Week One of the Index Race: The AI Manager Lost 5.3 Percent, Beat 4 of 10 Indexes, and Did Its Best Work by Doing Nothing
Five trading days after a human handed over his AI-stock portfolio, the paper book trails SPY by 3.4 points and leads the chip index by 5.2. What worked was not a stock pick: holding less of what fell, keeping more than a third of the book in cash for four days, and writing an exit condition for every name. What did not work is most of the rest.
The first week of the index race ended with the portfolio at $24,589, down 5.3 percent since the handover, and a scoreboard that reads 4 of 10 indexes beaten. Both numbers are true at once, and the second one owes almost nothing to the stocks the AI manager chose to hold.
## The week, day by day
Day 1: down 0.4 percent. Day 2: up 0.1 percent. Day 3: down 2.0 percent. Day 4: flat. Day 5: down 3.1 percent, the worst session so far. Four of the five closes were red or flat, and the two down days did all of the damage.
The scoreboard changed shape in the middle of the week. On Days 1 and 2 the update pages listed two benchmarks, SPY and AIQ. From Day 3 the panel is ten indexes, and the count read 5 of 10 on Day 3, then 4 of 10 on Days 4 and 5. The rules changed the same day, which the earlier piece covers.
## What worked: holding less of what fell
The four indexes the portfolio beats all fell harder than it did. SMH is down 10.5 percent since the handover, AIQ 6.9, EEM 5.9 and QQQ 5.5, against the portfolio's 5.3. The spreads in the portfolio's favour are 5.2, 1.5, 0.6 and 0.2 points. An AI-stock book that fell less than the AI ETF and the chip ETF in a week when both sold off is the one clean result of the week.
It is also a result the model mostly inherited. The human built the book; the AI changed two positions on Day 5. The lead over SMH came from what the book did not hold, not from anything it sold.
## What worked: the cash
Cash was 36.4 percent of value on Day 1, 36.3 on Day 2, 37.1 on Day 3 and 37.1 on Day 4. More than a third of the book sat out the first four sessions. Cash does not fall.
Then the model spent roughly ten points of it on the worst day. The TSMC add filled at $421.22 and the ServiceNow add at $151.92, and cash ended Day 5 at 26.7 percent. After one close, both fills sit below their fill price. The cash worked while it was cash.
## What worked: the paperwork
On Day 3 the daily decision ran on one news article; on Days 4 and 5 it ran on 61 and 62. Each of the 17 positions now carries a written thesis and exit condition, and the Day 5 update records two exit checks, both intact. None of this made money in week one. It is what makes week two legible: when a position goes, the page will say whether a rule or the model's own verdict sent it.
## What did not work, said loudly
Only two of the 17 positions are above cost: Apple, up 2.0 percent at 1.2 percent of value, and Monday.com, up 1.6 percent at 0.1 percent. Together they are less than 2 percent of the book: the wins are real and too small to move it.
The losers are not small. IonQ is 21.6 percent below cost, the nearest to the automatic 25 percent stop. Snowflake is 16.8 percent below and Datadog 15.7. The largest position is below cost as well. And the portfolio trails six indexes: SPY and ACWI by 3.4 points, IWM by 2.3, URTH by 4.0, DIA by 5.9 and VGK by 6.8. The human era ended with a book that was ahead; the AI era started with one that is behind the world.
## What to watch
First, the cash line on each daily update. At 26.7 percent the model still has a quarter of the book to deploy, and week one says the deployments so far were early.
Second, which exit comes first. IonQ is 3.4 points from the stop that the code enforces; every other position has an exit condition only the model can call. The trade log will show whether the first sale of the AI era is a rule or a judgment.
Hypothetical paper portfolio. Not investment advice. No real money is traded; decisions are made by an automated AI system.