Can AI Beat You at Chess? (Spoiler: Yes)

In 1997, Garry Kasparov — arguably the greatest chess player who ever lived — sat across from a machine and lost. The machine was IBM's Deep Blue, a purpose-built supercomputer that weighed 1.4 tons and could evaluate 200 million positions per second.

In 2026, I'm playing chess against an AI that runs on my phone. By typing moves in English. In a chat window. And it's still beating me.

The story of how we got from there to here is stranger and more interesting than you'd expect.

A Very Brief History of Computer Chess

Computer chess has always been the benchmark for machine intelligence. It's measurable, adversarial, and humans care about it deeply — which means every breakthrough gets attention.

1950s: The dream. Alan Turing wrote the first chess algorithm. By hand. On paper. It never ran on an actual computer — the machines of the time couldn't handle it. He played through it manually, taking about half an hour per move. It wasn't good. It played at a "terrible amateur" level. But it worked conceptually.

1970s-80s: The grind. Chess programs got gradually better. They were still beatable by any halfway decent club player, but they stopped making obvious blunders. The approach was brute force — evaluate as many possible positions as your hardware allows, pick the best one.

1997: The moment. Deep Blue beat Kasparov in a six-game match. It was programmed by a team of experts who hand-coded chess knowledge, opening books, endgame databases. It was a chess engine in the purest sense — a machine built for one purpose.

2017: The revolution. DeepMind's AlphaZero taught itself chess from scratch. No human knowledge. No opening books. Just the rules and millions of games against itself. In four hours of training, it became the strongest chess player in history. It then beat Stockfish — the reigning computer champion — in a 100-game match without losing once.

AlphaZero didn't just win. It played beautifully. Creative sacrifices. Long-term positional ideas that grandmasters called artistic. A machine had gone from "brute force calculation" to something that looked like intuition.

2024-2026: The weird part. Large language models — ChatGPT, Claude, and their cousins — learned to play chess despite never being designed to. They weren't programmed with chess knowledge. They learned it from the ocean of chess games in their training data. And while they're not Stockfish-level, they're surprisingly competent. Good enough to beat most casual players. Good enough to be fun.

This is the part that interests me. Not the superhuman engine. The imperfect, conversational chess partner that plays at a human-like level and can talk about the game while playing it.

How Text Chess Actually Works

If you've never played chess by typing moves, it sounds clunky. It's not. It's actually elegant.

You use standard algebraic notation — the same notation chess players have used for centuries. e4 means pawn to e4. Nf3 means knight to f3. Bxc6 means bishop captures on c6. If you don't know notation, you can also just describe moves in English: "move my queen to d5" or "castle kingside."

A typical exchange:

You: Let's play chess. I'm white. 1. e4

AI: 1... e5. The Open Game — a classic choice. I'll match your central pawn advance.

You: 2. Nf3

AI: 2... Nc6. Defending the e5 pawn and developing toward the center. Are you planning an Italian or a Spanish?

You: 2. Bb5 — let's go Ruy Lopez.

AI: 2... a6. The Morphy Defense. Asking the question: where will you retreat to, bishop?

This is what makes text chess different from playing against Stockfish on a chess app. Stockfish just moves. The AI talks. It comments on your strategy. It explains its thinking. It teaches while it plays.

You can ask mid-game: "Why did you play that?" And it'll explain. "I moved my rook to the d-file because your pawn structure is weakened there and I'm planning to eventually push d5." That kind of real-time commentary is something you normally only get from a human coach.

The Weird Joy of Imperfect Play

Here's a confession: I don't want to play against a perfect chess engine.

Stockfish at full strength is a miserable experience for anyone below grandmaster level. It never blunders. It never misses a tactic. It plays the objectively best move in every position. Playing against it feels like playing against mathematics. You're not exploring a game — you're being ground down by inevitability.

AI text chess is different. LLMs play chess at roughly an intermediate level — say 1200-1600 ELO depending on the model and the position. They make mistakes. They occasionally miss tactics. They sometimes play creative moves that aren't quite optimal but are interesting.

This makes them fun. I can beat the AI sometimes. Not always — it catches my blunders more often than I catch its — but sometimes. That uncertainty is what makes a game a game. The outcome is in question. Both sides have chances.

And when the AI does something surprising — a sacrifice I didn't see coming, a positional idea that only becomes clear ten moves later — it feels earned. Like playing against a human opponent who just found a great move.

Strategies for Playing Against AI

After dozens of text chess games, here are my observations about how AI plays and how to exploit it:

AI loves standard positions. It's been trained on millions of games, so it's strongest in common openings and standard pawn structures. If you want an edge, play unusual openings that lead to non-standard positions. The Grob (1. g4), the Bird (1. f4), or weird gambits tend to put the AI in territory where its pattern-matching is less reliable.

AI sometimes loses track in long games. In endgames — especially complex ones with many pieces traded — AI can occasionally lose the thread. It might move a piece to a square it's already on, or forget where something was three moves ago. This is where careful, precise play can exploit its limitations.

Tactics over strategy. AI tends to be decent at long-term strategy (it's absorbed thousands of games worth of positional ideas) but can miss short-term tactics, especially longer combinations of 4+ moves. If you see a tactical sequence, go for it — the AI might not see the whole line.

Ask it to play stronger. This sounds silly, but it works. "Play at maximum strength" or "play like you're rated 1800" actually produces different results than default play. The model adjusts its move selection based on the persona you give it.

Play multiple games. AI doesn't remember previous games unless you tell it to. Each new game is fresh. But you can say "you beat me last game with that bishop sacrifice on move 22 — I've prepared something for that" and it'll engage with the continuity.

Chess as Conversation

The thing I keep coming back to is that text chess isn't really about chess. Or rather, it's about chess and everything else that makes chess interesting: the psychology, the banter, the storytelling of a game.

I've had games where the AI offered a draw in a losing position — bluffing, essentially. I've had games where it congratulated a good move with genuine analysis of why it was strong. I've had games where it narrated the whole thing like a sports commentator, building tension toward the climax.

This is what dedicated chess engines will never do. They play optimal moves in silence. The AI plays with you — as a partner in the shared experience of a game.

On a-gnt.com, the Text Chess prompt is set up to give you exactly this: a conversational chess opponent that plays at an enjoyable level, commentates the game, offers analysis when you want it, and adjusts difficulty to keep things competitive. It's not trying to be Stockfish. It's trying to be a good chess buddy who's always free for a game.

The Future of This

I think we're in a weird transitional moment. Chess AI used to mean one thing — engine, superhuman strength, brute force. Now it means something else: a conversational partner, imperfect, at your level, available at 2 AM when you can't sleep and want to think about something structured.

The dedicated engines aren't going anywhere. They're essential for serious players analyzing their games. But for the vast majority of people who play chess casually — who want the experience of playing without the experience of being relentlessly crushed — text chess through AI is genuinely the most enjoyable chess opponent I've found.

It's also, I suspect, how a lot of people will learn chess going forward. Not from books. Not from videos. But from playing games where your opponent explains what's happening in real-time, answers questions mid-game, and adjusts to your level automatically.

Kasparov lost to Deep Blue in 1997 and called it the death of something. I don't think it was. I think we're only now finding out what human-machine chess was always supposed to feel like.

Not a competition. A conversation.

Your move.