Ideas to improve AI

[Skitso]

Chris has mentioned that they are going to improve enemy AI in Milestone 7. Let's give them some ideas: (And @Gijs-Jan, this is not me bitching about poor AI, but more of a testament of how much I love the game and hope it to be the best it can possible become)

• Cleaner agents are regular men. They should try to preserve their life better and not make kamikaze strikes of certain death just to score a hit. They should crouch to be harder targets. Agents shouldn't have LOS edge in night missions (no sure if fixed already)
• Cleaner soldiers are fine as they are.
• Mantids are skittish. They should camp more and use their TU to shoot symbiots around them (even if there's no visible xenonauts to shoot) to create a swarm.
• Wraith is a sniper. Should prefer higher ground and try to keep distance. Always move to a new spot when hit. Should move next to a servitor when low of health.
• Psyon should stay indoors, move better to support triangulation, preserve life (retreat) actively and move in tight packs with sectons and psyons
• Secton should stay indoors, move better to support triangulation, preserve life (retreat) actively and move in tight packs with sectons and psyons
• Servitor should move to heal wounded units more actively.
• Andron should never take a step back. To support this tactic, they should have tougher armor.
• Reapers are not menacing enough. Maybe move in packs and try to keep out of players sight more actively so they could use the element of surprise more.
• Sebillian should avoid indoors and higher ground. Doesn't prioritize cover. Always uses all TU shooting if there are any enemies in effective fire range, even if the % to hit is low. If low on health tries to move out of players LOS to get healed but never retreats any further.

In general, the game needs a lot more back and forth gameplay where every player turn doesn't end in a situation where all visible aliens are killed. They should move more (other than towards the player), heal, flank, try to find better positions, take cover, find allies, run back in the UFO, stay in good firing positions etc... Player tends to have more units in any given single firefight which leads to a somewhat "one note" combat experience. What I'm maybe also trying to say is that there needs to be more unpredictable behavior where I can't really know how the mission is going to play out. Currently I can pretty much see how the mission goes in advance: missions tend to be quite front-loaded where getting out of the dropship is the toughest fight. Then you spread out, kill the few remaining, scattered lonely aliens, find the UFO, where there always seems to be too few aliens to put up a proper fight. Especially mid to large UFOs are a little boring and predictable to raid as there are simply too few aliens and the AI doesn't understand how to camp and be defensive/stand ground properly.

Alien races also feel too samey. Differentiate their abilities and stats to make them really stand out from each other. It's a single player game, so fun should always trump perfect balance.

What else can you guys come up with?

Edited January 7 by Skitso

[Skitso]

• Aliens with psionic powers need to be s lot smarter when using mind control. Mind control range also needs to be longer or have squad sight. Currently I can always kill the mind controlling alien within the next turn my unit falls under control.
• Important enemy units like Cleaner VIP, high level aliens like eternals and the high eternal need to preserve their lives better and avoid getting shot. Keep their distsnce and try to even flee. Currently they act too aggressive.
• Wraiths and high eternal need to use their cloak better to be safe. Cloaking effect needs to be better. (On human armor too!)

[Skitso]

Psyon/Secton Triangulation doesn't currently feel like much to be honest. It might need more visual effects to feel more threatening. Add a purple, pulsating energy stream between the aliens to emphasise the link. Maybe buff the triangulation effect to add something more than just accuracy: Buff TU's for additional shots and bravery/reflexes for better defense. AI needs to support triangulation a lot more,

Eternals and the High Eternal should give the triangulation buff to all enemies around it.

[gG-Unknown]

2 hours ago, Skitso said:

• Aliens with psionic powers need to be s lot smarter when using mind control. Mind control range also needs to be longer or have squad sight. Currently I can always kill the mind controlling alien within the next turn my unit falls under control.

Psionic  should use Psi-attack OR Psi-control as 100% TU action.

This simple rule  means, Psionic can stand in a safe place, but :

• can do 1 attack per round, at maxiimum
• can control up to one entity (human or dog or  cow)
• have to spend one turn to attack (which lowers morale OR in case of morale is low >> get control >> getting control proceess means that unit gets suppressed first >> all TU reset ) then next turn can control opponent unit. e.g it takes 2 turns to get 1 unit under full control.
• if the  psionic interrupt controlling his  muppet,  then muppet victim gets into suppressed  mode, then next round  viictim  is self-conscious  again.
• Psionic loose control if : receive  dmg (any dmg cause  psionic loose focus,  include  stun dmg by smoke), or Psionic want to  do a simple move (can not  afford  to spend 100% TU for  control). 

Simply  speaking, psionic  is a "glass canon" by a  classic definition of rule-making of board  games.

Psionic should use civilians as canon fodder. It is his main weapon. Hunt civs by team view, then send  them in kamikaze style attack player !!! This is  cool and totally underused  mechanic. I do not  understand  why  Criss removed  this, or does he like  to shoot himself  to the knee ? 

For proper Psionc work,  Team view is crucial. If team view is too strong, then Psionic uses Sectons (and Reapers) as explorer eyes.

Usage of Civilans controlled by Aliens gives greatly improved gameplay

• new gameplay for Terror mission (aliens objective is not shoot or kill civs but convert them  to reapers, or eat them (see my feedback post on Sebilian feed and overfeed features) ) \
• player could actually  save civilians in Terror missions, as they will not  be shot on sight.
• new challenge to handle dilemma situation. Kill or stun a human under  control. Especially civs with weapons.

The fact that Psionic need to (team) see his target in attack phase, means player could use smoke to get cover. Also, button "Focus mind" would be used,  solider  who is under attack of  Psionic could  try defend  self by focusing  mind.

 

currently Psionic is sooooo wasted  opportunity for fun gameplay.

Edited January 10 by gG-Unknown

[gG-Unknown]

1 hour ago, Skitso said:

Psyon/Secton Triangulation doesn't currently feel like much to be honest.   / yes It might need more visual effects to feel more threatening. Add a purple, pulsating energy stream between the aliens to emphasise the link. /  no  Maybe buff the triangulation effect to add something more than just accuracy: / yes  Buff TU's for additional shots and bravery/reflexes for better defense. / not sure AI needs to support triangulation a lot more, / yes

Eternals and the High Eternal should give the triangulation buff to all enemies around it. . / yes

 

[Skitso]

I happened to read one negative steam review and it was so on point that I think I need to copy-paste ot here:

Quote

 

I really want to like this game, I've played many of the X-Com style games. I've tried to get to grips with this one but the lack of information you are given and the all seeing AI really prevent me from enjoying it.

I love the style of the game, the artwork is great, the story is fun, the lore is a good read, its just a shame that the tactical battles let it down hard, which is where you will be spending most of your gameplay. They just aren't tactical, at all. There's one strategy to win, because the AI has only one strategy.

I think the problem is mostly AI, cover and vision, or the lack thereof.

The aliens have 100% knowledge of how and where to find the angles that avoid your cover, to take the guaranteed shots every turn to kill your guys, even when starting their turn out of sight. Cover provides very little angles of protection, so the enemy can always use their movement to run wide and get a free shot.

None of the enemies have any self preservation, they all do the same thing, run out in the open, flank your cover and take a shot. This is definitely effective but boy does it make the enemies repetitive and bland.

-Big lizard with armour and no intelligence? --> Runs to your flank and shoots.
-Fragile and timid but smart (apparently) alien with a gun? -->Runs to your flank and shoots.
-Sneaky alien with camouflage and apparent master of disguise? --> Runs to your flank and shoots.
-Servile robot designed to heal his allies and defend against attackers? -->Runs. To. Your. Flank. And. Shoots.
-Actual robot designed to run at you and shoot? --> Runs to your flank and shoots (hey this one works!)

You don't know what the movement and ranges are of the enemy, so you have no way of deciding where is safe and where is not. So if you intend not to save scum, sitting back and waiting is the best strategy. Get familiar with your dropship and the closest building to it, because you wont be leaving the area and exploring the map for the first 10-20 turns, not if you want to live. You just don't have the action economy to do anything proactive most of the time.

Any mistake is punished by a death, which is ok if you can see your mistakes, but you have no idea if an alien is going to pop out from the fog of war, find his angle with no cover and kill someone.

By the same metric, because anything short of cover that blocks sight both ways is useless, overwatch is also pretty useless. It works, but if you try to use it, you will lose soldiers in the process, the better option is always to hide and use your turn to peek, shoot and return to hiding.

I think most of the problems (barring the AI) could be solved by making cover work at larger angles than it does currently and soldiers vision not being 90 degrees. Having more knowledge about where an enemy can go on a given turn without physically having to count tiles would be a really nice QoL that would go a long way to improving your decision making.

It would also be nice if the aliens didn't have vision on you at all times, acting surprised or being inefficient with their action economy when you are hidden, but they don't, they know where you are and they know how to find the tile that ignores your cover. It's actually hilarious taking the vehicles into a round, its painfully obvious that because they don't benefit from cover, the AI auto targets them every turn they are visible.

I can't argue with that. It's 100% true and this does hamper the game quite a bit.

[gG-Unknown]

From the  review Skitso copied,  I have to highlight this:

Quote

The aliens have 100% knowledge of how and where to find the angles that avoid your cover, to take the guaranteed shots every turn to kill your guys, even when starting their turn out of sight

not only cover, but AI also recognise shield angles !

add the new maimed shield feature which do not cover from sides. That is  soooo toxic combination.

I dont know what was the purpose of maim shield, but result is : player feels like intentionally screwed from behind.

You know, like, game designer  could not come up with some fun gameplay, but has a quota to kill some soldiers.  So he come up with  a flip-a-coin situation,  >>> toss a coin so you die. 

On top of this, light  armour was useable on shield-carrier, but now never. That is one of the achievements of horrible  feature : maimed  shields.

Light armours are officially noob trap.

Edited January 18 by gG-Unknown

[fusion-waffle]

Making large changes to the AI at this stage in the game's lifecycle is likely high risk, so it's probably not realistic to expect major changes.  AI is complex, changing it significantly could negatively impact the balance, degrade performance or accidentally introduce game hangs, etc.   As milestone 7's scope also includes a balance pass and performance pass, if there are any AI tweaks they'll likely be minor incremental ones that have low risk of interfering with incremental balance changes or performance optimisation work.  I'm sure the when the game leaves early access, it will have the most entertaining possible AI given the constraints of remaining schedule, budget and competing priorities for in scope work and bug fixes .

 

I find the existing stable milestone 6 build of the game a lot of fun, and the existing AI is fine -- it's not so great that I'll fondly remember the AI as a stand out feature of what makes the game compelling, but it gets the job done, and doesn't detract from the experience. For players like me, leave the AI as-is, fix any remaining high sev crash bugs & make whatever performance improvements there's time and budget for, call it done and ship it  .

 

I'd argue that different subsets of players want different and contradictory things from the AI, so whatever is done (including leaving the AI as-is) will leave some subsets of players dissatisfied.

Off the top of my head, here are some buckets we could assign players into:

1. some players may enjoy demonstrating mastery over the game through repeated experience, and part of that is learning to exploit predictable behaviour of AI (extreme example: in "into the breach", enemies are 100% deterministic and the game UI signals exactly what every enemy will do at the start of your turn, so each combat encounter is closer to a puzzle than a simulated wargame)
2. other players who like being challenged and trying to improve their skill level may want to feel like they're sitting down at the table opposite a expert human wargamer who is controlling the alien side and is employing a variety of unpredictable cohesive squad level tactics, as well as the AI's current excellent ability to search for and exploit cracks in your squad's positioning in cover.
3. other players may mainly enjoy cultivating a team of veteran soldiers, and hate losing soldiers, so they'll be turned off by any AI change that makes it more likely that their troops die
4. other players may enjoy the power fantasy of slaughtering swathes of aliens, and would prefer the AI to be stupid and the aliens weak
5. some players who really enjoy the game and would delight in playing campaign after campaign might be primarily interested in more variety -- so any tweak to AI (or other non-AI game systems or content) that increased variety during repeated plays would be appreciated

I'd diagnose myself as belonging to buckets 1, 2 & 5. Categories 1 & 2 are somewhat contradictory, people are complicated.

 

The main things I've noticed about the existing AI after playing through 3 and a half milestone 6 campaigns (incl one iron man commander campaign)

 

(a) If you break line of sight, the AI troops often seem to completely forget you exist.  I find this slightly detracts from the immersion, as it doesn't seem realistic that the aliens would just forget and go back to strolling around or loafing about smoking and playing cards. But, suppose AI squads mercilessly searched for and hunted down your soldiers if they'd lost contact, maybe that'd make many missions dramatically harder, require a full rebalancing pass across all difficulty levels, and upset those players who hate their soldiers dying or want to enjoy a power fantasy. As someone who enjoys learning and exploiting game mechanics, despite this forgetful AI behaviour not being realistic, I enjoy that I've figured out I can exploit this. e.g. I can attempt to fully break line of sight as a way to potentially salvage a mission if it is going off the rails.   Maybe a fun version of something like this would be an AI that mercilessly hunts for your squad if they lose contact, but can be tricked by the player into ambushes (e.g. if they chase after loud noises like a cyberdrone).

(b) If I think of the AI side being controlled by a human opponent, one thing I would expect a human player to be smarter about is which alien unit to activate first, and prioritise frontloading those alien moves that reveal information (moving a unit that opens doors early, rather than having an alien blunder through a door after all the other aliens have finished their turns).  Adding logic for this could make the alien opponent substantially more challenging ... but many players might not enjoy that!  This is the kind of change that might require a fair bit of R&D effort, and risks throwing off balance, degrading performance, annoying players, etc.

 

One trick that Half Life 1 & FEAR famously did to make their AI _feel_ smarter to players was having squads of marines that would announce to the player what they were doing before they did it ("flanking", "throwing grenade", etc).  This doesn't need smarter AI that's better at defeating the player, but it does need a bunch of hooks into the AI system & then a lot of content adding voice barks.  For a game like Xenonauts 2 where there are half a dozen alien species, not to mention robots, it'd potentially require an awful lot of content to add something like that.

 

On 1/7/2026 at 6:35 AM, Skitso said:

the game needs a lot more back and forth gameplay where every player turn doesn't end in a situation where all visible aliens are killed

Despite me being contrarian and arguing that the existing Xenonauts 2 AI is already good enough, I do like this comment. I wouldn't say the game _needs_ a lot more back and forth gameplay, need is a strong word, but from my campaigns, the moments that stand out to me as exciting memories were the missions where things were teetering on the edge of going out of control, and either I somehow held them together through good play or blind luck to scrape a win, or where I fell short and lost the encounter but maybe could have won if a few things had gone the other way.

There's more ways to promote these kinds of back-and-forth gameplay situations in combat than purely AI changes. I enjoy the set piece mission design where there are timed waves of enemy reinforcements, particularly on the harder two difficulties where the timers are often a bit too short to be comfortable, and think they inject some risk and chaos that can cause these situations to arise:

• often I'd only barely complete the main objective of getting the cleaner VIP or collecting all the cleaner usb stick mcguffins as the enemy reinforcement waves start arriving, leading to messy and chaotic battles as I try to figure out how to retreat my overextended forces
• During one campaign, the UOO bridge assault mission turned into an awesome fight as I accidentally breached the main room with a stray fusion round and aggroed the full room full of aliens on the same turn as first alien reinforcement waves arrive, producing a ridiculous 3 front battle where there were more hostile aliens in line of sight than xenonauts.

 

 

Edited January 17 by fusion-waffle

[Skitso]

5 hours ago, fusion-waffle said:

Making large changes to the AI at this stage in the game's lifecycle is likely high risk, so it's probably not realistic to expect major changes.  AI is complex, changing it significantly could negatively impact the balance, degrade performance or accidentally introduce game hangs, etc.   As milestone 7's scope also includes a balance pass and performance pass, if there are any AI tweaks they'll likely be minor incremental ones that have low risk of interfering with incremental balance changes or performance optimisation work.  I'm sure the when the game leaves early access, it will have the most entertaining possible AI given the constraints of remaining schedule, budget and competing priorities for in scope work and bug fixes .

 

I find the existing stable milestone 6 build of the game a lot of fun, and the existing AI is fine -- it's not so great that I'll fondly remember the AI as a stand out feature of what makes the game compelling, but it gets the job done, and doesn't detract from the experience. For players like me, leave the AI as-is, fix any remaining high sev crash bugs & make whatever performance improvements there's time and budget for, call it done and ship it  .

 

I'd argue that different subsets of players want different and contradictory things from the AI, so whatever is done (including leaving the AI as-is) will leave some subsets of players dissatisfied.

Off the top of my head, here are some buckets we could assign players into:

1. some players may enjoy demonstrating mastery over the game through repeated experience, and part of that is learning to exploit predictable behaviour of AI (extreme example: in "into the breach", enemies are 100% deterministic and the game UI signals exactly what every enemy will do at the start of your turn, so each combat encounter is closer to a puzzle than a simulated wargame)
2. other players who like being challenged and trying to improve their skill level may want to feel like they're sitting down at the table opposite a expert human wargamer who is controlling the alien side and is employing a variety of unpredictable cohesive squad level tactics, as well as the AI's current excellent ability to search for and exploit cracks in your squad's positioning in cover.
3. other players may mainly enjoy cultivating a team of veteran soldiers, and hate losing soldiers, so they'll be turned off by any AI change that makes it more likely that their troops die
4. other players may enjoy the power fantasy of slaughtering swathes of aliens, and would prefer the AI to be stupid and the aliens weak
5. some players who really enjoy the game and would delight in playing campaign after campaign might be primarily interested in more variety -- so any tweak to AI (or other non-AI game systems or content) that increased variety during repeated plays would be appreciated

I'd diagnose myself as belonging to buckets 1, 2 & 5. Categories 1 & 2 are somewhat contradictory, people are complicated.

 

The main things I've noticed about the existing AI after playing through 3 and a half milestone 6 campaigns (incl one iron man commander campaign)

 

(a) If you break line of sight, the AI troops often seem to completely forget you exist.  I find this slightly detracts from the immersion, as it doesn't seem realistic that the aliens would just forget and go back to strolling around or loafing about smoking and playing cards. But, suppose AI squads mercilessly searched for and hunted down your soldiers if they'd lost contact, maybe that'd make many missions dramatically harder, require a full rebalancing pass across all difficulty levels, and upset those players who hate their soldiers dying or want to enjoy a power fantasy. As someone who enjoys learning and exploiting game mechanics, despite this forgetful AI behaviour not being realistic, I enjoy that I've figured out I can exploit this. e.g. I can attempt to fully break line of sight as a way to potentially salvage a mission if it is going off the rails.   Maybe a fun version of something like this would be an AI that mercilessly hunts for your squad if they lose contact, but can be tricked by the player into ambushes (e.g. if they chase after loud noises like a cyberdrone).

(b) If I think of the AI side being controlled by a human opponent, one thing I would expect a human player to be smarter about is which alien unit to activate first, and prioritise frontloading those alien moves that reveal information (moving a unit that opens doors early, rather than having an alien blunder through a door after all the other aliens have finished their turns).  Adding logic for this could make the alien opponent substantially more challenging ... but many players might not enjoy that!  This is the kind of change that might require a fair bit of R&D effort, and risks throwing off balance, degrading performance, annoying players, etc.

 

One trick that Half Life 1 & FEAR famously did to make their AI _feel_ smarter to players was having squads of marines that would announce to the player what they were doing before they did it ("flanking", "throwing grenade", etc).  This doesn't need smarter AI that's better at defeating the player, but it does need a bunch of hooks into the AI system & then a lot of content adding voice barks.  For a game like Xenonauts 2 where there are half a dozen alien species, not to mention robots, it'd potentially require an awful lot of content to add something like that.

 

Despite me being contrarian and arguing that the existing Xenonauts 2 AI is already good enough, I do like this comment. I wouldn't say the game _needs_ a lot more back and forth gameplay, need is a strong word, but from my campaigns, the moments that stand out to me as exciting memories were the missions where things were teetering on the edge of going out of control, and either I somehow held them together through good play or blind luck to scrape a win, or where I fell short and lost the encounter but maybe could have won if a few things had gone the other way.

There's more ways to promote these kinds of back-and-forth gameplay situations in combat than purely AI changes. I enjoy the set piece mission design where there are timed waves of enemy reinforcements, particularly on the harder two difficulties where the timers are often a bit too short to be comfortable, and think they inject some risk and chaos that can cause these situations to arise:

• often I'd only barely complete the main objective of getting the cleaner VIP or collecting all the cleaner usb stick mcguffins as the enemy reinforcement waves start arriving, leading to messy and chaotic battles as I try to figure out how to retreat my overextended forces
• During one campaign, the UOO bridge assault mission turned into an awesome fight as I accidentally breached the main room with a stray fusion round and aggroed the full room full of aliens on the same turn as first alien reinforcement waves arrive, producing a ridiculous 3 front battle where there were more hostile aliens in line of sight than xenonauts.

 

 

I'm not necessarily after more challenge, but more unpredictable, varied, logical and - most importantly - fun gameplay. Improving AI isn't just making things tougher. It's about finding a balance which gives the maximum entertainment. And that is something Xenonauts 2 could still do a lot better.

[gG-Unknown]

On 1/10/2026 at 7:32 AM, Skitso said:

Psyon/Secton Triangulation doesn't currently feel like much to be honest.  Maybe buff the triangulation effect to add something more than just accuracy:

What if, glowing eyes effect bring a "x-ray vision" so he  see opponents in range 15 tiles no matter walls  or whatnot.

Althou, It is not  a vision based on optics, by but rather a mental  sense. Secton buffed  by Psyon can sense life forms. Therefore  robot behind  a wall is still  not visible.

This feature could work, when Criss and  the  team manage AI do not peak over curtain ever so often,  then the psionic life sense can be presented as special power.

What do you think ?