You are a game developer, and you have just typed a line of dialogue for a non-player character. It is a simple greeting, a few words of welcome for the player entering a tavern. You hit play. The character speaks your line, its mouth moving in sync with the audio file you recorded last week. This is the old way. Now, imagine you instead give that character a backstory, a personality, a goal. You connect it to a large language model and a voice synthesis system that runs in real time. You press play, and the player walks in. The character turns, assesses the player's gear, remembers their last interaction three hours ago, and improvises a line you never wrote. Its tone is wary because its personality is suspicious of outsiders. This is the bet Inworld AI is making: that the future of digital interaction is not scripted, but generative [Inworld AI].
Founded in 2021 by Kylan Gibbs and Ilya Gelfenbeyn, Inworld has raised a total of $120 million to build what it calls a Character Engine and Runtime for AI-powered NPCs [Google Cloud, recent]. It is not merely an API to an LLM. It is a layered system designed for the specific, punishing demands of real-time entertainment: millisecond latency, scalability to millions of concurrent users, and integration directly into game engines like Unity and Unreal [Lightspeed Venture Partners, ~2023]. The company's case study with Google Cloud claims it can scale to support over 10 million concurrent users, a figure that speaks to the architectural ambition behind the product [Google Cloud, recent]. For developers, the promise is to replace brittle dialogue trees with characters that can adapt, remember, and emote, creating a sense of living world that has been a fantasy of game design for decades.
The wedge is real-time orchestration
The technical differentiator for Inworld is not access to a foundational model. Many companies can pipe GPT or Claude into a chat window. The wedge is the orchestration layer that sits between the raw language model and the game world. This layer handles several critical tasks that are foreign to a typical chatbot. It manages character memory across a session, filters outputs for safety and narrative consistency, triggers appropriate animations and emotional expressions on the character's face, and ensures all of this happens with the low latency required for natural conversation in a fast-paced game. This is the proprietary 'engine' the company is selling. It abstracts away the complexity of stitching together voice synthesis so voice lines are generated on the fly with appropriate emotional inflection, and crucially, it ties the character's cognitive layer to animation and rigging systems so that a decision to be 'angry' results in the correct facial expression and body language [Lightspeed Venture Partners, ~2023]. This is the hard, unglamorous work of making AI feel present in a 3D space, and it is where Inworld has focused its engineering.
A founder pedigree built for conversation
The founding team's background reads as a direct prelude to this company. Kylan Gibbs was a co-founder of API.AI, the conversational AI platform acquired by Google in 2016 and rebranded as Dialogflow [TechCrunch, Aug 2023]. He later worked as a product manager at Google DeepMind [Business Insider, 2026]. His co-founder, Ilya Gelfenbeyn, was the founding lead of the Google Assistant Investments program [TechCrunch, 2019]. They have spent their careers at the intersection of natural language understanding, developer platforms, and scalable systems. This is not a team experimenting with AI for the first time; they are applying a decade of domain knowledge in conversational interfaces to the specific problem of game characters. The rest of the leadership team fills out the operational and technical picture, with a CTO, a VP of Operations, a Chief Science Officer, and strategic advisors like John Gaeta, known for his work on The Matrix films [Inworld AI].
| Role | Name | Notable Background |
|---|---|---|
| CEO & Co-Founder | Kylan Gibbs | Co-founder of API.AI (Google Dialogflow), ex-Google DeepMind PM |
| Executive Chairman & Chief Strategy Officer | Ilya Gelfenbeyn | Founding lead, Google Assistant Investments |
| CTO | Michael Ermolenko | Not specified in public sources |
| Chief Science Officer | Igor Poletaev | Not specified in public sources |
| VP, Operations & Corp Dev | Florin Radu | Not specified in public sources |
| Strategic Advisor | John Gaeta | Creative pioneer, The Matrix visual effects |
Traction through partnerships and scale
Inworld's go-to-market strategy hinges on embedding its tools into the pipelines of game studios and platforms. Its most significant partnership is with Microsoft and Xbox, working to develop generative AI tools for game developers, including an AI character engine [Inworld AI, 2026]. Other named customers and partners include Niantic (of Pokémon GO fame), ILM Immersive, and LG UPlus [Inworld AI]. A notable case study is the game Death by AI, which Inworld says used its custom APIs to reach profitability with 20 million players, though this metric is sourced from the company's blog [Inworld Blog, recent]. The broader signal is that the platform is moving beyond experiments into shipped, scaled consumer experiences.
The company's funding history shows a steady drumbeat of support from a who's who of strategic investors, reflecting belief in both the technology and its market fit.
2023 Series B | 50 | M USD
2023 Undisclosed | 30 | M USD
Total Raised | 120 | M USD
Investors include Lightspeed (which led the last major round), Microsoft's M12 fund, Intel Capital, Samsung Next, Meta, Disney Accelerator, Founders Fund, and Kleiner Perkins [Lightspeed Venture Partners, ~2023]. This capital is likely being deployed to further build out the runtime infrastructure and expand the business development team, as evidenced by open roles for a GTM Lead for Partnerships and senior machine learning engineers [Inworld AI Careers].
The risks of commoditization and cost
For all its technical specificity, Inworld operates in a field that is attracting intense competition and faces fundamental economic questions. The risks are not trivial.
- The LLM layer. Inworld's orchestration is valuable, but it sits atop large language models that are rapidly commoditizing and improving. If the core reasoning of an NPC becomes a cheap, high-quality commodity, the value of the middleware could be squeezed. Inworld's answer is that its deep integration with game engines and its proprietary systems for memory, safety, and animation are the true moat, not the LLM call itself.
- The cost question. Generating unique voice dialogue and processing complex interactions for millions of players in real time is computationally expensive. Inworld's Runtime launch specifically touted a 95% reduction in scaling costs for one customer, indicating this is a primary battlefield [Inworld Blog, recent]. The business model depends on driving these costs down faster than the price developers are willing to pay.
- Proving the use case. While novel, generative NPCs must prove they make games more fun, engaging, or efficient to produce, not just more technologically sophisticated. The industry is littered with advanced tools that designers found too unpredictable or difficult to direct. Inworld's partnerships with major studios are the early test of this premise.
The cultural question beneath the code
Every new tool for creators answers a silent cultural question. The spreadsheet answered, "What if we could model business decisions before we make them?" The camera phone answered, "What if everyone could document their own perspective?" Inworld's Character Engine is answering a question that has simmered through decades of interactive fiction, from text adventures to massive online worlds: what if the people in these worlds could remember us? The product is a set of APIs and SDKs, but the implicit promise is a shift in the nature of digital companionship. It is betting that players, and eventually users of all sorts of apps, will develop a deeper, more meaningful attachment to characters that exhibit continuity and agency. This is not about replacing writers, but about providing a new kind of brush,one that paints with behavior instead of just words. The next twelve months will be about seeing what the first generation of artists chooses to paint, and whether players, in the quiet moment after the quest is over, bother to go back and talk to the bartender just to see what he says.
Sources
- [Inworld AI] Inworld AI | Top-ranked voice AI for realtime applications | https://inworld.ai/
- [Google Cloud, recent] Inworld case study | https://cloud.google.com/customers/inworld
- [TechCrunch, Aug 2023] Inworld, a generative AI platform for creating NPCs, lands fresh investment | https://techcrunch.com/2023/08/02/inworld-a-generative-ai-platform-for-creating-npcs-lands-fresh-investment/
- [Lightspeed Venture Partners, ~2023] Building With Inworld,The Character Engine for AI NPCs | https://lsvp.com/stories/inworld-ai-npcs-character-engine/
- [Inworld Blog, recent] Introducing Inworld Runtime | https://inworld.ai/blog/introducing-inworld-runtime
- [Inworld Blog, recent] How Inworld Helped Death by AI Reach Profitability | https://inworld.ai/blog/how-inworld-helped-the-ai-game-death-by-ai-with-20-million-players-reach-profitability
- [Business Insider, 2026] AI's first wave was about cutting costs. The second wave is about building things we've never seen. | https://www.businessinsider.com/ai-second-wave-redefines-startups-new-products-2026-2
- [TechCrunch, 2019] Drivetime nabs $11M from Makers Fund, Amazon and Google to build voice-based games for drivers | https://techcrunch.com/2019/09/09/drivetime-nabs-11m-from-makers-fund-amazon-and-google-to-build-voice-based-games-for-drivers/
- [Inworld AI] Inworld AI: Realtime, interactive AI for gaming and media | https://inworld.ai/gaming-and-media
- [Inworld AI Careers] Inworld AI Careers | https://inworld.ai/careers