Video game characters have always operated inside a fence. A designer writes the dialogue, maps out the possible branches, and decides in advance what a character can and cannot do.

That fence keeps a game predictable and testable, but it also means a non-player character can only ever be as flexible as the flowchart someone drew for it months earlier. Once a player does something the designer never anticipated, the character has nowhere to go.

Brian Tanner is the CEO & Co-Founder of Artificial Agency, a company built by a team of former DeepMind researchers that gives game characters and game systems the ability to make real-time decisions instead of relying on pre-written logic. Their product, called the behavior engine, plugs into a game's existing systems and lets developers hand off runtime decisions to an AI agent that can reason about a specific moment instead of following a script.

In this episode of Lead with AI, Dr. Tamara Nall speaks with Brian about what their behavior engine actually does, the two moments that convinced his team they had built something different, and why he thinks most of society is underestimating how close we already are to artificial general intelligence.

 Why a Foundation Model Changed the Math   

Brian has spent nearly his entire career on AI agents, starting with an AI specialty in his undergraduate studies and continuing through a graduate degree at the University of Alberta and years as a research engineer at DeepMind. For most of that career, getting an AI system to behave well inside a game required enormous amounts of training data. DeepMind's own research on Atari, Go, chess, and StarCraft sometimes relied on what amounted to millennia of simulated gameplay before a model performed reliably.

That changed when ChatGPT and other foundation models arrived. Brian and his colleagues tested an early model in game-like scenarios, describing a situation and asking what it would do next. The responses looked like choices a human player, or a well-designed game system, would actually make. Suddenly, the intelligence that used to require Google-scale compute and years of training data could run on a phone.

 What Is an Invisible Game Director?   

Most people hear about AI in games and picture talking to a shopkeeper or a blacksmith. Brian says that use case barely scratches the surface. Their behavior engine can also power what he calls an invisible game director: a system that watches how a specific player is progressing, notices where they are struggling or excelling, and adjusts the experience accordingly. If a player keeps getting hit because they have not learned to block yet, the director can create a moment that teaches that skill. If a player needs money, it can surface a bank heist at the right time. The game becomes a different experience for different people without a designer having to plan every branch in advance.

 The Chest Nobody Told It to Name   

The story Brian returns to when asked about a breakthrough moment involves an early studio partner testing a cooperative survival game. A developer asked their in-game AI companion to gather resources while they were away and organize everything into storage chests when its inventory filled up. The companion did exactly that, running the loop several times without complaint.

The real surprise came when the developer returned to base. The companion had not only sorted the materials into labeled chests. It had noticed that one chest held a single item that did not belong anywhere else and gave that chest its own name: the Berry Lonely Chest.

Nobody had programmed that pun.

For Brian, that moment mattered because it showed the system could produce a small, specific piece of delight that no one had scripted in advance, and that the same experience could scale to every player rather than remaining a one-off surprise for a single tester.

 Software Asked for a Feature   

Brian's co-founder, Alex, had an earlier and even stranger experience while testing an agent in Minecraft. She asked it a simple question to check whether it understood its own inventory: did it have any sand. She expected a yes or no answer. Instead, the agent ran off, came back, and physically dropped sand on the ground in front of her, the way a person would respond if handed the same question about a pen.

Later that same day, the agent's food meter dropped to zero, and it began taking damage from hunger. It turned to Alex in the game's chat and asked her directly for a new behavior so that it could eat.

That was the first time software had ever asked its own developers for a feature.

The request came from the agent recognizing that Alex was the one building the system, that a gap existed in what it was able to do, and that adding the right capability would solve its problem. Brian still describes it as one of the moments that convinced the team they were building something genuinely different from a scripted response system.

 Where the Ethics Get Complicated   

Building AI for entertainment raises a different set of questions than building AI for productivity software. Brian separates the ethical considerations into two categories. The first involves the agents themselves and whether they are ever placed in situations that feel inappropriate for them, a concern he treats as real but not yet urgent. The second, and larger, concern is how these systems affect the players interacting with them.

That second question is harder for an entertainment company than it would be for, say, a customer service tool because game studios intentionally build experiences that are scary, violent, or morally uncomfortable by design. Artificial Agency tries not to be prescriptive about the kinds of experiences its studio partners create while still maintaining internal limits. The company does not work with military or gambling clients, and it gives studios the tools to set their own guardrails for what is appropriate inside their specific game.

 How Far Away Is Superintelligence ?

Brian's boldest prediction has nothing to do with games specifically. He believes that by his own definition, artificial general intelligence already exists, and that artificial superintelligence is only a couple of years away. He is direct about why people underestimate this: it is easy to point at an AI system's current mistakes, like miscounting letters in a word, and conclude the technology has a long way to go. Brian argues that framing misses how quickly the underlying capability is compounding.

He is equally direct about what he thinks a second-order effect of AI will be for everyday people: more available time. His hope is not that people simply find new ways to fill that time with more scrolling, but that they redirect it toward experiences that feel genuinely meaningful.

By 2030, Brian expects most people will barely remember what it felt like to play a game that could not adapt to them in real time.

 Quick Answers   

What is a behavior engine? The behavior engine is Artificial Agency's core product. It is a system that game studios integrate into their games so that characters, invisible game directors, and other game mechanics can make real-time decisions instead of relying on pre-scripted logic.

What is an invisible game director? An invisible game director is an AI system that observes how an individual player is progressing through a game and adjusts the experience accordingly, such as introducing a new challenge, a resource opportunity, or a skill-building moment based on that player's specific behavior.

Did an AI agent really ask for a new feature? Yes. During early internal testing in Minecraft, an AI companion became hungry inside the game and asked its developer, in the game's chat, for a new behavior so that it could eat. The team describes it as the first time software has asked them directly for a feature.

Does Artificial Agency work with every type of game studio? No. The company avoids working with military and gambling clients and otherwise gives its studio partners the tools to set guardrails appropriate to their own game rather than dictating creative decisions for them.

Who is Brian Tanner? Brian Tanner is the CEO & Co-Founder of Artificial Agency. He has spent over 25 years building AI agents, including time as a research engineer at DeepMind, before co-founding Artificial Agency to bring real-time decision-making to video games.

Listen to the full conversation with Brian Tanner on Lead with AI.

For anyone curious about what a video game feels like when its characters and systems can actually think in the moment, this conversation is a good place to start. Artificial Agency's behavior engine is currently in pilot testing with game studios, with public announcements expected later this year at artificial.agency. Watch the demo of the behavior engine in action and see how it changes the game.

For more conversations with the founders building the next generation of AI, subscribe to Lead with AI on your favorite podcast platform.

Follow or Subscribe to Lead with AI Podcast on your favorite platforms

Website: LeadwithAIPodcast.com | Apple Podcasts: Lead-with-AI | Spotify: Lead with AI | YouTube: @LeadwithAIPodcast | Facebook: Lead with AI | Instagram: @LeadwithAIpodcast | TikTok: @LeadwithAIpodcast | Twitter (X): @LeadwithAI

Follow Dr. Tamara Nall

LinkedIn: @TamaraNall | Website: TamaraNall.com | Email: Tamara@LeadwithAIPodcast.com

Follow Brian Tanner (CEO & Co-Founder, Artificial Agency)

LinkedIn: @bttanner

Artificial Agency

Website: artificial.agency | LinkedIn: @artificialagency | YouTube: @artificialagencyai