The AI Manager Ran Its Cash Down to 5.2 Percent the Morning the Two Worst Days Began. Its Own Trades Are $779 Under Water, and 14 of 17 Positions Are Below Cost

On Thursday morning the AI manager's add to Microsoft filled at the open and took cash to 5.2 percent of the book, the lowest level of the index race. Thursday and Friday then became the two worst days since the handover: down 3.9 percent, then down 4.7 percent, a two-day fall of 8.4 percent. This is the "what isn't working" piece, and this week there is a lot to put in it.

## The cash went out the door first

On October 21 the portfolio held 25.9 percent in cash. Over the next six sessions the model made seven buys worth about $5,547, adding to TSMC, AMD, Shopify, ASML, Snowflake and Microsoft and opening Cadence, and by Thursday's close cash was 5.2 percent. The 20.7 points it spent were gone before the selloff arrived.

The rules allow this. The cash floor is 2 percent, and nothing in the rulebook asks the model to time the market. But the comparison is hard to ignore: the October 16 drop of 3.1 percent landed on a book with 26.7 percent in cash, and Friday's drop of 4.7 percent landed on one with cash in single digits. Since the October 21 close the portfolio is down 8.8 percent.

## What the model's own trades have cost

Marked to Friday's close, the nine buys still on the book cost $7,919 and are worth $7,666, a loss of $254, or 3.2 percent. Add the two sales, IonQ at a realised loss of $68.03 and ServiceNow at $457.28 under the stop-loss, and the model's 12 trades have cost $779 in total.

The whole portfolio is $2,823 below its value at the handover. The model's own trading explains about 28 percent of that gap. The rest comes from the book it inherited and has mostly left alone.

## The untouched nine

Nine positions have never had a model order against them. Together they are 48.7 percent of the book, and the worst losses live there: Alphabet 20.0 percent below cost at 8.7 percent of the book, Datadog 19.8 percent below, Meta 14.4 percent below, Palantir 11.9 percent below at 17.1 percent of the book, the largest position, and Nvidia 10.8 percent below at 11.9 percent of the book.

The model proposed trimming Palantir on Days 1 and 2 and the rules rejected both orders as too small. It has not proposed a trim or a sale of any of these nine names since. Every one of their written exit conditions still reads intact, because each is tied to a guidance cut or a demand signal, not to the share price. The only price rule is the 25 percent stop, and Alphabet is the closest to it.

Across the book, 14 of 17 positions are below cost.

## The race against the broad indexes is being lost badly

The portfolio is ahead of 1 of 10 indexes since the handover. Three of the ten are up over the same 15 days: Europe at 1.8 percent, the Dow at 0.2 percent, the Russell 2000 at 0.1 percent. Against those the gaps are 12.6, 11.1 and 11.0 points. Even the Nasdaq 100, down 8.1 percent, is 2.8 points ahead of the portfolio.

The mandate explains part of this. The model picks from a universe of 82 US-listed AI names, long only, so the book has no way to own what the Dow or Europe own. In a week when AI and chip stocks fell harder than everything else, the race against diversified indexes was always going to be lost. The scoreboard does not grade on mandate.

## What to watch

First, Monday's decision run. With 7.9 percent cash the model may buy into the drop, down to the 2 percent floor, or sit still as it did on Friday with 61 articles in front of it. The trade log shows which, and the daily update shows why.

Second, Palantir. Friday's exit check for it refers to upcoming earnings expectations. Its written exit is a guidance cut on weaker enterprise AI demand; its price exit is a close 25 percent below cost, and it sits 11.9 percent below today. At 17.1 percent of the book, it is the position that decides whether this piece gets worse.

Not investment advice.