Angry Robot is my system for researching market data and algorithmic strategies. With agents in Arcanada, I could return to the project and make sense of everything that had accumulated in it. They help recover context and prepare tests. Decisions involving real money remain mine.
I founded Angry Robot several years ago. When I returned to revisit an old trading hypothesis, I found myself a long way from the first useful change. I first had to remember how the project worked and why I had made earlier decisions.
Over the years, each hypothesis became tied to data, tests, and code that could place an order on a live account. Returning to the project began to cost more time than I could give it. Whenever I recovered one connection, another one slipped out of view.
I chose one part and asked the agents to bring back everything around it. They found earlier decisions, matched the code to its data, and prepared the tests. I checked those connections and could finally see where to resume.
Now I do not try to recover all of Angry Robot at once. I take one hypothesis, rebuild the context it needs, test the change, and continue from there. Years of accumulated work are useful to me again.
I currently run three strategies on live accounts, and they make money.
Agents can explore a new idea, prepare the code, and run the tests. I stop the process before it reaches a live account. Then I read the results, judge the risk, and decide whether the change should touch real money.
Without my decision, the change stays in testing. That boundary lets me continue developing Angry Robot.