The prisoner's dilemma is a simple setup: two players, each choosing to cooperate or betray, where betraying looks smart in the short run but wrecks you both if you both do it. In 1980, political scientist Robert Axelrod ran two computer tournaments where scientists submitted strategies to play the game 200 times in a row against each other.
The winner of both was the simplest program entered: tit-for-tat, submitted by Anatol Rapoport. It cooperates on move one, then just copies whatever the other player did last round. The greedy "betray first" strategies racked up quick points and then collapsed, because everyone learned to punish them.
Axelrod found the winning strategies shared four traits, and these are the ones worth stealing for real decisions:
Source for the tournament and the four properties: Axelrod's tournaments, PLOS One review and the Axelrod-Python tournament docs.
The trap in almost every hard call is that you model it as a one-off. "Should I undercut them on price?" feels like a single move, so the aggressive answer looks free. But business is repeated: the same competitor, client, or partner is still there next quarter, and they remember what you did. The one-time win quietly costs you the relationship, the trust, or the price floor for years.
Claude is genuinely good at this because it can hold every branch open at once, name the second and third move you can't see yet, and stay unemotional while you're the one with skin in the game. It is not predicting the future. It is forcing you to look past move one.
Paste this into Claude (works in Claude.ai or the Claude app), then fill in your real numbers. The more honest your payoffs, the sharper the answer.
I'm stuck on a real business decision and I want you to war-game it using game theory, specifically the repeated prisoner's dilemma. The decision: [describe it in one line, e.g. "do I undercut competitor X on price or hold my rate"]. The players: me and [the other party]. The payoffs as I see them: - If we both cooperate (both hold): [what happens] - If we both defect (both undercut / race to the bottom): [what happens] - If I defect and they cooperate: [what I gain, what they lose] - If I cooperate and they defect: [what I lose] Important: model this as a REPEATED game, not a one-off, because I'll be dealing with this party again. Walk me through: (1) what the naive one-off move would be and why it's a trap, (2) how the game changes when it repeats, (3) the strategy that compounds in my favour over many rounds, and (4) the exact first move you'd recommend and what signal it sends. Point out where a "win once" move quietly costs me the relationship.
Same decision, different angles. Run whichever fits.
Now play the other side. You are [the competitor / partner]. Given my likely move, what's your best response, and your response to that? Run three rounds of move and counter-move so I can see where this actually lands, not just round one.
Re-run the analysis but change one assumption: what if this is the LAST time I'll ever deal with this party (a true one-off)? Show me how the right move flips, so I understand exactly how much the "we'll meet again" part is driving my decision.
Stress-test your own recommendation. Argue the strongest case AGAINST the move you just told me to make. What would have to be true for the opposite choice to be right, and what's the cheapest thing I could do to find that out before I commit?