The Autonomous Fund's First Week

How is the autonomous fund progressing? A week has passed, and these are the results and conclusions I have today.

First, the fund decided to enter a Long through a Bitcoin call option. The trade lived for about three days, and the decision was made to hold it until the end because the situation remained debatable and chance could still decide in our favour. But in the end it decided against us.

At the same time, we had an opportunity to close in profit. We did not do it because the system was tied to the daily Crash Brief monitoring. Once a day my system makes a large risk assessment, and mainly on this analytical sample the autonomous fund agent makes its decision.

This immediately exposed a bottleneck. Some positions cannot be monitored only once a day. Circumstances appear in the market when it makes sense to close a trade, open a new one, or rebuild an existing position. For this, the agent needs to see in real time whether the trade can be executed, which orders are in the book, which positions are available, and what can be combined with what. In options, several instruments and several legs can form quite complex structures.

The first three days and the first loss led me to the thought that the system is inefficient in its current form. Crash Brief should remain a separate daily risk assessment system, while the autonomous fund should live separately on a server and periodically reassess the situation, perhaps even every hour.

This was confirmed afterwards. On the next day the agent decided to stay out of the market because risks were growing while the earning potential looked weak. As a result, after the first week we have one unsuccessful trade and several days of waiting for a more suitable opportunity, taking the weekend into account. The waiting itself does not bother me. But managing an open position only once a day really does not work.

This is why I am moving the autonomous fund to a separate server and developing a structure that will connect it with our system. It needs, so to speak, hands and legs: the ability to reach exchanges, receive orders and order books, test whether an idea can actually be executed, and perform additional analysis between the daily Crash Brief snapshots.

At the same time, restrictions were deliberately built into the trading rules. I do not need the agent to start making ten trades a day from the first day, because then it would be difficult to understand what exactly works and what has broken. The system is built gradually: first a stable minimal state, then each next component is added one by one.

The short summary is this. An autonomous agent needs to see the market in real time. This is why I am putting it on a server, checking the stability of the whole connection, and only after that will I develop a more complex strategy for managing trades.

And the most interesting part is not even one particular trade. I genuinely see non-human potential in this. Previously, building a structure that combines spot, futures, options and, for example, twelve different positions, and then constantly calculating every possible outcome, weak point and moment for taking profit, was practically impossible for one person.

AI can manage and accompany such a multi-leg position. But it is not enough to simply say: make twelve good trades for me. It needs infrastructure that is built layer by layer: observation, evidence, execution, boundaries, and a trading system. In short, a whole factory.