"We’re pushing the boundary of how AI can be adopted in games"

Hello and welcome back to AI Gamechangers, our regular deep dive into how artificial intelligence is being put to practical use across the games industry.

This week, we’re chatting with James Wu of GameBot, whose team is deploying AI-driven teammates, opponents and even evolving MMORPG bosses into live games like PUBG Mobile and Wild Rift. We explore how “invisible” bots can smooth onboarding, reduce frustration and improve retention, and what it takes to integrate this kind of system without touching source code.

Next week, the games industry takes over San Francisco thanks to the GDC Festival of Gaming and adjacent activities. We’re going to be there, and at PGC Summit San Francisco on Monday, 9th March, too - join us.

If you’re new here, welcome. Our archive of conversations with industry leaders is free to browse, and there’s plenty more to come.

James Wu, GameBot

Today, we discover deep-learning agents that mimic player behaviour and exist inside titles such as PUBG Mobile, League of Legends: Wild Rift and MIR5. Bots like this can help improve new-player onboarding and boost retention.

Our guide is James Wu, General Manager of AI Partnerships at GameBot. With nearly two decades in games spanning EA, Nvidia, ByteDance and more, Wu has built consoles, cloud gaming platforms and strategic partnerships across Asia and beyond. Now at GameBot, a company founded by alumni of Tencent AI Labs, he’s focused on developing bots that can become visible teammates, invisible assistants, or even adaptive bosses that learn in real time.

Top takeaways from this conversation:

1. GameBot deploys AI teammates and opponents into live multiplayer titles to reduce frustration, smooth onboarding and measurably lift retention (without players detecting they’re bots).

2. In MMORPGs, AI-driven bosses can learn from player behaviour in real time, evolving during battle, creating replayable, less predictable encounters.

3. Integration does not require access to source code or sensitive player data. Bots are trained on game rules and deployed via the cloud, connecting to servers like normal players.

4. James Wu advises mid-sized studios to go “all in” strategically on AI, but tactically partner with specialists to prototype measurable use cases before building internal AI capability.

AI Gamechangers: Please give us an overview of your background. How did you get to where you are today?

James Wu: I’ve been working in the gaming industry for two decades already, spanning EA, Nvidia, Xbox, Logitech, ByteDance Gaming: basically multinational companies as well as local Chinese companies.

I built two startups during that time. We tried to build a gaming console for China; to challenge Xbox and PlayStation really sounds crazy, and obviously didn’t come up with the best success.

Recently, I joined GameBot. GameBot is a startup that’s now seven years old. The core team are originally from Tencent AI Labs. We deliver AI agent solutions for top-tier developers and publishers, including the League of Legends Wild Rift, PUBG Mobile, Garena Free Fire, and recently, MIR5 from Korean developer Wemade. We help them deliver measurable results by boosting user retention as well as the number of active users.

I’m chasing a career goal to help more developers achieve their goals. Working with AI bots and AI agents is one way to help game studios embrace AI better. My plan here is strategy and expansion because we have been working with giant companies in China, then Japan, and now Europe. We’ve already been talking with Ubisoft for a while, and I want to meet more and more European developers.

When did you personally first come to understand the importance of AI in games?

When I left my last job, which was the head of China gaming product at Logitech, I tried to build my own business again. When I talked to my investor friends, I was shocked – everyone was asking me, “Is your new project AI-related? If not, I don’t want to waste my time!” Everybody in the industry is talking about AI, too. So it was in 2023 that I started to take it seriously. And then I started to use AI more often myself, and I realised that it’s not just a trend; I’m a believer.

“Recently, we released a technical demo called COTA, our ‘chain of thinking agent’ – we believe that’s the world’s first shooting game experience with AI agents. It’s based on LLM-driven agents, but the response time is less than 100 milliseconds”
James Wu

On my career path, I’ve spent a lot of time studying how I can help the whole industry. That’s a big dream. I try to build platforms (like gaming consoles, hardware platforms and a cloud gaming platform business), and my hope is to create those platforms so that developers and publishers can benefit, and the users can get good content. At the end of 2023, I realised that AI could be a tool that helped me to approach my goal.

Then the offer came! The core team and founders are really good, and the technology doesn’t have direct competitors back in China.

AI means different things to different people. LLMs, ML, art… But what are you doing with AI, and how does it work?

Our product, GameBot, is pretty unique. Our AI is not like if they install an LLM chatbot – we’re happy to help game developers on the back end rather than the front end. We develop AI bots to be actors in the game. The bots can play the same game with you, and they perform as your teammates or your opponents. Real gamers cannot find out that they are AI.

They can help with new-player onboarding (in place of other real gamers who might be too difficult to survive against at first). This is the way we provide emotional values to gamers while also helping lift retention.

“We’ve been asked: Are you cheating? Are you misleading your players? I don’t think so, because that’s just one part of the game’s features – the same as they are already doing with behaviour-tree AI. We’re just making a smarter move”
James Wu

We might also provide “visible” (obvious) AI characters too, who act as NPCs or bosses, or even some of the co-pilots who help you to play the game better.

Traditional AIs in games are based on simple behaviour trees. They can be real dummies – and what we’re doing is replacing them with our deep-learning bots, so that they can play more like human beings. And in this way, our bots could help you make more money.

The team has a secret recipe for training the bots, which I obviously can’t go too deep into here. They have built their own models. They’re not LLMs, of course, but special models.

For example, we work with PUBG Mobile. We check their requirements, what they need us to help them with. For example, perhaps they want a smoother onboarding for new players. We get the right bots trained on the servers. It can take three to six weeks, and after that, we submit them for an acceptance test, and then deploy the bots onto the cloud servers – they will then connect to the game servers, just as normal players do!

Please talk us through the advantages for a player. Does the presence of AI bots in the game make it a better experience for gamers?

The first thing is that we have invisible AIs: bots whose presence you don’t feel. For example, when you play League of Legends, if you lose three times in a row, you probably get so frustrated that you AFK or quit the game. But we have a mechanism where our AI will detect if you lose three times in a row, and then it triggers a frustration mitigation route where your teammates and opponents will all be our bots. You can’t detect that, and you will feel a confidence boost. That’s a typical example of how we help in games like this.

“The bots can play the same game with you, and they perform as your teammates or your opponents. Real gamers cannot find out that they are AI. They can help with new-player onboarding (in place of other real gamers who might be too difficult to survive against at first)”
James Wu

Sometimes we’ve been asked: Are you cheating? Are you misleading your players? I don’t think so, because that’s just one part of the game’s features – the same as they are already doing with behaviour-tree AI. We’re just making a smarter move.

For the visible AIs, we recently released a partnership with MIR5. MIR5 is an MMORPG releasing this year. With MMORPG bosses, everybody knows at what level they will cast what spell or do something, and yet people still play against the boss every week or every month. Without touching the source code, without changing anything on the server side, and without revealing any sensitive user data, we help the developer turn their boss into an AI-driven boss. The boss will have memories, its behaviour is all AI-driven, and from the very first second the players get into battle with it, the boss learns from them – how they move, what their strategy is. Theoretically, the boss will evolve: the longer the battle, the harder the boss gets. That creates a brand-new gaming experience.

Think of replacing those traditional MMORPG bosses with the monsters from the Monster Hunter series. And that’s not to mention games like Rainbow Six, which are usually very unfriendly to new users. Our game bots can grow together with newcomers. Those bots won’t leave you. They won’t think, “Oh, you’re such a poor player, I don’t want to play with you.” That’s something we can already help with in the game.

Companies are considering how to get started with AI. Individuals are worried about losing their jobs. If a mid-sized studio were to come to you tomorrow asking how to get started with AI bots, what would your advice be to them?

First of all, AI is new to everybody. If you’re a mid-sized company, you cannot take too many risks – you should do things smart and step by step. Strategy-wise, you should go all in. You should embrace AI and use it as much as possible in your daily life. But tactically, the smart move is to let the professionals do their job. Solution providers like us can help you come up with a demo or prototype with measurable results and controllable risk. And then, during the partnership, your team will learn and grow, and you will build your own AI skill base.

That’s the thing – there’s a fear of missing out, I think. Your competitors are using AI to help lift their retention already. Don’t get left out!

What’s next on the roadmap for GameBot? Tell us about what the next 12 months of product development look like for you.

We’re pushing the boundary of how AI can be adopted in games, especially in the operation part. And recently we released a technical demo called COTA, our “chain of thinking agent” – we believe that’s the world’s first shooting game experience with AI agents. It’s based on LLM-driven agents, but the response time is less than 100 milliseconds, which is very quick.

The good part is that we review all the thoughts of these AIs, real-time, within this high-pace environment, and see how they’re strategising. Why are they making this move? Why are they rushing to point A instead of going elsewhere? We believe that that’s helpful for pro training as well as balance testing. It’s also about being transparent to gamers. That’s one thing that we recently released.

We are thinking, down the line, of releasing a tool to help individuals develop their own games, perhaps inside of a platform like Roblox. You might use natural language, and then a tool will help you to generate in the Lua language something you can bring into Roblox, to make your own game experience easily.

You can integrate AI bots without changing the game. Please tell us about the onboarding process and what the experience is like for a company that wants to try this.

First of all, I’d like to know what your needs and requirements are. Then the two teams (technical and product teams) will work closely for six to eight months. It takes about eight weeks to come up with what we call a game core – a dedicated server – because we need to feed our bots with the structure of the data instead of just a visual thing. AI doesn’t see things! So we feed them with how the game plays, its rules, so that they can understand. We get it trained, and then we will spend one or two months working on iterations and tests. Finally, we deploy in the cloud, and it’s on demand.

[From start to finish] it’s six months, maybe nine months at most – and there’s no touching of source code. We don’t need any sensitive data from the user side! And the cost is affordable. We first charge an NRE [Non-Recurring Expenses charge] as a development fee that covers the cost of the training. Most of the costs are in training. And the second part, when we deploy on the core servers, the model is SaaS [Software as a Service], and it depends on how many bots you’re using concurrently monthly.

Subscribe now! It’s free, and you’ll receive interviews with industry insiders and studio leaders in your inbox. Coming soon: Unico, CALM Consultancy and more, following sessions at PGC and GDC.

Further down the rabbit hole

Some useful news, views and links to keep you going until next time:

• According to a recent interview, Xbox is facing “no pressure from Microsoft” to use AI under the new leadership of Microsoft Gaming CEO Asha Sharma.

• Krafton appointed Kangwook Lee as chief AI officer to oversee the company-wide R&D roadmap. He will focus on “elevating gameplay, enhancing operational efficiency, and securing new growth drivers”.

• It was recently revealed that Amazon’s AI coding tool Kiro caused a 13-hour AWS outage in December after autonomously deciding to “delete and recreate” its environment. Amazon claims user error, not AI error, but the story raises questions about how much autonomy is handed to AI agents, as does the story of Meta’s Summer Yue, whose OpenClaw agent began deleting her inbox.

• Games Factory Talents is hosting an AI and investment-focused careers fair in Helsinki on 23rd April. Probably worth a look if you're hiring, job-hunting, or just curious about where the industry is heading.

• AI-powered storytelling platform Giant has raised $8 million to scale and expand its interactive offerings for children.

• Saga (whose CEO Rebecca Liao we interviewed last year) has partnered with Trivia Crack studio Etermax to turn mascot Willy into an AI agent designed to chat with players on social platforms and drive game installs.

• It’s your last chance to get cheap tickets to PGC Summit San Francisco, taking place on Monday, 9th March, right before GDC. Speakers include reps from Meta, Scopely, Discord, Tencent, Arcanix, 10six and loads more.

• Ridl-E: The AI Riddle Bot is out now and turns your surroundings into a scavenger hunt by using AI to generate riddles on your phone from photos.