
Replying to @MortemLabs
@MortemLabs we ship at speed of light 🚀
The best trading agents won’t just have better prompts.
They’ll have better memory of their own failures.
Not memory as chat history.
Memory as evidence:
what failed,
why it failed,
and what changed after.
1
5
110
Mortem Labs retweeted
To stop your bot from lying about where your money went.
As always, @superteamin never forgets to support us. Huge s/o to @paarugsethi for the WIN #14 🫡
Most bot failures are not dramatic.
They are boring
❯ stale context
❯ bad retry logic
❯ missing checks
❯ silent tool errors
❯ overconfident execution
Boring failures are expensive.
Trading agents need more than logs
They need evidence:
‣ what the agent saw
‣ what tools returned
‣ what changed before execution
‣ what landed onchain
‣ whether the mistake was avoidable
Mortem is studying how autonomous trading systems fail before they burn more capital
If an agent uses a quote, waits, reasons, signs, and executes, the quote may no longer describe the market it traded in.
6
85
A bad trade often has three timelines:
‣ model reasoning
‣ tool outputs
‣ on chain execution
Most teams only inspect one.
1
8
83
The interesting question after a bad trade is not “what was the prompt?”
It’s “what did the agent believe was true when it acted?”
5
69

















