Virtual reality has always promised more natural ways for players to interact with games, but talking to AI-controlled coworkers while trying to keep a dimension-hopping fast-food restaurant running sounds like a particularly chaotic way to put that idea to the test. During Gamescom, we met with Owlchemy Labs and Artificial Agency to try out the upcoming single-player mode for Dimensional Double Shift and see how its T.E.M.P. coworkers handle the unpredictable nature of human communication. Rather than using generative AI to create dialogue on the fly, the two studios have taken a more controlled approach, combining hand-crafted content with AI-driven interpretation and behavior. With the mode still in development ahead of its planned 2027 release, we spoke with Owlchemy Labs CEOwl Andrew Eiche and Artificial Agency co-founder and CEO Brian Tanner about how the system works, what the team learned from Gamescom players, and whether a robotic coworker can capture some of the chaos of playing alongside another human.
Why Dimensional Double Shift?
The new single-player mode relies on natural language comprehension and behavioral decision-making rather than generative content. Why is Dimensional Double Shift such a good fit for that approach?
Andrew Eiche, Owlchemy Labs: We built Dimensional Double Shift as a game that is “better together”. We wanted to lean hard into playing with other people. Communication is a core part of the experience. When we set out to make single player, we wanted an experience that felt as similar to the multiplayer experience as possible. People love the game and the mess that happens when people try to collaborate in a chaotic setting. Using AI is a natural fit. Language is an organic thing. Every statement contains layers of meaning. Large Language Models are our current best effort at making sense of how we speak to each other.
Brian Tanner, Artificial Agency: You shouldn’t have to stop a chaotic shift to remember how to talk to your robot coworker. The words are only part of the picture. The Behavior Engine combines what the player says with what’s happening in the game and the T.E.M.P.’s own role. That context helps it interpret an incomplete or imprecise request, rather than requiring the player to use a particular keyword or carefully constructed command. We’re trying to make talking to a T.E.M.P. part of the teamwork, not another interface you have to learn.
Putting the T.E.M.P.s to the test
You tested the mode with players behind closed doors at Gamescom. What did those sessions teach the team?
Andrew Eiche: It confirmed that we’re not done yet! The mode still needs a lot of polish before it’s time to ship, but it is already off to a great start.
One interesting thing we saw during Gamescom is that player preferences don’t necessarily need additional work. For instance, we had a player ask if the T.E.M.P.s could speak less. Typically, this would require a new menu option or some other control. We told the player to ask the T.E.M.P.s to speak less, and they did! The agents made a note of the request and honored it throughout the session.
That ability to understand context also seems important when it comes to deciding whether a player is actually asking for something. How does the system distinguish a request from someone simply thinking out loud or joking around?
Brian Tanner: The unique thing about what we’ve built is that the agents don’t technically need to wait for clearer instruction at all. A T.E.M.P. understands relevant game state, its role and objectives, and the boundaries it operates within, giving it the context to distinguish a request from a passing comment. It’s similar to playing with another human teammate: you can tell when they’re instructing you to do something, but you don’t stop playing or making decisions when they aren’t giving you instructions.
Keeping the humans in control
That sounds like it also gives Owlchemy plenty of room to let its own sense of humor come through. How do you write fixed, human-created dialogue for a system that has to react to unpredictable player behavior?
Andrew Eiche: We’ve been working on landing humor in a fully interactive environment since Job Simulator. AI is not too much different from allowing a player to do whatever they want with their hands. We have a rough hierarchy for what types of jokes work best in VR. The jokes players make themselves are typically the most funny. A great deal of our work is to create an environment where the player is comfortable laughing. From there we let them explore and play. We write and design to react to the player’s intentions.
This manifests in letting the player mess with the T.E.M.P.s and having lines to acknowledge that the player and the T.E.M.P.s may not always be working collaboratively. The goal is not to cover every possible action a player could conceive of, but to make sure they feel seen as they play.
Brian Tanner: Our team likes to get the T.E.M.P.s to eat weird things, start food fights, and ask them existential questions about the universe. Not necessarily the most productive shift, but that’s part of the point.
The opportunity for developers is to give characters personalities and attitudes, then let those qualities show up in situations they haven’t individually scripted. The creative direction still comes from the developer. The Behavior Engine gives those characters more ways to express it through their behavior: helping, pushing back, goofing off, or getting pulled into a game within the game.
For Owlchemy, that unlocks a single-player mode with some of the social energy of multiplayer. You have teammates to coordinate with, but also someone to mess around with. It’s not just about whether you finished the shift. It’s about what happened along the way.
Keeping up with the chaos
All of that has to happen quickly, though. How are you keeping the system responsive when a player is issuing commands in the middle of a hectic shift?
Brian Tanner: In a game like this, responsiveness is part of the gameplay, so you can’t send every decision through the same reasoning loop. We use a fast, game-specific model in the cloud alongside our on-headset Reflex Architecture, which separates decisions requiring deeper reasoning from immediate reactions that need to happen quickly. We also specialize the intelligence around Dimensional Double Shift’s constraints rather than using a general-purpose model. That combination is designed to reduce latency while still giving the T.E.M.P.s enough reasoning capability to handle messy, fast-changing shifts.
Andrew Eiche: We’re perpetually tweaking response times. Humans have a ton of small non-verbal and non-linguistic ways to give them time to think and respond when spoken to. We have to add all of that into our NPCs manually. It’s a bit of a meditative experience to observe the moments between words and attempt to recreate them synthetically.
Making XR easier to access
Single Player Mode is arriving alongside controller support and a Steam Frame version. How do those additions fit into Owlchemy’s broader vision for the future of XR?
Andrew Eiche: We are always thinking about how XR becomes the mainstream medium we know it can be. Supporting multiple ways to play our game and bringing accessibility in at the early stages of development are ways we can invite a broad set of players to dig into Dimensional Double Shift and hopefully become lifelong users of XR.
We’re still in the early days of XR. There is a generation that is already using it daily and, as the technology matures, we want it to be as accessible as possible!
That seems to tie back nicely to the original goal of making the single-player experience feel as close to multiplayer as possible, even when the player isn’t actually playing with friends. What does that mean in practice?
Andrew Eiche: If someone wants to play Dimensional Double Shift, we want to make it as easy as possible for them to jump in. That might mean playing solo because your friends aren’t around, using a controller, or picking up the game on a new platform like Steam Frame. We want people to be able to get into DDS in whatever way works best for them.