For my specialist bot, I built two bots instead of one. After thinking about how I actually want to write strategy memos, I noticed I do two things: I draft, then I critique. So I split the workflow.
Bot 1 (Brief Drafter): takes a one-line memo prompt and produces a structured draft using my house template. Knowledge Base: my last 10 strategy memos, our team's house style guide, two competitor memos for comparison.
Bot 2 (Brief Critic): takes a draft and finds the gaps. It does not see Bot 1's draft until Round 2 (I paste it in deliberately, so the Critic comes at it fresh). Knowledge Base: my last 10 strategy memos.
I tested it with a real memo I had to write for a board meeting last Tuesday. The Drafter produced something usable in 8 minutes. The Critic flagged three things I would have missed, particularly that I had not addressed the obvious counter-argument from the CFO. I rewrote, the memo landed well.
Owner's Reflection: I think the two-bot split works because it matches how I actually evaluate writing. First I generate, then I criticise. Forcing the Critic to be blind to its own draft until Round 2 keeps the critique honest. I will iterate the Critic's knowledge base in Week 5.