AI Brewed Me Two Commander Decklists

Commander is often called the format of limitless creativity. You can do anything in Commander, with your only limitations being your knowledge of the card pool (and thankfully we have Scryfall to fix that). Commander decklists are often creative extensions of the deckbuilder; in fact, you can learn a lot about a person by looking at their decklist. So what happens when you ask AI to make you the perfect Commander deck?

Well, this week I got a little bored and thought, why not test this out? For this little experiment, I asked ChatGPT to create two Commander decklists for me: the perfect decklist and the worst decklist. Can a computer truly make something perfect? Can it make something dreadful? Let’s see. 

The Perfect Decklist

When I prompted the AI to make me the perfect Commander decklist, I was given a brief explanation of this deck and why it should be considered perfect.

As far as I can tell, this seems like a great decklist description. Decks want card draw, win conditions, powerful plays, and a suite of interaction. My hopes were starting to get high. Let’s see if the AI met the mark, shall we?

Zur the Enchanter

ChatGPT

Test deck on 06-30-2023

Commander

Creatures (15)

• 1 Solemn Simulacrum
• 1 Trinket Mage
• 1 Teferi, Mage of Zhalfir
• 1 Venser, Shaper Savant
• 1 Mulldrifter
• 1 Sower of Temptation
• 1 Glen Elendra Archmage
• 1 Ethersworn Adjudicator
• 1 Sun Titan
• 1 Consecrated Sphinx
• 1 Phyrexian Metamorph
• 1 Snapcaster Mage
• 1 Baleful Strix
• 1 Archaeomancer
• 1 Torrential Gearhulk

Lands (20)

• 1 Strip Mine
• 1 Wasteland
• 1 Reflecting Pool
• 1 City of Brass
• 1 Underground Sea
• 1 Tundra
• 1 Scrubland
• 1 Polluted Delta
• 1 Flooded Strand
• 1 Maze of Ith
• 1 Watery Grave
• 1 Godless Shrine
• 1 Hallowed Fountain
• 1 Academy Ruins
• 1 Exotic Orchard
• 1 Reliquary Tower
• 1 Marsh Flats
• 1 Bojuka Bog
• 1 Command Tower
• 1 Mana Confluence

Spells (48)

• 1 Sensei's Divining Top
• 1 Nevinyrral's Disk
• 1 Counterspell
• 1 Vampiric Tutor
• 1 Grim Monolith
• 1 Reanimate
• 1 Propaganda
• 1 Swords to Plowshares
• 1 Necropotence
• 1 Mana Vault
• 1 Animate Dead
• 1 Sol Ring
• 1 Demonic Tutor
• 1 Fellwar Stone
• 1 Mana Crypt
• 2 Aura of Silence
• 1 Lightning Greaves
• 1 Phyrexian Arena
• 1 Talisman of Progress
• 1 Talisman of Dominance
• 2 Mystic Remora
• 2 Vanishing
• 2 Rhystic Study
• 1 Pithing Needle
• 2 Copy Enchantment
• 1 Dimir Signet
• 1 Orzhov Signet
• 1 Azorius Signet
• 2 Paradox Haze
• 1 Pact of Negation
• 1 Austere Command
• 1 Oblivion Ring
• 1 Path to Exile
• 1 Swiftfoot Boots
• 1 Chromatic Lantern
• 1 Supreme Verdict
• 1 Cyclonic Rift
• 1 Detention Sphere
• 1 Swan Song
• 1 Toxic Deluge
• 2 Smothering Tithe

Human Analysis

The commander the AI chose was Zur the Enchanter, a popular enchantment-based commander known for its power and often oppressive gameplay. When I saw this, I genuinely had high hopes. I’d expected something with five colors, but Esper enchantments wasn’t bad either. Then I saw the list and realized just how flawed it was going to be.

First off, the deck only has 83 cards. It’s not even complete. Heck, the AI generated list even included doubles of cards like Mystic Remora, Vanishing, Smothering Tithe, and Aura of Silence, counting them as both “Spells” and “Enchantments”. In fact, when we remove all the doubles, the decklists clocks in at 76 cards, which is far from perfect or complete.

Even if we look past this, I have to say it plainly: this deck sucks. The creatures in this list have virtually zero synergy with Zur. The creatures are more focused around instants and sorceries with pieces like Archaeomancer, Snapcaster Mage, and Torrential Gearhulk. The instant and sorceries themselves are standard, run-of-the-mill Esper good stuff. They’re nice spells but ultimately generic and don’t really support the commander or the theme of the deck.

The artifacts are also generic, and the enchantments are a bit of a mess. Sure, there are some good ones there, but they ultimately don’t really do much past provide you value. Most Zur decks fiddle around with Auras in some shape or form. This list has none. No Ethereal Armor, no All That Glitters, no Steel of the Godhead. While EDHREC isn’t always the end all be all for commander decklists, there is little to no crossover between the average EDHREC Zur list and the one the AI made. It didn’t even pull data from the #1 Commander deckbuilding website. And to top it all off, this incomplete decklist (at its cheapest) is $2099.07….that is far from perfect.

The Worst Decklist

For funsies, I decided to see if an AI could make a terrible Commander decklist. I was unsure of what to expect, but I knew it was going to be bad. This time around, I was definitely not disappointed. The AI gave the following description about the deck before showing the final list:

Chimney Imp

ChatGPT

Test deck on 06-30-2023

Commander

Creatures (15)

• 1 Drudge Skeletons
• 1 Craw Wurm
• 1 Clockwork Beast
• 1 Zephyr Falcon
• 1 Grizzly Bears
• 1 Feral Shadow
• 1 Wood Elemental
• 1 Moss Monster
• 1 Volrath's Shapeshifter
• 1 Scaled Wurm
• 1 Aven Fisher
• 1 Scornful Egotist
• 1 Norin the Wary
• 1 Thraben Heretic
• 1 Chimney Goyf

Lands (15)

• 1 Remote Isle
• 1 Smoldering Crater
• 1 Drifting Meadow
• 1 Tranquil Thicket
• 1 Lonely Sandbar
• 3 Sorrow's Path
• 1 Rupture Spire
• 1 Sejiri Steppe
• 1 Transguild Promenade
• 1 Crumbling Vestige
• 1 Holdout Settlement
• 1 Forsaken Sanctuary
• 1 Geier Reach Sanitarium

Spells (18)

• 1 Cursed Scroll
• 1 Stasis
• 1 Time Stop
• 1 Armageddon Clock
• 1 Last Chance
• 1 Endless Whispers
• 1 Grafted Skullcap
• 1 Anarchy
• 1 Teferi's Realm
• 1 Reins of Power
• 1 Delusions of Mediocrity
• 1 Psychic Battle
• 1 Touch of Invisibility
• 1 One with Nothing
• 1 Sorcerer's Strongbox
• 1 Psychic Surgery
• 1 Steam Augury
• 1 Worst Fears

Human Analysis

What in God’s good name is this deck? It definitely follows the prompt. It certainly is awful. The deck doesn’t use a legendary creature for the commander. Of course, Pauper Commander exists, and so do Rule 0 decks, but wow… Chimney Imp? Really? The decklist is once again incomplete, capping out at 49 cards. The list doesn’t even follow the color identity rule. The commander is black, but the list is all five colors. Did I mention that this list also runs three copies of Sorrow’s Path? Three! 

The creatures leave a lot to be desired. Grizzly Bears, Clockwork Beast, and Moss Monster are bad…which technically does fit the criteria I asked for. To be fair, the AI was actually pretty good at choosing wacky and bad spells. Anarchy, Worst Fears, One with Nothing, Time Stop, Delusions of Mediocrity, and Stasis are hilarious and definitely bad. While I do believe that there is no such thing as a bad card, these definitely do not shine here. There is no win condition, and no clear path or direction. This decklist is absolutely insane and bad…which means the AI succeeded.

Experiment’s End

So what did we learn? Well, AI is absolutely trash at making cohesive Commander decklists. Both lists were very jumbled, with no clear win conditions or synergies. Both decks also didn’t hit the 100-card criterion for a Commander deck. The lists also included multiple copies of cards, which also violates the rules of Commander (unless stated otherwise, of course). The only good thing the AI did was make a horrible Chimney Imp deck, which I find hysterical.

All in all, when it comes to deckbuilding, it’s clear to see that the uniqueness and ingenuity of humans beat out the AI by a mile. Happy generating, deckbuilders.