Looking for feedback on my Monster A.I. Zine


Thanks @MonsterAI for pointing that stuff out. I didn’t think about that, but that’s really good feedback. Maybe I could put stuff like Patrol and Retreat mode on the back so that becomes more of Combat side on the front and those additional Roleplaying-esque states on the back.

That would leave more room for the actions like your sprint suggestion, which totally makes sense.


Fuck yeah, flipping cards over is my jam. That’s like when the Innistrad werewolves transform in MTG. Man y’all are spoiling the ideas I had for future releases lol.

And you don’t really need to write down a sprint action, that’s already in the rules for checking conditions. Just remember that if enemies are too far away for your monster to target, that’s when it performs a sprint action.


Alright, call me crazy.

Here’s what I came up with for a double sided Monster/Enemy AI. The front side (left) is combat focused and the back side (right) is more Roleplaying focused.

I also threw in a little bit of flavor text and a space for Loot Ideas because, well LOOT!

Formatting is still going to need work and I’m not happy with the empty space below the Hearts on the back side. Harrumph!


This is rad. I’m just gonna be an annoying MTG rules judge for a moment and say that the last combat behavior won’t always target the enemy that’s at or below 1/4 health unless you specify in the behavior text “Target that enemy” because the rules for targeting say if there’s multiple enemies that can be targeted by an ability then the monster will go after its target priority. But we both know what it’s supposed to do, so it’s fine lol.

But hey you got this, go forth and have fun at your play session.


Yeah totally. I think the booklet is laid out in a similar logic, it starts with your two basic goblin enemies (warriors and archers), turn the page and a more advanced goblin thief plus a basic orc warrior, last two pages are the shaman and berserker. But you know what, I think what I’ll do is something like what you said but each A4 page is for one race or faction and then I’ll make 4 enemy schemes for each. That way you have a sheet for goblins and one for orcs, for example. Thank you, this was some helpful feedback.


Oh shoot scratch that, the last condition would literally never activate because the first condition already checks if 1-2 enemies are close. What you want to do is make the last behavior the first one to check.


Those look great, and I’m so glad the ideas just keep flowing. Only thing I am confused about, if you are playing ICRPG or mixing systems?

If ICRPG, the NPC would never make a check to see the players, the “room” has a target number…it represents the difficulty of the room, taking out the need of a wis check for alert mode.

I only mention this cause it’s a great mechanic, and you can adjust room difficulty on your GM turn.

So target can be 18. Thief type with all the stealth loot only needs a 15+ and that means rolling a nat 5+. If he attacks the guard it’s still easy and double effort, with bonuses to sneak attacks, PC thief needs a nat 12+.

Assuming a miss when the GM turn comes up his AI kicks in, he runs and screams the alarm!!! But the room difficulty goes down to 13 , cause he is hard to get past unnoticed but not the best fighter around with that bum knee.

Anyway really long way of saying, he doesn’t really need to roll against a player, the player rolls against the target. Less numbers floating in the table.


It’s very probable I’m screwing something up. I am playing ICRPG but maybe I’m missing some things haha.

I included the WIS check because I was under the impression that room target applied to the monsters/NPCs too. So when it was the guards turn to go on patrol they would roll against the target just like everybody else.


this is way off topic but how else can I find out if I am wrong.

however you are playing it, as long as you and your group are having fun…It’s great.
but my take is this…I’m probably wrong.
This game is shockingly streamlined and made to be run by sensitive GMs who want to turn it up to 11, but not be at fault.

  • PCs have a set target number, easy normal and hard based on the Target

  • NPCs/Mobs target the Players armor to hit or The PC makes a save vs the Target

  • Many of the Monsters have attacks equal to a dice D4, D6, D8…you can use that dice to randomize their attacks. To speed play roll an attack type Die, a D20 and the damage Die…but that reduces anticipation of whats going to happen next.

So Target number = Armor, perception, magic protection, difficulty to pick pocket, open locks, disarm traps…
Mosters/NPCs/Mobs don’t need many stats, just their pluses and damage…if you have vigilant guards or guards who see in some weird way…let me know where you are going and give me a hard sneak roll. Or change the room Target when combat begins on your turn.

bringing it back to topic.
Also to make goblin AI, think of the goblin squad…as they loose numbers then they get less reckless, then downright cowardly.

So we should have individual AI and group AI, and exception, Stone heart warriors might not flee, while the warriors might.

ICRPG is very streamlined, and with something like Monster Logic you can shorthand things even more to add complexity to your monsters.


That makes sense and I appreciate the feedback. It’s nice to have more experienced people to help out :smiley:

I think I got into the weeds when I was plotting out how enemies would do things that weren’t combat-related. So like a monster who is patrolling a room. I thought to myself, “well… they wouldn’t roll against armor so they must make a WIS check.”

After reading your explanation, I see that my logic took a left turn and it should have been “The monster is going to stay on patrol until the PC’s fail a stealth check.”

At that point they would transition from Patrol Mode to Alert Mode and the PCs would need to make another successful check, albeit slightly more difficult, to avoid being found out, right?


Ok, this is going to be one of my long answers I think…so skip it if it’s not important to you.

Depends… if party getting past the guard is central to their want…I’d give them a second roll. If PCs are just doing it to see if they can backstab the guy and steal a key or uniform… he detected them…he might go into attack mode.

What did the failure represent, making a sound, like a twig snapping, or the character running past as they thought he looked away, but he turned back and saw them in all their glory?

By default all creatures for me follow Jeff Cooper color code of awareness, https://www.bsr-inc.com/awareness-color-code-chart/

Look it up, there is a lot more info on the interwebbs but that was a quick link I found explaining it.

In the case of a competent guard, he is at condition yellow, the condition you should be when driving.

He sees someone approaching, he moves to Orange, same as if you see someone swerving a distance ahead of you on the freeway and in other lanes.

He assessed the situation (AI or random roll) and reacts. This might be to attack, (condition red) hide himself ( condition red), call the alarm ( condition red), call out to the approaching individuals and ask “who goes there?” ( condition orange).

Also his range of awareness is line of sight, and range of hearing. In games terms I would just use Far.

So… if all creatures have a default awareness meter, we can we can use it for shorthand to make more informed AI script.

Bad guards might be in condition white, playing dice in the far corner

Getting back on topic. And using the above as an example. I think a few symbols and or colors could really augment this Monster AI. In providing a shorthand for complex concepts would allow great flexibility readable at a glance. But lengthen the instruction manual. Also allowing random reactions…roll a D6. Would allow for diversity of actions for the GM. Not all creatures need it. But it would greatly broaden the possibilities.


Definitely color code if that makes the most sense for you, I think that’s actually really cool for organizing your AI.

I think @KaneDriscol had the right idea for separating alert modes from the combat behaviors and providing conditions for entering these modes. Maybe you’ll find a good balance between more contextual conditions and behaviors rather than mechanical. The rules that I wrote create automatons, but if that’s inconducive to narrative and gameplay, throw out the rules or find a balance.

On the subject of random rolls: I wrote ML with a strict philosophy against random decision making. Rolling a dice and doing what it says is completely arbitrary-- there’s no logic behind that decision, and from experience it often leads to dumb decisions. BUT there is a place for randomness. ML’s core system is 100% deterministic which means it’s predictable. Therefore we need some randomness to make gameplay less predictable and more challenging. A random turn order each round, random amounts of damage, random chance to hit etc. It’s up to you to find a good balance of randomness and determinism. Hack your systems, find what works for you.

If you want more diversity of behaviors, CREATE MORE MONSTERS. Give us differentiated enemies with unique special traits and/or actions (e.g. bloodthirsty goblin, cursed goblin, wisecracking goblin) and make those distinctions clear to the player so they can make more tactical decisions about who to single out, how you engage with that enemy, who’s a bigger threat. It will make your gameplay more interesting to provide unique enemies that behave differently. In the future I’ll have lots of variant actions and traits to swap out to make unique monsters and perhaps a method of procedurally generating them too.


Using night watchman doing his patrol on the docks.

She hears a sound ( failed stealth check) and notices 2 dark clad figures about 50 feet away, she attacks the trespassers.

Selectively Random
She hears a sound ( failed stealth check) and notices 2 dark clad figures about 50 feet away, roll D6.

  1. “A” She attacks the trespassers.
  2. “O” She stands there and observes.
  3. “C” Shouts “who goes there” while grabbing the hilt of her baton.
  4. “S” Blows on her whistle to get more watchmen.
  5. “RHO” runs off to hide and observe.
  6. “RS” runs off to get more watchmen.

Written as
:six: 1:A,2:O,3:C,4:S,5:RHO,6:RS
After shorthand established
*_note_challenge in the “who goes there” or “what’s the password” sense. Not the “I challenge you to a dual” sense.

I’m not trying challenge your concept, just clarifying my view of short hand in a thoughtful implementation of randomness.

In this way, avoiding making 6 distinct versions of night watchmen, I give them 6 distinct possible reactions in one line of text.


This is super cool. I apologize if I came off as defensive, I was just laying out my point of view. This back and forth has been super productive and informative by the way, and definitely makes me think about how to expand on my rules. I also apologize in advance for the long text and going deep into what I mean about making distinct enemies to create more tactical considerations.

I made rules specifically for combat because non-combat encompasses so much non-linearity that determinism would be restrictive and unfun. So let me see what we can do about non-combat behaviors.

Let’s look at your selectively random method. It has a contextual condition (failed stealth check), and many of the actions require the GM to arbitrate what happens. (The guard observes or she shouts a challenge, but then what? The GM would decide what happens next unless we make more steps for deciding the next behavior.) Personally I would skip the rolling and let the GM choose from the available options, unless we’re playing GM-less, in which case I would let one player act as GM for the guard until we get into combat.

Let’s pivot to talk about making distinct versions of monsters for combat. It doesn’t have to be a lot of work, but what it adds to gameplay is worth it. Players can now distinguish between enemies and make tactical decisions based on those distinctions. Making behavior random does not make distinct enemies; if every enemy is the same, they just act randomly on their turn, then players will treat them exactly the same because it’s arbitrary. That’s why we naturally distinguish our enemies, we make archers and magic casters and front line warriors who behave differently from each other and our players react to them differently. But we can also make different versions of those archetypes (e.g. the guard commander has more hit points, he crits on 19 and 20) and it only adds to making interesting combat. And you can do this randomly too by making unique random results for different variants of enemies. So it can work both ways, random or deterministic.

Now let’s try this for non-combat. Guard commanders can’t be negotiated with or easily tricked. Dumb guards can be tricked or bribed. Drunk guards can be knocked out easily or have a chance of falling asleep. Making these distinctions clear, through narrative description and gameplay, means players will take that information into consideration. This is super easy to do, it makes gameplay interesting, it makes your world interesting. If a random roll is the only thing that distinguishes behavior, then we’ve only added unpredictability without adding tactical considerations for how players should react to their enemies. Thanks for coming to my TED Talk.


I’m here for the intellectual stimulation, so all good. Text in a fast fashion is a terrible medium to confer thoughts and passion.

You’re process and concepts are logical and thoughtful.

So for the most part you are engaging an AI or flowchart for combat mostly?
Now makes a bit more sense.

I had followed the AI for behavior train.
Perhaps I’ve been gaming too long, but I’m not rolling a posture randomness each round, I’m really just setting up variables.
Night watchmen from the above examples, calls her challenge. PCs respond by ignoring her, and going about her night, I’m assuming they are now on the enemy list, what does she do to people who make the enemies list and outnumber her and are not attacking anyone yet?

By the same token, orc berserker target priority closest, moves and such… closest is closest enemy? What if it’s held in a magical barrier, does it just run for the nearest enemy not moving? Or is the barrier now it’s nearest target.

GM input is going to be involved always. If wanting to replicate one player tactical war games??? Goal becomes different and really other than learning the rules of a multiplayer game a computer game really is the better medium for one player games.

I think your idea has broad appeal to take some of the mental bandwidth off the GM, not eliminating the GMs input.
Unless I’m running 20 goblins they are not going to eat up mental effort, magi with 10 different attack spells…probably not but now we are talking.
20 magi with 10 spells each…now we are talking. But to not be boring we go with 5 magi subtypes, each with various AI aspects.

Not sure I’m seeing the difference’s to randomly rolling on their attack spell list.


I can’t really tell you what happens with the nightwatchman, it sounds like she’s not actually in combat. Or if she was, then we would check her AI sheet.

I’m not sure who’s being held in a magical barrier in your scenario (or what magical barrier does). If the Orc Berserker was unable to move or to target an enemy with its attack, then it wouldn’t do anything. Orc Berserker targets the closest targetable enemy, so if an enemy couldn’t be targeted (maybe because it’s held in a magical barrier) the orc would ignore that enemy. If it can be targeted, then yes the orc would target that enemy (which presumably means the damage gets soaked by magical barrier. If it does, awesome). The barrier would not be the target unless it can be destroyed by damage and there isn’t another enemy to target, because then that would be maximizing the effectiveness of the orc’s attack.

The difference is this: If all of your magi have access to the same actions, but they only act differently based on what you roll, then they’re all the same. They’re not actually distinct, so there’s no reason why the players should prioritize attacking one Magi over another. Now you can still roll for AI, but you want to distinguish the magi with descriptors, traits, and unique actions so that players can make meaningful decisions regarding their enemies. If one of these magi can resurrect and heal its allies, the players will want to prioritize him. Or if another magi has a higher chance of rolling a freeze spell than the other magi because he’s the elite cryomancer, that’s a reason for the players to prioritize him as a target. Differentiated enemies create more interesting situations, more tactical decisions, more things for the player to consider.


Version 0.2 is now up. It’s not a huge overhaul, just some graphical changes and it’s been made to work more explicitly with D&D and Pathfinder-like systems.



Oooh! Shiny! It looks really good @MonsterAI.

I’ve been really excited about this topic because something like Monster AI is really helpful. I’m not as comfortable running complex monsters or interactions yet.

Sure, I can run a nightwatchman scenario as described above, but it’s fairly often that I don’t consider all the options and cool roleplaying/combat situations that might arise. More often it ends up being something like “Make a stealth check. Uh oh. The guard spots you and attacks.” I know that’s not a bad way to do it, but I want to improve as a DM and that seems to be my go-to.

When I ran the guards AI yesterday, it was really nice to have some guidelines and suggestions for how things could go down. It reminded me that I could call for help and bring in some more guards and it gave me a clear signal of when it made sense to do so. It was also nice to have some ideas on roleplaying moments too and that there were clear steps for me to take to achieve a particular goal (Add suspense as the PCs sneak past the guards).

From the perspective of a guinea pig, I guess what I’m looking for is some very clear building blocks to mix and match with to include in both roleplaying scenes and combats. I personally think it could work to have something selectively random to determine how the scene/combat is going play out but some deterministic steps to facilitate running through that scene or action.


That’s awesome! Just the way you’ve engaged with it and the discourse I’ve had on A.I. design and best practices has given me lots of ideas to expand upon the rules and make a larger booklet for my next version with suggestions on how to program A.I., find a balance between determinism, randomness and roleplaying, and mixing and matching rules to find the playstyle that suits you. I’m glad you found value in my pet project. It’ll only get better with more playtests and feedback.