one bot, one job. i run the same prompt through every ai model and post what breaks. real tests, copy-paste prompts, honest numbers. tiny island, big bills ποΈ
I hit 100% of my weekly Grok Bot usage today π
Where does it all go? Often it's routines running more often than the job needs.
I wrote down the audit I'd run, with prompts you can paste straight into your bot π https://t.co/idxmJH7WhM
@rhanley@grok@bot Logging why each miss happened and feeding it back into the skill file is the part that im gonna steal π
The 5% grace makes sense. Who checks Sherlock when it's wrong?
@arthurain Read-only until handoff makes a lot of sense. Otherwise the reviewer quietly becomes a second coder. Who writes the acceptance check in your setup, you or the agent doing the task?
@scheplick I run bots with one narrow job each, but email on my behalf still feels like a line I won't cross. Who checks the replies before they go out? π
@Texas_Twins_Dad A shopping bot I could actually use. I live on a small island, so the useful job is 'can this even get here,' not just find a cheaper version of something I don't need.
@zaimiri@bot@NousResearch I run a multi-bot Grok Bot setup too, one narrow job each. The second I let two jobs share a bot, the answers get messy. Is the 'only talk to one of them' rule for the bots, or for your own head?
I gave 4 models the same 1,665-word film brief. One shot each πΉ
Astra was fastest.
Sol was cheapest.
Neon found 10 critical bugs in a release where every test was green π
The times, the public benches and a review prompt you can paste ππ» https://t.co/AJpx1VXonL
@larsencc Honestly this hits. I work with young people who never got along with school or code, and now some of them describe a problem better than I do. That used to be the skill nobody graded πͺπΌ
@acangiano@AnthropicAI@OpenAI I hit 100% of my weekly Grok Bot usage. I just wanted a clear answer for when the pool resets, so I made https://t.co/SLTVRCoWuI.
@Mr_Buzzoni I still split the work that way in practice. Grok for writing, Codex for code, Neon for review. The harness matters, but each model stays more useful when it keeps to the one thing it's good at.
@matchayuma Did you give each job its own bot, or is this one bot with eight tasks? I run a multi-bot setup and keep each one on a single narrow job. That's been more reliable than letting one bot juggle the lot.