I have been playing competitive shooters since I was a teenager. Kofi started around the same time. Lena was an FPS player before she was an engineer. Yuki has spent more hours in ranked queues than she has spent sleeping this decade. We started M10 because the four of us, after years of playing these games seriously, had run out of patience with a set of problems that nobody seemed to be treating as technical problems that could actually be fixed.
This post is about what those problems are and why we think they are fixable. It is not a pitch. We have not shipped a game yet. This is just an explanation of why four people who could be doing other things decided this was worth two years of their lives so far.
Problem One: Bots That Never Learn Anything
Most competitive shooters have practice modes with AI opponents. Those opponents are, with a few exceptions, scripted in a way that has not fundamentally changed since the early 2000s. They have a difficulty setting that controls their accuracy and reaction time. At higher difficulty they shoot more accurately and react faster. They do not learn your habits. They do not counter your strategies. They do not get harder to play against in any way that requires you to actually change how you play.
This is not a resource problem. These studios have enormous teams. It is a priority problem. Single-player campaign AI gets real investment because it is visible and reviewable. Practice mode AI is treated as a utility, not a product. The result is that the largest segment of competitive player time spent outside of ranked matches, the time players use to warm up, develop new strategies, and practice specific mechanics, is spent against opponents that provide almost no useful resistance.
When we say our opponents adapt to how you play, we mean something specific. We mean they track your positioning habits, weapon preferences, and engagement timing over the course of a session and adjust their behavior to counter what you are actually doing, not what an average player at your rating would do. The system we are building is not smarter in the sense of having better aim. It is smarter in the sense that it builds a model of you specifically and acts on that model.
Problem Two: ELO Hell Is Real and Nobody Is Solving It
The term ELO hell describes a state that competitive players know intimately: you are stuck in a rating bracket where the quality of your teammates is so variable that your individual performance has very limited effect on your match outcomes. You win matches because your teammates happened to be having a good day. You lose matches because they were not. Your rating stagnates regardless of how well you personally play.
The standard industry response to this is to say ELO hell is a myth, that ratings systems are designed to surface individual contribution, and that players who blame ELO hell are not taking responsibility for their own performance. This response is technically defensible in the aggregate but misses what is actually happening at the individual match level. A single match in a team-based game can be dominated by one player performing extremely well or extremely badly in a way that other players cannot compensate for. When those outlier performances cluster, they produce exactly the rating stagnation players report.
Our matchmaking approach tries to address this at the source by reading skill signal in real time rather than relying entirely on historical rating. A player who is performing above their rated level in the current match influences the match composition engine directly, not through a slow rating update that takes dozens of matches to propagate. This does not eliminate variance in match quality, but it gives the system more information to work with per match, which reduces the worst-case clustering of extreme performance outliers.
Problem Three: The Skill Ceiling Problem Nobody Talks About
The problems above are well known in the competitive gaming community. This one is less discussed, probably because the players most affected by it are the ones with the least reason to complain publicly.
Very skilled players in large competitive titles have no meaningful way to continue developing once they reach the upper tiers of the matchmaking system. They have beaten the bots. They have climbed to a rating where every match is against players at or near their level. At that point, their improvement curve flattens. The mechanics they can still develop are extremely narrow, and the matches available to them do not provide the kind of varied challenge that would push them to develop new strategies.
An AI opponent that adapts to you continuously raises its effective ceiling as you improve. There is no "easy" setting once you understand the system, because the system's difficulty is a function of your own behavior. If you become better at varying your playstyle to confuse the AI, the AI has more to work with and builds better counters. The growth is bilateral, and there is no hard ceiling to hit.
We want the game to be worth playing for the players who currently have nothing left to practice because they have already beaten everything available to them. That is not a mass market goal. It is a product quality goal that we think produces a better game for everyone, including players who are nowhere near that level yet.
What We Are Building and What We Are Not
We are building one game, not a platform, not a framework, not a suite of tools for other developers. The AI and matchmaking systems we are developing are built to serve this specific game, and we have designed them with the constraints that game imposes: a competitive shooter format, team-based play, and a population of players who take the game seriously enough to care whether the opponents are worth their time.
We are not building the most technically complex AI system in the industry. We are building one that is technically sophisticated enough to produce the player experience we are targeting. Those are different goals, and conflating them would lead us to build something that impresses engineers and confuses everyone else.
We are also not claiming we are the only people who understand this problem. There are smart people at large studios who have thought carefully about every issue we have described here. The reason these problems persist is not ignorance. It is incentive structures. Fixing the practice mode AI does not ship features that go on the back of the box. Rebuilding the matchmaking system from scratch is a multi-year risk with no guaranteed payoff. For a studio with 400 people and an existing product to maintain, those trade-offs make a certain kind of sense.
For four people who have no existing product to protect and who are building from scratch specifically to address these problems, the trade-offs are completely different. That is the advantage of starting small and starting focused.
Where We Are Now
We have an internal prototype with the core AI adaptation mechanic working. We have a team that has been building this for two years. We have a sense of what the finished game needs to do and a realistic picture of how long it will take to get there. We do not have a release date because we are not willing to commit to one until the game is playable by people outside this room and we have confirmed that the thing we are building actually delivers what we are describing.
When it is ready, we will let people in. Until then, if you want to follow along, this is where we will be writing about what we are learning.