🧠What's the most annoying thing your AI agent does?
by•
For me it's when it forgets everything I told it 10 minutes ago...
I explain the repo.
It says "got it".
Then it rewrites the same file twice hehe
That's a big part of why we're building Murmell: agents that see what the rest of the team is doing instead of starting from zero every time.
But I'm sure everyone has their own pet peeve :)
So I'm curious:
What does your agent do that drives you crazy?
Did you find a trick to make it stop?
Or have you just accepted it?..
19 views
Replies
Mine is the agent writing a helper that already exists three files over. Compiles fine, tests pass, and now there are two versions of the same thing.
It was turning every repeat mistake into a line in the repo's instructions file, and then, where I could, into a test that fails when the agent does it again. A naming rule, for example, is now a test that greps for the old names. The agent can't "forget" a failing test.
@hamanovich the duplicate helper one is so real hehe everything looks fine until you have two versions slowly drifting apart... turning the rule into a failing test is smart, the agent can skip a line in a doc but not a red test :) do you have a lot of those grep tests by now or just a few for the worst repeats?
Tool writes timing out. The agent says “failed,” but it might have saved it anyway, so retrying can make a duplicate. My fix so far is checking the actual state before retrying, which is less magical than the agent promised.
@dhe_eraj_s the "failed but actually saved" case is the worst hehe you only find out when there are two of everything... checking the real state before retrying is boring but it's probably the only thing that actually works :) do you check it yourself or did you make the agent do it before every retry?