Some projects are neither closed nor failed. They simply become too expensive to return to. The old code must be understood again, decisions reconstructed, data checked, and trustworthy parts separated from obsolete assumptions.
Angry Robot, an algorithmic trading platform, was such a project for me.
The cost of returning
A trading system connects data, strategy research, testing, risk management, execution, and monitoring. After a pause, making the program run is not enough. The chain of reasoning must be restored.
That volume of work once had to compete with current commitments and other projects. An agentic workflow made it possible to separate the recovery into focused lines: component analysis, documentation, mismatch detection, test preparation, and hypothesis verification.
AI did not accept financial risk for me. It made returning to the project manageable.
That gave the project a second life. I did not return to an archive that was merely pleasant to remember, but to a system I could work with again: question it, find weak points, and continue its development carefully. For me, this is one of the most personal results of AI-First.
This was where I first felt that agents could do more than help create something new. They could also bring accumulated intellectual capital back into active work.
Two levels of evidence
A public fact and personal testimony must not be merged.
The public site confirms that Angry Robot and its products exist. Three strategies run on real accounts and, by my account, make money. I do not publish amounts, percentages, or equity curves because this package contains no anonymized export that would let a reader verify those figures.
The precise statement is therefore:
- the platform exists and can be checked publicly;
- the operation and positive result of three real-account strategies are author testimony;
- no future return is promised.
This is not investment advice or an invitation to give money to the system. Any trading decision requires its own risk assessment.
The practical result
For me, the main result is not limited to profit. The project became manageable again. Decisions can be recovered, hypotheses separated from real-account actions, and changes checked before they affect money.
In Cubrim, a published metric settles the argument. In Angry Robot, part of the result remains private and within my responsibility. The two cases require different language.
AI-First is useful both for creating something new and for returning to a complex existing system. The higher the cost of error, the more important the boundary between agent analysis and human decision becomes.
Source and verification date
- Angry Robot public site, checked on July 21, 2026.
The link confirms the platform's existence, not profitability. The public surface must be checked again before publication.