Ubisoft's first artificial intelligence presentation was all about conversation. The second option, Companions, focuses on combat and features a group of three individuals trained by their own narrative team for you to join in battle with.
It is the product of an expanded internal team (now up to 80 people) developing proprietary middleware that connects a wide range of external AI models to the company's Snowdrop and Anvil game engines and helps internal teams build them into their games. At the same time, they are developing their own “game experiences” (the team is careful not to call them “demos”) that highlight potential applications. The company allowed me to team up with teammates at an event in Paris this week and said it was being tested by both internal studios and external testers, touting reviews from “hundreds” of users about its product. Ubisoft Connect program.
The teammates are a noticeable change from the Neo NPC team's debut at GDC 2024, which was based on conversations with AI-NPCs that unlocked RPG-style relationship effects.. According to Manzanares, it was received favorably but reticently by internal teams who didn't see it fitting with the types of games they were making. In response, the teammates seem like something that could easily carry the Tom Clancy prefix.
It's based around a pair of AI cyborg soldiers that you can command in a few simple skirmishes against small squads of bullet-sponge robots, led by a third AI companion who has GM-like control over the game. The soldiers, Pablo and Sofia, each have their own personalities (stoic and playful respectively, the demo offers several different options for both) and backstories that can be revealed through conversation during the game. Companion, Jaspar, delivers GM-like insights with Ryan Reynolds-style cunning. All this runs on the Google Gemini platform.
At the start of the mission, you have teammates but no weapons, and you need to guide them from cover for their first encounters. Using voice commands brings back unpleasant memories of the late and unloved Kinect for Xbox, but it works in a way that Kinect has rarely managed – I could give quick instructions like “stand on these panels to open the door” or “flank the soldiers and stay behind cover,” and Pablo and Sofia would do so with varying degrees of sass depending on which personality was assigned.
Both periodically offered attack plans or abilities, such as the ability to concentrate fire, using the so-called initiative system. This became increasingly useful as the level progressed, because even if a weapon was provided, it was particularly useless and took time for each enemy unit to grind it down. A failed tactic will result in Pablo or Sofia collapsing and requiring resuscitation, and you can request that they do the same to you.
Beyond just bossing them around, choosing to divert from the conversation, we discovered different backstories for each of them; I managed to tease Pablo about the fact that the previous squad had been destroyed, and Sofia cheerfully glossed over the practicality of the cyborgs' dining habits. I didn't want to waste my demo time painstakingly searching them for knowledge, but by a stroke of luck I discovered them, the result of what the narrative development team says was careful preparation, instruction, and training.
Meanwhile, Jaspar was truly annoying: smug and at times demeaning, although he seemed responsive to requests to be less tactful. He had more control over the game – I could ask him to highlight soldiers on the HUD, change the color of the HUD, or tell me what I just picked up so I didn't have to go into the D-pad menu and figure it out.
It's also responsible for giving out achievements, about ten in number, as in the demo the unlock was unlocked when the AI thought the requirement had been met rather than when the player met fixed criteria. It “sounds simple,” says Manzanares, “but it’s quite complex. We have generative AI that produces content that is not just conversation, but a game system – like a complex quest system. How can an LLM play the role of a live producer? That's what we're working on.” It also points to significant accessibility potential, with a range of applications ranging from customizing gameplay for players with disabilities to reminding them what they should do after returning to a long-lost save.
I didn't enjoy spending time with Jaspar's witty one-liners (apparently testers find him better company), but that was because of his character, and it truly represents the project's success: Narrative Director Virginie Mosser was delighted when I called him “sassy” because that was a key characteristic of what the team calls a “character sheet” for each teammate.
“We want to create characters,” she says. “The worst thing is when someone says, 'I don't like him, I don't hate him.' I want the player who says “I hate him” because then that's the feeling. They need to provoke something.” She is keen to highlight the amount of work this requires from the creative team at a time when there is growing concern in the industry about creatives losing their jobs to generative artificial intelligence.
“We have to work hard to prepare, almost as if they were actors,” she says. “It takes a lot of time to instill in them the rules of improvisation. Whatever you write in your backstory will affect your future play.” The key element here is creating a “character sheet” that describes the NPC's backstory and personality, which is then entered into a tooltip and then tweaked, tested and iterated, along with working with the voice actors. “It's a lot of creative work, and that's good news for us,” Mosser says, “because it's good news for the creators.”
Manzanares believes this will be a key differentiator for Ubisoft in a potential future where AI NPCs become ubiquitous. “Character panels and tooltips are a really, really big deal for the team,” he says. “This is where we put a lot of effort into the tools and how to write with them. Because in the future, that’s where they’re going to make a difference versus the general model that maybe everyone has.”
Teammates is just one result of the team's ongoing efforts to create a central platform that will allow Ubisoft developers to incorporate artificial intelligence tools into their games. “We have to first create middleware that connects to the engines so that these teams don’t have to deal with that,” Manzanares says. “Anything related to the AI generation – the tools, the models, what's happening with Gemini, whatever the new announcement is – they don't have to worry about that. We'll take care of it. Our goal is to connect and work so they can focus on their game.”
In the meantime, the company will continue to develop prototypes like Teammate to demonstrate the capabilities. “We have to continue to create small experiences, things that are playable,” says Manzanares, “so that creative directors and maybe even players can start checking them out and maybe take them further.” The similarities of the teammates to existing Ubisoft products are expected to help with this. “It's a first-person shooter and the squads and even the settings are familiar,” says Remy Laborie, director of data and artificial intelligence, “and I think that speaks to the team. This is their vocabulary. It helps to think about how to create new gameplay cubes with AI if you already have a vocabulary for the game you're trying to work with.”
The R&D team's focus is on rapid iteration, with the new build in players' hands in October and the roadmap calling for monthly updates through January. “We'll add more features to Jaspar so you can do more,” says Manzanares. “We will add an emotion system: if you talk badly to Pablo, he will not be as happy, we will change emotions in real time.” We are expected to test, learn and respond to user feedback, as well as implement any changes that come with new AI models, as the platform means they can be swapped in and out as needed.
Ubisoft is expected to develop its own AI modes, but none have been named or mentioned; The team will not be distracted by plans or costs, preferring to focus on the benefits of choosing from currently available external models. Given the significant financial and environmental costs associated with these models, this reckoning will come at some point, but for now the priority is to identify potential use cases.
The initiative, as presented, appears to be much more creator-centric and more voluntary than the stated AI mandates from Crafton and EA. Ubisoft's approach is to create a central resource and make it available to internal game development teams to find a use for it (the team hopes they'll be enticed by “gaming capabilities” like Teammate). While development of the tool is progressing quickly and internal reception is described as positive, it will take some time to implement it into shipping games.
“A lot of internal people are asking, 'Okay, does this fit my vision for the game? I need to think about it.” Some people are more like, “I need to ship soon, so I’ll come back to you next year,” says Reynaud Francois, the project’s creative director. “A lot of people see potential – I don't know if that potential matches their vision of the game today. But maybe the next one.”
