Bots that break games (and make them better)


July 9th 2025


Why hello Reader,

In June, I spoke at the XP Summit here in Toronto—a gathering of game developers from across the industry. Instead of focusing on the usual competitive angle, I talked about something that doesn’t get as much attention: how the bots we build could actually help make games better.

Think about it—most studios are drowning in QA work. Testing is repetitive, slow, and expensive. Meanwhile, our bots already do what testers can’t: run the same scenario over and over without losing focus, break things in unexpected ways, and explore every edge case without complaint.

Studios like Ubisoft and EA are already leaning into this. They’re using bots to balance systems, run overnight tests, and catch bugs humans miss.

So here’s the idea I pitched at XP: what if our bots weren’t just good at playing games, but actually improving them? What if the next step for your project wasn’t a stronger opponent—but a smarter ally for game devs?

There’s a real opportunity here. Not just to sharpen your skills, but to build something useful—something that makes the whole game better.


Here's the full talk if you want to see how I laid all this out at XP Summit :

video preview

For Your Radar

we're back for Week 3 of Probots, elimination week! don't forget to set your notifications so you don't miss it at 5:30 PM EDT

🗒️ ./run Notes:

A couple of ideas from my XP summit talk, if you want to build skills for using a bot (in this case a SC2 Bot) for QA.

Heatmaps & Pathfinding

  • if you haven't yet, give my pathfinding challenge a try.
  • Practice using A* or Dijkstra to get the probe to the goal
  • In the second part use heat maps to prevent the probe from dying

Adversarial Bot Loops

There are still no good training environments for Python-sc2 that I have found (though Sharky for C# just cooked one up). You could mess with Stablebaseline3 + Zoo or Gym but that's a whole mission. (Let me know if that's something you'd want to see.)

To keep it simple though, with a microbot:

  • Run BotA vs BotB locally using python-sc2.
  • Download a Bot Micro Practice Map - (shout out to Sharknice for his) or you can create your own.
  • Track metrics like win rate and time to win.
  • Mutate one bot slightly after each round (e.g., add aggression, change unit comp, tweak micro timing).
  • Use the results to evolve the weaker bot—or stress test the stronger one.

Got ideas this sparked? I’d love to hear what you're testing. But that should be enough to get you started


Ok before we go,

Need a real project for your resume or portfolio?

Join a weekly build session and leave your mark on something that ships. → [Link]

Ok, that's it from me this week,

Happy Coding!

Drekken
Founder, VersusAI

📧 Drekken@versusai.net | 💬 Discord: drekken1

May the Bugs Be Ever In your Favour🪲

One Last Thing...

Email Preference:
Unsubscribe | Update your profile | 113 Cherry St #92768, Seattle, WA 98104-2205

Pownz Clan

README is a bi-weekly batch of Gaming AI competitions, practical bot-building tips, and behind-the-scenes insights— keeping you one step ahead

Read more from Pownz Clan
frustrated drekken sitting at a desk with post-it notes around his screen , starcraft 2 is on the screen

June 24th 2025 Why hello Reader, While adding my build to my bot and deploying it to the ladder, I was reviewing a replay from a Protoss bot. In that replay, the bot completely countered mine. Naturally, I jumped into solving that problem — tweaking things, adjusting timing — only to review another replay where a Reaper harass tore my bot apart. And then another one. Before long, I was reacting to everything and making no real progress. I didn’t even feel like working on the bot anymore....

bot maker riding a disruptor unit

June 10th 2025 Why hello Reader, Ever watch your disruptor fire a perfect shot... right into your own units? Yeah. RIP my army. You know, when I started I thought I had everything figured out. Disruptors look simple—just aim for the biggest group of units and boom, free wins. But turns out, in StarCraft, disruptor shots aren’t even spells. They’re units in the API. Which means the bot needs to steer them. Every. Single. Frame. This is nothing to say the risk factor of 100 true damages it does...

playing poker with star wars droid

May 28th 2025 Why hello Reader, I was catching up on Andor Season 2 (fantastic show btw, highly recommend), and there was this scene that caught my attention. K-2SO, everyone's favorite sarcastic security droid, was struggling at a poker-like game, totally baffled by the concept of bluffing. Which had me thinking: how is it that an AI like K-2SO can stomp an enemy in combat, but put him across a table from someone trying to mislead him, and he falls apart? In Case You Missed It Progress made...