and then you want to track orders states, and then you want to track exit strategies - trailing stops that are sometimes internal, sometimes sent to the order book - profit targets, and then you want to track settlement statuses as balances change on margin, and how you get filled
all while dealing with different and complex broker APIs and routing to different exchanges that have their own rules and limitations
on the other hand, agents just do it and handle edge cases themselves
Have you actually put together trading strategies by having the agent drive? I've never trusted it that far and I use agents a lot at my job right now. The way I usually do it is, I break out pen and paper to do an analysis of what I want mathematically. I then have the agent build out some Python that lets me backtest and analyze my work. I read through the code (which is usually fairly compact since numpy/scipy and various finance libraries do most of the heavy lifting for me), make any changes as needed, then run my analysis. Then I run it in a production setting if I like it. But the actual strategy is something I come up with on pen-and-paper.
I have, and agents come up with the strategy and execution based on my contribution of what sectors to look at and alternative data sources I tell them to look at a certain way. My time horizons are quarters, as well as signal conversion into a variety of single and multi leg options trades
all while dealing with different and complex broker APIs and routing to different exchanges that have their own rules and limitations
on the other hand, agents just do it and handle edge cases themselves