README

making our bots matter outside the ladder

Drekken

Why hello Reader,

“90% of all AI research gets discarded because it never gets used in the real world.”

I heard a guy from Ubisoft’s bot division say that on a podcast recently, and it really struck me.

You know what came to mind? AlphaStar. It came, made a massive splash, and then disappeared.

DeepMind obviously had a real-world application for their research, but the bot itself? Mothballs. And it makes you wonder how many of our own bots share that exact same fate. You spend hours on it, working on your win rate and tweaking it, only to have it sit on a shelf later, collecting dust.

It begs the question: is there more to building a bot than just farming MMR on a ladder? If so, what do those use cases actually look like?

You give a bot purpose by working backward from a real problem.

Shifting from competitive tuning to utility-based design turns a hobby project into a professional asset. Professional game AI development starts with a studio bottleneck and builds the bot to solve it.

That utility falls into two functional buckets.

Player-Facing Utility

Player-facing bots do a lot more than just fight the user.

Dropping a brand new player into the deep end to get completely destroyed by veterans is terrible game design. A throttled bot serves as a dynamic guide to walk them through the complexities of the game.

They also serve as drop-in teammates. When a human disconnects from a multiplayer match, a bot seamlessly takes over that slot to keep the game alive.

They provide opponents across a sliding scale of skill. This gives everyone from a day-one beginner to a seasoned pro a caliber of match that actually fits them.

Testing & QA Utility

Anyone who has worked in game production knows testing is a massive bottleneck. A reliable bot turns manual QA into an automated system.

You use a bot’s navigation mesh for reachability tests. It automatically verifies if two points on a newly designed map are actually connected.

You deploy twenty agents simultaneously to stress-test the game’s performance and track exact FPS drops.

You pit bots against each other over thousands of iterations to surface core gameplay imbalances. This verifies mechanics without requiring a human to playtest every single numerical tweak.

We are engineering autonomous systems. Optimizing purely for a ladder win rate restricts the bot to a single, narrow environment.

Applying Problem-Backward Design makes your code indispensable to a future game studio. Build solutions that help ship better games.

I did a talk about this in depth at a Game Conference,
check it out👇🏾

video preview

⌨️ Next Commit

Look at your bot’s current architecture. Pick one bucket—Player-Facing or Testing & QA. Identify one specific studio problem your bot could solve, whether that is guiding a brand new player or running reachability tests on a map. Write down the first three steps required to repurpose your code to do it.

Are there other use cases that I might have forgotten? Let me know.

May the Bugs Be Ever In your Favour🪲

Email Preference:
Unsubscribe | Update your profile | 600 1st Ave, Ste 330 PMB 92768, Seattle, WA 98104-2246