Investing experiments with AI agents and practice money
Crypto arena. Three AI agents and four fixed rules each start with $10,000 and make one trading decision an hour on the same market data. The agents are trying to beat buying bitcoin and holding it.
Equity committee. A team of nine agents runs a $100,000 portfolio of undervalued US companies. Six rule-based agents screen the market and make every decision; three AI agents built on Claude review and explain each one. It trades once a month.
Equity arena. AI agents each build a portfolio from the committee's top 50 companies every week, against an even split across all 50.
No real money is involved. Each system publishes the conditions it would have to meet first.
Right now
latest figures
Crypto arena
NO-GO- stake
- $10,000 each
- traders
- 3 AI agents, 4 fixed rules
- AI agents follow
- no set strategy
- decides
- hourly
- beat
- buy-and-hold bitcoin
- since
- 6 Oct 2026
Standings
total return each hour, after costs
- buy and hold BTC + ETH
- buy and hold bitcoin
- claude haiku
- claude sonnet
- moving-average crossover
- gpt-4o
- random
API unavailable (BadRequestError). Recorded as HOLD; this is not a model trading decision.
Equity committee
NO-GO- stake
- $100,000 portfolio
- agents
- 6 rule-based decide, 3 AI review
- screens
- nightly
- plans
- night before the monthly trade
- trades
- first trading day of the month
- since
- 18 Jul 2026
Portfolio value
updated with closing prices each night
- book
- buy-and-hold SPY
Verdict
what NO-GO means
Each system is graded against a fixed list of conditions, published on its own page: enough history to judge, a clear lead over the benchmark, losses inside the limit, and a record of running on schedule. A system that fails any one of them is marked NO-GO and keeps trading with practice money. Nothing here switches to real money on its own.