One of the new feelings I have with “AIDEs” is the fear of them stopping :) Sure they are faster and smarter but they also stop for a multitude of reasons that a real person wouldn’t (at least real people would start talking to other people and maybe tap you on the shoulder) but agents sometimes just stop. Period.
We used to have “range anxiety” with EVs, now I have “stall anxiety” (if I setup agents to finish something and some decided to ask me a question half way into the work and/or stop because it forgot I told it to never yield until it gets to a particular end state)
Sometimes it’s context window limitations for sure so there’s this odd question I have on “how close am I in this conversation to a compaction or truncation?”, or “do I have enough “space” for an agent or orchestrator to finish what set it out to do?”
@sampullara Yeah, one thing I noticed after spending more time in Asia (after 2 decades in the US) is that things actually work… a day after a category 4 hurricane, very minimal downtime =p
Traditional WM is really going to be challenged given GPT 5 Pro + some scripting against @monarch_money can yield really good analysis and recommendations.
The context window is enough for it to carry out monthly/quarterly analysis and recommendations.