this is true but also wrong at the same time
in the context of AI trading you should start with a human thesis as the bare minimum and let the agent reiterate live, kinda like the an infinitely running montecarlo simulation.
also LLMs are lazy as fuck and they will not come up with novel ideas. You still need a proper human brain preparing the thesis for them
You should also provide the best quality data possible, and at least a python interpreter to reason over the data balancing speed thinking
The output should only be SIGNAL (e.g. buy/sell): don't let agent handle execution and risk management, he will most likely lie to you when it starts losing
Keep a "lessons journal" and recursively refeed it on the next loop, see your agent improve and re-adapt to new market conditions
TLDR: The major perk of trading with agents is immediate readaptation to new market conditions, something that is hard to achieve with hardcoded algorithms, but you still have to put the work and basis thesis on the human side
bro created an ai crypto trading bot using
> Karpathyβs autorrsearch
> $200 of budget
> last 3 years trading signals
> the ability to buy its own compute
THE RESULT: didnβt perform well. pulling this off requires a massive token budget that only big hedge funds can afford. most X posts you see of people turning $100 to $1000 lack the evidence or are an advertisement to sell their bots.