It’s Not about the Method
Originally through a textbook I sought to classify how well your bot learned by asking about hard-coded or machine learned and everything in between.
That frame was wrong. It was never about the method.
The better question is: how autonomous is your bot?
Autonomous here doesn’t mean you’re AlphaStar or bust. It means how many decisions your bot can make on its own without you hard-wiring every response. How much of the game can it figure out on its own?
The Ladder
That textbook gave me four levels a bot can move through.
Level 1: Simple Reflex. Your bot reacts to what it sees right now. If supply blocked, build overlord. Just plain rules.
Level 2: Model-Based. Your bot remembers what it scouted and uses that memory to make better calls.
Level 3: Goal-Based. Your bot picks actions based on what it’s trying to achieve. It plans toward objectives.
Level 4: Utility-Based. Your bot weighs competing goals against each other and picks the best tradeoff.
Most bots start at 1, reach 2 and plateau out. It works. It’s real code. But it stops short. So let’s talk about the next level
The Level 3 Crossing
The jump from 2 to 3 is where your bot stops just reacting and starts planning. Many of the top Probots occupy this level.
1. Give your bot a goal, and have it work towards it.
Right now your bot probably does something like: “if I see enemy units, build defense.” That’s reflex. A goal-based bot asks: “what am I trying to achieve this game?” Your bot should know why it’s expanding, not just when.
2. Rank your goals.
Your bot has to choose. Expand, defend a drop, or tech up. A level 2 bot picks one and ignores the rest. A level 3 bot keeps a priority list. First priority, second priority, fallback. It follows the list. That alone takes your bot from tunnel vision to juggling.
3. Make scouting change the plan.
A level 2 bot scouts and stores info. A level 3 bot scouts and adjusts its strategy. Fast expand? Shift to aggression. Rush incoming? Pull back expansion timing.
The Level 4 Build
A level 3 bot follows a priority list. A level 4 bot doesn’t need one. It scores options and chooses the best thing by expected value. It can intentionally take a short-term loss for a longer-term gain because the math says so (for example lose workers to kill tech, delay third to secure map control, skip upgrades to hit a timing) . Very few bots occupy this space. But LLMs are changing that possibility.
There are already ChatGPT controlled SC2 bots out there, so think sub-agents controlling the levers of each decision. And if you’re at level 3, your bot already has goals and priorities. It just needs to weigh choices and your systems do the rest.
You’re Closer Than You Think
Whether it’s level 1 or 2, the path to higher autonomy is a few deliberate design choices away. Either way, you’re not building “just a bot.” You’re building an autonomous agent. Now make it look like one.
Get more insight out my original breakdown of the 4 levels🔽:
