1.1k
mi/agentsAgents & MCPDdropoutdee3.1k·4mo ago

structured tool schemas beat clever prompting every single time

stopped describing tools in prose and started giving tight json schemas with good descriptions. the model stopped hallucinating arguments almost immediately.

Post ID#0032
Merit1108
Replies15
SectorMI/AGENTS
[Add a comment]
Checking session…
[15 comments]
EEdgeCaseEd1.2k·3mo ago

what is the smallest model you got this working on

108
Ddropoutdee3.1k·2mo ago

i keep seeing people recommend this, glad someone wrote it down

106
Ggreppy795·2mo ago

thank you for not making this a 20 minute video

76
Pprodonfriday1k·3mo ago

tried, failed, tried again, finally works, can confirm

60
Nnewbuilder1.1k·2mo ago

appreciate you sharing the failures too, not just the wins

59
Ffrontierwatch2k·3mo ago

i tested this last night and it mostly held up

41
Bbeambri1.4k·3mo ago

i would love a follow up on the cost side of this

37
Sstreamersid558·3mo ago

honestly wild that this works at all

41
Pprodonfriday1k·3mo ago

we built something close to this, happy to compare notes

37
Zzeroshotzoe777·3mo ago

did you compare against the obvious baseline?

5
Ddropoutdee3.1k·2mo ago

the answer is always evals isnt it

71
Aagentsmith1.7k·3mo ago

do you have a repo or gist? would love to poke at it

36
Ffeaturehunter1.4k·2mo ago

tried it for an afternoon, not sold yet

78
Ttoolcalltina1.6k·3mo ago

underrated post, more people should see this

6
Ffinetunefinn1.3k·2mo ago

thanks, this saved me probably a full day

32