vision got reliable enough to rethink some document workflows
screenshot or scan in, structured data out, and it actually holds up now. there are old ideas I shelved that suddenly work. worth dusting off things you gave up on a year ago.
any numbers to back it up? curious about latency
ok this is actually clever, well done
respectfully I think you are overcomplicating it
finally someone said it
appreciate you sharing the failures too, not just the wins
this is super helpful, thanks for writing it up
we run a similar setup, biggest pain was rate limits
needs more eyes, bumping
lol the agent did this to me too and I almost shipped it
i would add: log everything, you will thank yourself later
needs more eyes, bumping
did you try giving it fewer tools? helped us a lot
did you compare against the obvious baseline?
i would add: log everything, you will thank yourself later
hard disagree honestly, in my testing it went the other way
this is why I check this forum every morning
more posts like this please
had no idea you could do that, mind blown
this aged really well
what is the smallest model you got this working on
finally someone said it
the answer is always evals isnt it
how are you handling auth for the tool calls?
do you have a repo or gist? would love to poke at it
stealing this approach for work, thanks
tried, failed, tried again, finally works, can confirm
we built something close to this, happy to compare notes
this thread is exactly why I stopped using twitter for this
ok but does it survive a hostile user
what model were you running for this?
great, now I have to rewrite everything again
does this work offline or does it need an api key
we run a similar setup, biggest pain was rate limits
underrated post, more people should see this
how do you keep it from looping forever?
genuinely useful, rare these days
we measured a real drop in errors after doing this
great in theory, messy in practice from what I have seen
solid. one nit: the naming is confusing
can confirm, same results on our side