies used to buy a powerful machine for every engineer or designer because that was the only practical option. AI changed the economics behind this. High-end GPU systems are expensive to buy, expensive to power, and expensive to cool.

Instead of buying one workstation for every employee, companies are putting those machines into server rooms, colocation facilities, and cloud environments. Making this shift improves asset utilization, data security, and hardware management. The hardware stays in one place, while employees access it from wherever they happen to be working.

Why are traditional desktop workstations no longer enough for teams working with AI, real-time rendering, simulation, CAD, 3D, and visual computing?

It’s not about traditional desktop workstations failing people. The goalposts move every time a new generation of GPUs becomes available. Whenever there’s more VRAM, ISVs use it to generate more realistic pixels, better simulations, and smarter AI. The workloads are changing, so companies need to upgrade their hardware. Training AI models, rendering complex scenes, running simulations, and working with modern visual computing software all demand far more GPU power than they did even a few years ago. Modern professional workstations can draw well over 1,200 watts for graphics processing alone, which makes them difficult to deploy at scale across an organization. Companies still need that performance. They’re just delivering it differently by centralizing the hardware and letting employees connect to it remotely.

A lot of AI infrastructure coverage focuses on models, GPUs, data centers, and inference. Where does remote desktop access fit into the AI infrastructure stack?

These focus on the token generation that becomes an input into how humans interact with software on computers. We think that more of the computers people interact with will live elsewhere. That means a creator, engineer, designer, or developer will have the ability to access a high-performance computer on any screen. The remote desktop is the access layer between users and the computers they’re relying on. Graphics and latency still matter when humans interact with their computer, especially when working in CAD or 3D. Phaze prioritizes performance for those use cases and gives everyone access to the compute when they need it no matter where they are.

Phaze talks about both employees and AI agents needing access to powerful machines. How does the definition of a “user” change when AI agents begin accessing computers alongside people?

At work, people are trying to accomplish goals with computers whenever they use them. Now, a human can interact with the computer to accomplish that goal, or they can employ an agent to do the same on their behalf. Either way, the outcome is the same, it’s just a different user. We think companies need to treat agents just like employees and manage and secure their access accordingly.

What kinds of tasks will AI agents need remote workstation access for that cannot be handled cleanly through APIs alone?

There’s a debate brewing between computer use models versus CLI/API-first design. I think most software providers will deliver an interface for AI agents to consume. But there are scenarios when the goal is to test how a human would use software or when the job-to-be-done requires interacting with a full computer. In those two scenarios, a computer use agent will be the best solution. So, if you’re QAing a new application for customers or you’re trying to resolve an end user’s issue on their computer, I think you’ll need a computer use model.

Remote access accelerates accomplishing those goals. Sometimes deploying agents to end user devices is cumbersome or requires a lot of technical knowledge. There are also scenarios where having a central management of agents that are connected to many computers is advantageous. Additionally, people may not want to run a computer use model on their primary device, so they can request tasks to be accomplished while they’re focusing on other items on their computer.

How should enterprises think about the security risks of giving AI agents access to full machines with files, applications, permissions, and compute resources?

I don’t believe there’s a significant increase in risk relative to giving employees the same access. Companies have a lot of experience ensuring computers are siloed, secure, and only have access to the files that they should. I suspect computer use agents will inherit the same permissions as the user who requests the agent to do something. There’s a benefit as well in having computer use agents operate computers via remote access because then the agent isn’t installed directly on the computer it’s controlling. A human can watch the agent taking action through the remote desktop product, or the scope of inputs it can control can be managed through the remote desktop protocols.

Remote desktop has existed for decades, but many users still associate it with lag, poor visuals, or clunky IT workflows. What has changed technically that makes this category more relevant now?

For a long time, remote desktop deserved that reputation. Things have changed around the technologies remote desktop protocols use. The internet has gotten much better. Nearly everyone has a high bandwidth connection at home. And now, if your airline makes a deal with Starlink, you can have a high bandwidth connection in the air too. Additionally, hardware was released that includes chips specifically designed to decode and encode video streams quickly. This combination, along with technical innovations, makes the past experiences unacceptable today.

People accepted lag, blurry visuals, and unreliable performance because there wasn’t a better alternative. Building Parsec forced us to solve those problems because gamers simply wouldn’t tolerate them. If the experience wasn’t seamless, they stopped using it.

Today those same expectations exist across professional computing. Engineers, designers, developers, and creators all expect remote access to feel like they’re sitting in front of the machine. As companies move workstations into shared infrastructure, that level of performance isn’t optional anymore.

Which industries do you think will adopt remote AI-ready workstations first: gaming, architecture, engineering, film, manufacturing, robotics, AI labs, or something else?

The first wave is already pretty clear. It’s any industry where people spend their day working with GPU-intensive software.

That includes engineering, architecture, manufacturing, visual effects, film, game development, simulation, and AI research. These organizations already depend on expensive hardware, so they see immediate value in centralizing it instead of distributing machines to each person. People who work with the most intense simulations, visual scenes, design the most complicated products and buildings all rely on the latest GPU-powered workstations.

Over time, that group will expand as AI becomes part of more professional software. More employees will need occasional access to powerful compute even if they don’t need a dedicated workstation.

Looking five years ahead, do you think the enterprise workstation will still be a personal device, or will it become a shared pool of GPU-powered infrastructure accessed by both humans and AI agents?

I think we’ll look back at dedicated workstations the same way we look at dedicated servers.

There will always be exceptions, but for most organizations, it makes more sense to build shared pools of high-performance compute and let people connect to whatever resources they need. That’s a better use of expensive hardware, and it gives organizations far more flexibility as workloads continue to evolve. We’d like to see a world where people have access to the compute they need no matter where they are and without thinking about how often they need to upgrade their hardware. We think that’s possible if the way you access the hardware is good enough to disappear into the background and not remind you that you’re remotely accessing the compute. 

The workstation isn’t disappearing. It’s moving into the data center. The real challenge is making that transition invisible to the people using it. That’s what we’re building.

Thank you for the great interview, readers who wish to learn more should visit Phaze.