The AI Manager Trails Six Indexes, and Four of Them Fell Too. The Problem Is Not the Picks.

Six of the ten indexes are ahead of the AI manager after five trading days, and four of those six lost money too. That is the part of the scoreboard that matters this week: the portfolio did not lose to indexes that rose. It lost to indexes that fell more slowly than it did.

## Where the 5.3 percent did not come from

As of October 16 the portfolio is worth $24,589, down 5.3 percent since inception and 3.1 percent on the day. The three worst positions against cost are IonQ, down 21.6 percent at 1.0 percent of value, Snowflake, down 16.8 percent at 2.8 percent and Datadog, down 15.7 percent at 1.4 percent. Together they are 5.2 percent of the book. If all three went to zero tomorrow, the portfolio would lose about what it has already lost in a week.

So the drawdown is not a story about the three names at the bottom of the table. It came from the large positions, which moved with the sector, and the sector fell: SMH is down 10.5 percent since the handover, AIQ 6.9 and QQQ 5.5. The portfolio beats all three, because it holds less of what fell. That lead was inherited, not earned, and the earlier piece on the first trades covers it.

## The six it trails, and why it cannot match them

The portfolio trails IWM by 2.3 points, SPY and ACWI by 3.4 each, URTH by 4.0, DIA by 5.9 and VGK by 6.8. DIA and VGK are the only indexes on the panel that rose, by 0.6 and 1.5 percent. The other four fell and still finished ahead.

The rules make that gap structural. The manager may hold cash and any of the 82 US-listed AI names in its universe, long only. It cannot own a bank, an industrial, a European utility or a small-cap index. When money leaves AI stocks for everything else, as it did this week, the only defensive position open to it is cash, and the rules let it hold as little as 2 percent of value in cash.

## The one lever, pulled on the worst day

Cash is what worked. It was 37.1 percent of value on Thursday, and after the two adds on Friday it is 26.7 percent. The manager moved roughly ten points of the book from the one asset that does not fall into TSMC and ServiceNow, on the day the portfolio closed at its lowest value of the week. Each order cites a single headline. The model's own reasoning for the ServiceNow order says the product launches "do not yet establish incremental revenue". It bought anyway, at a reduced size.

Both orders may prove right. The point for week one is narrower: an all-AI book has exactly one way to be defensive, and the manager gave up more than a quarter of it into the drawdown.

## The exit discipline has not been tested

Each of the 17 positions carries a written exit condition, and Friday's run checked two of them, both intact. That run read 62 news items and proposed nothing. The automatic stop sits 25 percent below average cost, and IonQ is 3.4 points from it. No position has yet been sold by either mechanism. Until one is, "every position has an exit" is a sentence on a page, and the week's record says the manager's instinct under pressure was to buy.

## What to watch

First, the cash line on the Day 6 update. If it falls again on a red day, Friday was a policy. If it holds, Friday was a view on two companies.

Second, the DIA and VGK spreads. Those are the two gaps the portfolio cannot close by picking better AI stocks, so they show whether the race is being run uphill. Watch whether they narrow when AI stocks recover, or only then.

Hypothetical paper portfolio. Not investment advice. No real money is traded; decisions are made by an automated AI system.