I dont usually write at those sites; but just created an account to reply this thread, as I saw many things posted that are simply not true.
I am a intermediate-level C# programmer; and legitimate wow-player since 2008, now playing at classic.
I always played as a priest, healing hardcore guilds; but I got really pissed when I simply had no gold to buy a flask to raid. (distilled wisdom). Then I decided to write a bot to grind stuff for me. I had no previous wow-bot-programming knowledge.

It took me 3 weeks from scratch to have a fully functional bot, that can walk following a list of waypoints of map coordinates, and kill mobs better than an average player, including looting and skinning.

So there is not any technical "impenetrable barrier" to any average programmer to write such program. From my (programmer) point of view, limitations are heuristic detection (e.g.Playing 48 hours nonstop), direct proccess detection (like an anti virus would), followed by banwaves, and legal persuit aginst companies that used to sell this kind of aid.

It's true that memory reading is not a reality anymore (as blizzard is using "shadowed" / random memory address ) since 2017, but a new generation of very efficient pixel-reading bots are the new reality.

Map coordinates, player location, player facing angle, player and target health and mana are available to any addon via lua code. The interface wow-bot now is made via addons, that draw a small multi coloured square (e.g 10x10 = 100 pixels) at player screen, and each pixel will carry colour code of 3x 255 (R-255-G-255-B-255): a single pixel at the screen for instance can carry player health at R,player mana at G, and player "facing" ate B, and another pixelwill carry X,Y coordinates and combat status. The bot software reads those pixel colors, and send the info to the bot itself, that will use that info to move the player and play.

LUA functions interface (addons) can supply a plethora of info that is more than enough to automate playing proccess: Player coordinates, Player "facing direction", player health, mana, combat status, moving speed, spell "isinrange", incorrect facing of the target after trying to cast stuff (e.g. you are facing the wrong way), target health, mana, cooldowns, etc. That is more than enough to make a fully functional virtual player (aka bot), without reading a single memory byte. Yet, detecting stuff (e.g. mobs) at the screen is way easier than complex visual machine stuff. You can simply read the red pixels from mobs target bars to know where mobs are, for example.

Its really simple to do, so there is no real challange. So whenever you guys think of "technical limitations" on botting, remember that there is no such thing. limitations are in fact from banning botters.