Global gaming giant Ubisoft unveiled its latest AI game research project “Teammates” in November 2025, following the Neo NPC prototype demonstrated at GDC 2024—another significant experiment in generative AI game applications. Teammates is a first-person shooter game prototype where players can control AI teammates Pablo and Sofia through natural voice commands. These two AI NPCs can understand tone, intent, and environmental context, dynamically generating real-time responses and actions, showcasing how generative AI can fundamentally transform NPC interaction experiences in games.
Teammates Game Concept
Dystopian FPS Setting
Game Background
Teammates is set in a dystopian future world where players assume the role of resistance organization members tasked with infiltrating enemy bases to locate 5 missing teammates.
The game employs first-person shooter (FPS) mechanics, combining stealth, combat, and puzzle-solving elements.
Core Gameplay Experience
Unlike traditional FPS games, Teammates’ core isn’t player’s individual shooting skills, but how to collaborate with AI teammates.
Players need to:
- Issue tactical commands via voice
- Observe environment and formulate strategies
- Adjust plans based on teammate feedback
- Build trust relationships with AI teammates
Game Length
Current prototype version gameplay lasts approximately 30-45 minutes, representing a short experimental demo rather than full commercial game.
AI Teammates: Pablo and Sofia
Jaspar: AI Assistant
First AI players encounter is Jaspar, an AI assistant residing in player’s communication device.
Jaspar serves as mission briefer and tactical advisor:
- Provides mission objectives and intelligence
- Analyzes environment and enemy situations
- Suggests action plans
- Answers player questions
Pablo: Combat Specialist
Pablo is first human AI teammate players meet, positioned as combat specialist:
- Excels at frontal firepower suppression
- Can execute offensive tactics
- Straightforward personality, action-oriented
- Proactively offers suggestions based on combat situations
Sofia: Stealth Specialist
Sofia is second AI teammate, specializing in stealth and flanking operations:
- Excels at flanking and reconnaissance
- Can execute precision assassinations
- Cautious personality, strategic thinking
- Warns players of potential dangers
Dynamic AI Personalities
Pablo and Sofia aren’t just command-executing robots, but AI possessing “personalities”:
- Express opinions on player decisions (agree or question)
- Change emotions based on mission progress (nervous, confident, frustrated)
- Remember player’s command style and adapt
- Exhibit natural interaction similar to human teammates
Voice Control Core Mechanics
Teammates’ greatest innovation lies in completely natural voice-based command system.
Natural Language Understanding
No Memorizing Commands Required
Players don’t need to memorize fixed voice commands (like traditional voice games’ “attack,” “defend”), but can communicate with AI teammates using natural conversation:
- “Pablo, can you flank from the left?”
- “Sofia, check what’s in that room”
- “Let’s split up, meet at stairs in five minutes”
- “How’s the situation now?”
Tone and Intent Recognition
AI not only understands literal meaning but also interprets tone and intent:
- Urgent tone → AI judges situation critical, accelerates action speed
- Hesitant tone → AI may proactively offer suggestions
- Commanding tone → AI executes directly
- Questioning tone → AI first explains then asks for confirmation
Environmental Context Awareness
AI understands commands based on game environmental context:
- Player says “clear there,” AI judges target based on line of sight
- “Cover me” meaning changes with current tactical context
- AI understands relative positions (“left,” “behind,” “upstairs”)
Bidirectional Dialogue System
AI Proactive Communication
Not just player commands, AI teammates proactively communicate:
- “I see enemies, should I fire?”
- “Running low on ammo, need to conserve”
- “This room’s clear, what’s next?”
- “Sounds like activity ahead, be careful”
Request Confirmation and Clarification
AI proactively inquires when uncertain about understanding:
- Player: “Go there” → AI: “You mean the left corridor or right room?”
- Player: “Ready?” → AI: “Ready to move anytime”
Building Team Sense
This bidirectional communication creates genuine team cooperation feeling:
- Player isn’t commanding officer but team member
- AI teammates provide feedback and suggestions like real players
- Forms natural tactical discussion and decision-making process
Technical Implementation
Google Gemini Integration
Teammates’ AI system powered by Google Gemini:
- Gemini’s multimodal understanding capability processes voice input
- Natural Language Processing (NLP) parses player commands
- Contextual memory maintains conversation coherence
- Real-time generation of contextually appropriate responses
Ubisoft In-House Middleware
Between Gemini and game engine lies Ubisoft-developed middleware:
- Converts AI-generated semantics into in-game actions
- Manages AI decision integration with game physics, collision
- Ensures AI behavior complies with game rules and balance
- Optimizes response speed, reduces latency
Snowdrop Engine
Game developed using Ubisoft’s proprietary Snowdrop engine:
- High-quality graphics rendering
- Dynamic lighting and particle effects
- Physics system supporting environmental interaction
- Deep integration with AI systems
Voice Recognition Challenges
Multilingual Support
Currently Teammates primarily supports English, but Ubisoft plans multilingual expansion:
- Semantic understanding complexity differences across languages
- Accent and dialect recognition accuracy issues
- Cultural context impact on command understanding
Environmental Noise
Actual gameplay may face:
- Game sound effects interfering with voice recognition
- Home environment background noise
- Need to calibrate microphone sensitivity
Response Latency
AI understanding and response generation requires time:
- Ideal latency < 1 second, avoiding disrupting gameplay fluidity
- Need to balance cloud AI processing and local inference
- Network connection quality impacts experience
Relationship with Neo NPC
Teammates represents continuation of Ubisoft’s generative AI game research.
Neo NPC Prototype (2024 GDC)
Static Environment Demonstration
At GDC (Game Developers Conference) 2024, Ubisoft demonstrated Neo NPC prototype:
- NPCs possessed novel cognitive and natural language capabilities
- Could engage in natural conversation with players
- Remembered player behavior and adjusted reactions
- But NPCs existed in static environment, unable to execute complex actions
Proof of Concept
Neo NPC primarily served as proof of concept:
- Demonstrated LLM-driven NPC dialogue feasibility
- Tested player acceptance of AI NPCs
- Explored technical limitations and optimization directions
Teammates’ Evolution
From Static to Dynamic
Teammates extends Neo NPC’s dialogue capabilities to dynamic game environment:
- NPCs not just converse but execute complex actions (movement, shooting, cover)
- AI understands game environment and tactical context
- Real-time response to player commands and environmental changes
From Demonstration to Game
Neo NPC was technical demonstration, Teammates is playable game prototype:
- Complete game loop (objectives, challenges, progression)
- Traditional game mechanics (FPS combat, level design)
- AI technology integrated into core gameplay experience
Scale Expansion
Teammates development scale far exceeds Neo NPC:
- 80-person development team (Neo NPC approximately 10-15 people)
- Several months development cycle
- More complex technical integration
Development Team and Technology
Development Scale
80-Person Cross-Disciplinary Team
Teammates developed by Ubisoft internal 80-person team:
- Game designers: Design core mechanics and levels
- Programmers: AI integration, game logic, engine optimization
- Art team: Environment, characters, animation
- AI researchers: Gemini integration, NLP, middleware
- Audio design: Voice synthesis, sound effects, mixing
- QA testing: Functional testing, AI behavior testing
Development Timeline
From Neo NPC to Teammates approximately 12-18 months:
- Early research and prototype development
- Mid-stage engine integration and content production
- Late-stage testing and optimization
Technology Stack
Google Gemini
- Multimodal AI model supporting text, voice, image input
- Natural language understanding and generation
- Contextual memory and reasoning capabilities
Snowdrop Engine
- Ubisoft’s proprietary next-generation game engine
- Used in The Division, Avatar: Frontiers of Pandora, and other works
- Supports high-quality graphics and complex systems
In-House AI Middleware
- Bridge between Gemini and game engine
- Semantic understanding → game command conversion
- AI decisions → NPC behavior execution
- Game state → AI context input
Voice Recognition and Synthesis
- Real-time Speech-to-Text (STT)
- Text-to-Speech (TTS) generating NPC voices
- Possibly using Google Cloud Speech API or proprietary technology
Technical Challenges
Real-Time Performance
Game requires high frame rate smooth operation while executing complex AI inference:
- Trade-off between local vs cloud AI processing
- Reducing AI response latency to acceptable range
- Avoiding AI computation affecting game frame rate
AI Reliability
LLMs may produce unpredictable outputs:
- Avoiding AI issuing dangerous or nonsensical commands
- Ensuring AI behavior complies with game rules (no clipping, no cheating)
- Handling AI “hallucination” issues
Player Experience Consistency
AI behavior differs each playthrough, potentially affecting experience:
- Balancing AI randomness and predictability
- Ensuring core narrative and game objectives achieved
- Avoiding AI excessive randomness causing player confusion
Closed Beta and Player Feedback
Teammates currently in limited closed beta phase.
Testing Scope
Invited Players
Ubisoft invited select players for testing:
- Ubisoft internal employees
- Selected community opinion leaders and content creators
- Game industry professionals
- Randomly selected players (through application system)
Testing Objectives
Collect player feedback to evaluate:
- AI voice understanding accuracy
- Player satisfaction with AI teammates
- Game mechanics’ entertainment value
- Technical stability and performance
Initial Feedback
Positive Reactions
Testers reported positive experiences:
- Extremely Immersive: Voice interaction makes players feel genuinely commanding real teammates
- Novel: Completely different experience from traditional games
- Impressive AI: Pablo and Sofia’s responses exceed expectations in naturalness
- Tremendous Potential: Seeing future gaming possibilities
Technical Issues
Also players reporting technical flaws:
- Occasional Misunderstanding: AI didn’t correctly understand commands, executed wrong actions
- Response Latency: Sometimes AI responses slow, disrupting fluidity
- Strange Behavior: AI occasionally gets stuck or performs illogical actions
- Voice Recognition: Recognition fails with heavy accents or unclear pronunciation
Gameplay Questions
Some players questioning game depth:
- Post-Novelty: AI interaction initially interesting but may lack long-term appeal
- Shallow Gameplay: Besides AI teammates, FPS itself fairly basic
- Replay Value: Fixed narrative, lacking traditional game progression systems
Industry Significance and Impact
Teammates isn’t just a game but important milestone in gaming industry AI application.
Strategic Significance for Ubisoft
AI Technology R&D
Teammates serves as testing ground for Ubisoft AI technology research:
- Accumulating generative AI game application experience
- Training AI professional talent
- Establishing AI technology infrastructure
Future Game Integration
Ubisoft may integrate Teammates technology into commercial games:
- Assassin’s Creed: AI NPCs providing more natural dialogue and quests
- Rainbow Six: AI teammates providing tactical support in single-player mode
- Watch Dogs: Dynamically generated NPC interactions enriching open world
Market Positioning
Demonstrating Ubisoft’s leadership in AI gaming field:
- Attracting investor and shareholder confidence
- Recruiting top AI talent
- Establishing industry discourse power
Implications for Gaming Industry
NPC Revolution
Teammates demonstrates NPC evolution from “scripted props” to “intelligent partners”:
- Breaking traditional game NPC dialogue tree limitations
- Achieving truly dynamic dialogue and behavior
- Pointing direction for open-world game NPC design
Voice Interaction Potential
Proves voice control can become core game mechanic:
- Not just auxiliary function but primary interaction method
- Suitable for strategy, simulation, cooperative game genres
- Natural extension for VR games
AI Game Design Paradigm
Traditional game design needs adapting to AI technology:
- Game level design must consider AI behavior unpredictability
- Balancing narrative control with AI-generated content freedom
- New QA testing methods (AI behavior difficult to exhaustively test)
Competitive Dynamics
Other Major Publishers Following
Teammates may trigger industry chain reaction:
- EA, Activision Blizzard, Take-Two following with AI game investments
- Indie developers using open-source LLMs developing similar games
- Forming AI game innovation competition
Differentiation from Traditional Games
AI games may open new market segments:
- Attracting early adopters interested in new technology
- Complementing rather than replacing traditional games
- Forming “AI Game” category
Controversies and Ethical Issues
Teammates also sparked controversial discussions about AI games.
Creative Worker Concerns
Voice Actor Unemployment Crisis
AI voice generation threatens voice acting industry:
- NPC dialogue AI-generated in real-time, no need for pre-recorded voiceovers
- Voice actor job opportunities drastically reduced
- Particularly affecting supporting and secondary NPC voice work
Writer Role Change
AI-generated dialogue diminishes traditional game writers’ role:
- Shifting from writing specific dialogue to defining AI personality and rules
- Losing precise control over narrative details
- Creative expression limited by AI model capabilities
Technology Dependency Risks
Google Dependency
Teammates highly dependent on Google Gemini:
- Gemini service interruption would render game unplayable
- Google may adjust API pricing or policies
- Technical sovereignty in external company’s hands
Online Requirements
AI games may require persistent internet connection:
- Cloud AI processing needs network
- Cannot play offline single-player
- Game permanently unplayable after server shutdown
Player Acceptance
Novelty vs Long-Term Appeal
Skeptics question AI interaction durability:
- Initially novel and interesting, but long-term may become repetitive and boring
- AI randomness may cause inconsistent experience
- Lacking traditional game careful design and polish
Preferring Traditional Design
Some players prefer traditional game design:
- Carefully written scripts surpass AI improvised dialogue
- Predictable game mechanics fairer for competition
- Doubting AI can achieve human design depth
Commercial Prospects
Teammates currently research project, commercialization plans unconfirmed.
Possible Release Models
Free Experience
Released free as technical demonstration:
- Attracts players to try, collect data
- Build brand image and market buzz
- Not pursuing direct revenue
Paid Full Version
Developed into full game for sale:
- Expand content to 5-10 hours gameplay
- Add levels, missions, AI teammates
- Price $20-40
Subscription Service
Integrated into Ubisoft+ subscription:
- As subscription-exclusive content attracting users
- Continuously update new content and AI features
- Experimental nature suitable for subscription model
Technology Licensing
Not released separately but license technology to other games:
- Integrate AI NPC technology into existing Ubisoft games
- License to third-party developers
- Become game development tool/middleware
Challenges and Risks
Development Costs
AI game development and operation costs high:
- Gemini API call fees (each interaction incurs costs)
- 80-person team labor costs
- Long-term maintenance and optimization investment
Market Uncertainty
AI game market not yet mature:
- Uncertain how much players willing to pay for AI experience
- Competitors may quickly follow, eroding advantages
- Rapid technology progress may render product obsolete
Technology Dependency
Risks depending on third-party AI services:
- Google adjusts Gemini policies or pricing
- Competitors can use same AI technology
- Lacking exclusive technology moat
Future Development Directions
Ubisoft hasn’t announced Teammates long-term plans, but possible directions include:
Short-Term (2026)
Expand Testing
From closed beta to open beta:
- Invite more players to participate
- Collect large-scale data
- Optimize AI models and game mechanics
Technical Optimization
Improve current technical bottlenecks:
- Reduce AI response latency
- Enhance voice recognition accuracy
- Increase AI behavior reliability and consistency
Medium-Term (2027-2028)
Content Expansion
If feedback positive, may develop full version:
- Longer story campaign
- More AI teammate characters
- Diverse mission types
Multiplayer Mode
Explore AI teammates mixed with real players mode:
- Real players + AI teammates cooperative mode
- AI filling teammate vacancies in multiplayer games
- AI reinforcements in PvP mode
Long-Term (2029+)
Technology Integration
Apply Teammates technology to Ubisoft flagship game series:
- Next-gen Assassin’s Creed AI NPCs
- Rainbow Six tactical AI teammates
- Far Cry dynamic NPC dialogue
Open Ecosystem
Establish AI game development platform:
- Open tools letting developers create AI-driven games
- Mod community creating new content for Teammates
- Form AI game ecosystem
Conclusion
Ubisoft’s Teammates represents bold gaming industry experiment in generative AI application. From 2024’s Neo NPC static demonstration to 2025’s playable FPS prototype, Ubisoft demonstrates capability to transform AI technology from concept to actual gaming experience.
Technological Breakthrough
Teammates proves advanced LLMs like Google Gemini can drive real-time natural language interaction within games, with AI teammates Pablo and Sofia exhibiting intelligence and naturalness surpassing traditional scripted NPCs.
Industry Implications
This project points direction for gaming industry: future NPCs will no longer be static dialogue trees but intelligent entities capable of understanding player intent, remembering interaction history, and dynamically generating responses. Voice control may become core game mechanic rather than auxiliary function.
Challenges and Unknowns
But Teammates also faces numerous challenges: technical stability, cost control, player acceptance, creative worker concerns, and fundamental question of whether AI games possess long-term appeal.
Future Outlook
Teammates currently remains experimental project, Ubisoft hasn’t committed to commercial release. But regardless of ultimate fate, this project has left mark in gaming history, becoming early milestone in generative AI gaming era.
2026 may see more major publishers launching similar AI experiments. The path Teammates opened—where will it lead the gaming industry? Will players accept this new gaming experience? Can AI truly change game design fundamentals?
Answers will gradually unfold in coming years, and Teammates stands as one of this revolution’s pioneers.
Sources:
