Possibly the most interesting employee I met was Chris Woods (Lead Engineer, Hex Entertainment), who is responsible for designing the AI behind PvE battles. Unlike a typical RPG, you can't create unique enemies by giving them different clothes and a different weapon. In a card game, enemies need to ACT very specifically differently with each other, while still playing the same card game. This is a challenging undertaking.
It's important for players to not feel like they're simply playing deck after deck – that each game feels unique. So AI must be both predictable based on their deck, but unpredictable based on their personality. So Human AI may bluff, acting like they have a big play when they're really only holding a resource card (shards) instead of that instant spell card they could have. Wild animals may make very hasty or reckless decisions, simply throwing their creatures against you even in situations where they won't win.
All of this is controlled by a background set of sliders and variables, which the AI uses to decide what actions to take. An AI may be timid or bold, may love or loathe a certain race, which can lead to some fascinating scenarios such as the AI singling out specific creatures because of the type of card they are. It turns out the HEX PvE AI can be just a little bit racist. Or even A LOT racist. For example, the enemy AI may weight a 3 attack, 1 defence card with a value of "4" — making blocking it with their 3 attack, 3 defence card valued at "6" an unwise trade-off. However, if that card is say, the race of their mortal enemies, then the AI may actually artificially double the weight of that card — meaning sacrificing its 3/3 to block it makes perfect sense to the PvE AI.
Furthermore, to ensure the AI is never entirely predictable, three of the "best" possible moves (after considering personality, prejudices, and other background measures) are given values. The AI then has an 80% chance to make the "best" move, a 10% chance to make the second "best" move, and an additional 10% chance to make the third "best" move, challenging players to think about what exactly may be up the AI's virtual sleeve. Is this a bluff? Is this a real move? Did they goof and go for the less optimal play? You won't know until the turn is complete.
The PvE AI is also adaptable and can make educated guesses based off the game's current popular trends. For example, if a player places down a green shard, the AI knows there's a high percentage chance your deck will contain at least one Wild Growth (but due to the current game trends, it won't assume you have a Strength of the Redwood). Or if you play a blue shard, the AI will feel pretty confident that you'll have at least one Buccaneer. This means you can technically trick the AI by using cards it will not predict...with some fascinating results! However, as word of this way of playing spreads, eventually the AI will adjust its statistics and potential exploits will eventually wear away.
Overall, the AI is designed with a different mindset, which instead permits the AI to cheat and look at future moves and cards. This also means there's no one way to "solve" an AI, making battles consistently challenging, even for players that have large existing libraries, while remaining approachable to newbies. The system is designed to adapt. Now let's just hope they don't take over the HEX offices while we're not looking...