Loading…
19
A guy at the Austin meetup told me my AI agent was 'just a chatbot with extra steps' and it's been living in my head for 3 weeks
I showed him my agent that books my own calendar and drafts replies to client emails at the coffee shop on South Congress, and he just laughed and called it a fancy wrapper. That stung mostly because he wasn't totally wrong, my setup was really just chaining prompts with no memory between runs, so it broke the second a reply needed context from two weeks back. For the folks actually running agents in production, how are you handling long term memory without it turning into a giant messy pile of past chats?
1 comments
Log in to join the discussion
Log In1 Comment
green.noah6d ago
Ha, okay but the "just chaining prompts" part is actually the real problem, not the memory itself. You can have perfect long term memory and still fail if every run starts fresh with no idea what the last run decided. The fix isn't stuffing all your old chats into context, it's saving the outcome of each run (what got booked, what got sent, what the client wants) as a small set of facts you pull back in later, kind of like notes to yourself instead of a transcript. Whole chat logs are a trap, they get huge and mostly useless fast.
8