When the Machines Clock In

Same seeds, fiat money, five AI levels. With no AI, output reaches 1.4× its starting level in 20 years. At Modest (the app default) it reaches 3.5×, at Substantial 6.2×, at High 21× and at Extreme 125×. Jobs go the other way: year-20 unemployment rises from 4.6% to 47% at Substantial.

An AI agent’s real cost halves every two years by default, and the Substantial preset adds an energy breakthrough in year 8. Barry’s view is that agent prices should fall dramatically as agents become more capable and new power sources such as fusion arrive. We compared four paths under Substantial AI. Without agents, output reaches 3.1×. With agents whose cost barely falls (half-life 20 years), 5.96×. On the baseline path, 6.24×. With a one-year half-life and a 90% fusion cut in year 3, 6.22×.

With Substantial AI we turned two dials. Dropping UBI cuts typical consumption from 3.51× to 2.97×. Tripling it to 75% of AI-driven GDP lifts it to 4.57×, but inflation rises from 0.5% to 3.2% a year. Spreading AI profits broadly (ownership concentration 0.1 instead of 0.8) reaches 4.77×, with inflation at 0.7%.

With choice on (speed 0.05) and default trust, bitcoin ends year 20 at 16% of money with no AI, 15% with Substantial AI, 15% with cheap agents and 11% at Extreme. So AI nudges the mix slightly toward fiat, and cheap agents change nothing. Trust is what moves it: from 10% at trust 0 to 53% at trust 2. At this speed every trust level peaks in year 7 (81% at trust 2), when the BTC/USD path first spikes.

The tipping point. Pay currency is chosen economy-wide: when bitcoin’s score beats fiat’s, every household and firm switches to bitcoin pay together. Firms pay their agent-compute bills from their dollar account, so after the switch agent sales almost stop and output follows the no-agent path: 6.3× at trust 1.2, 3.1× at trust 1.6 (no agents: 3.1×). Setting legal-tender pull to 0 has the same effect (3.1×).

On a pure bitcoin standard with Substantial AI, prices fall 10.0% a year and end at 8% of their starting level; debt deflation brings 13 bank failures in the median run (11–18). That is a limit of the bitcoin mechanism in this model, not of the hybrid. The hybrid never gets there, because goods stay priced in dollars and fiat policy keeps running. Its average inflation is 2.3%, against 0.5% under pure fiat.

Lowering legal-tender pull doesn’t bring deflation either: it does the opposite. With no AI, inflation rises from 3.9% at pull 2 to 8.3% at pull 0. A likely reason: bitcoin rides a rising BTC/USD path, so a bigger bitcoin share means richer holders with more dollars to spend. Bitcoin-heavy settings bring a few bank failures (median 5 at trust 2 and pull 0, no AI).

With choice off, holdings never move, so bitcoin’s share grows with its price to 52% under Substantial AI. At speed 0.01, people follow slowly and it ends at 39%. At 0.05 or faster, the mix holds near 15%. Under Substantial AI, the poorest fifth holds 1.8% of wealth with choice off, 1.8% at 0.01 and 2.6% at 0.05 (20 of 20 paired seeds higher). The wealth Gini falls from 0.70 at 0.01 to 0.65 at speed 1. Output and typical consumption barely change.