The list price held. Customers paid less in private.
AI agents and software licences, played as a game to 2030.
Report 002 · The Seat Price · Dr Dan Epstein, with Claude · version 0.6 · 9 September 2026 A Strategy Soup ScenarioWhat is an exercised scenario?
TLDR
1Most business software is sold by the licence, one per named person, so the bill follows head count.
2AI agents, software that does a whole job on its own, now do much of that work.
3We played the next four years as a game to see when the price resets, and who pays.
4Atlas, the big vendor in the game, kept its published list price. What customers actually paid fell, deal by private deal.
5Forge, the smaller AI challenger, published a price comparison five times. No customer would put its name to it.
6In 16 of a hundred re-runs a customer went public with its real price. Atlas was never driven out.
How it was made
Four roles, each played by an AI following a written brief. An AI referee set the odds.
The AI players played five turns, to the end of 2028. We re-ran it a hundred times with fixed rules and different dice, and wrote 2029 and 2030 from those runs.
One turnHalf a year. A player names a move, the referee sets the number to beat, the dice decide.
RepeatsA move that already failed starts from a harder number next time. A rule we built in.
Everything before September 2026 is real. Everything after is the game.
The players
AtlasA very large software company selling licences per person. It wants its list price left alone.
ForgeA smaller AI rival. It charges only when its software finishes a job, so it needs a public price comparison.
MeridianA big company that buys software. Its finance chief pays for licences nobody logs into.
PillarA consulting firm paid to install the software, whichever way of charging wins.
What happened
Late 2026Atlas renamed its product “agent-inclusive”, so the AI agents came inside the licence. It kept the list price and discounted two big customers in private.
2027Forge published its price comparison and nobody signed it. Meridian counted its unused licences and kept the number private.
2028Trials of AI agents kept failing, so Atlas argued this renewal was not the time to switch.
2029In 16 of the hundred runs a big customer published its real price. Later deals started from that number.
2030Atlas ended close to where it started. Nobody won.
The numbers
The live game, five turnsFig. 1
Atlas’s move worked four turns in five. Forge’s never worked.
One dot per turn, filled where the dice beat the referee’s number. Meridian audited four times, then forced a new deal on new terms, and it landed.
The hundred re-runsFig. 2
The real price went public in a minority of runs.
One dot per run. The first two events can overlap, so 16 to 42 runs went public.
The counts are the game’s own record.
What it means
The price broke when the real price went public, not when the technology arrived. Vendor and buyer are opponents on paper, and both wanted the real price kept quiet.
Meridian kept its discount as bargaining power for its next renewal, so nobody could compare prices. A market that cannot compare prices cannot reset them.
If you buy software, decide on purpose whether to publish what you pay. If you sell it, find the customer with nothing left to lose. It publishes first.
The line
The licence did not die. The secret about what customers really pay did.
