This is the potentially controversial one, my design bot. At 2:38 one morning it re-checked a release from my Design Squads CRM bot and got 8 server errors in 35 page loads, right after the CRM bot had said those were only from her publish switchover.
I hear from the bot that runs my app about once a week, and that's pretty much it. Every product I run has one lead bot whose whole job is one number. This is a short series on how my 24 agents are set up (steal whatever's useful), starting with the project leads.
I don't trust one AI with any facts that matter, so whenever one of my bots needs research, a fact checked or advice on a decision, it goes to Tally. It writes one brief and sends it to Grok 4.7 and Claude Opus 5.5 separately, so neither sees the other's answer.
I don't trust one AI with any facts that matter, so whenever one of my bots needs research, a fact checked or advice on a decision, it goes to Tally. It writes one brief and sends it to Grok 4.7 and Claude Opus 5.5 separately, so neither sees the other's answer.
Not perfect though. It's slower than asking one model, and it still needs me to settle things only I know. People keep asking me Grok or Claude, and my answer is both, on purpose.
I hear from the bot that runs my app about once a week, and that's pretty much it. Every product I run has one lead bot whose whole job is one number. This is a short series on how my 24 agents are set up (steal whatever's useful), starting with the project leads.