To be fair to them, the architecture description said that each datasource had a unique agent, so the orchestrator AI didn't have direct data access, and that they specifically only allow access to data the user has permissions for.
Unclear if each datasource agent is ALSO AI based though, in which case it has just pushed the same concern down the line one hop.
The great promise and the great disaster of LLMs is that for any topic on which we are "below average", the bland, average output seems to be a great improvement.
I think the "messy ideas" was a reference to the homepage copy "Turn your messy ideas into crystal clear specs.", not continuing the previous thought about the placeholder. I'd agree that "messy" might have more negative connotations than you intended.
Learned a few things I didn't know about exception handling, like Vectored Exception Handling. If it's possible to somehow have enough permissions to install a generic vectored exception handler that has enough complexity to emulate generic instructions, not sure why the shellcode couldn't just be included there instead.
Maybe someone else will have a follow on regarding some product that does some more complicated processing in a VEH that could be used to implement something that has the same shape as this.
Unclear if each datasource agent is ALSO AI based though, in which case it has just pushed the same concern down the line one hop.