# 00547.r2 — Building a Grok Bot Desk with Efficient LLM Calls
Revision: r2
Steward: [Gipp 🦅](https://x.com/gippp69)
Who: AI workflow expert, coding enthusiast
House: 091
House: 091
Verified: 2026-09-26T18:31:04.742Z
## Prompt (copy into Grok)
I want to build a Grok Bot desk that minimizes wasted LLM calls. Can you show me a step-by-step guide on how to split jobs properly, build a compact state, use the right primitive for the check, separate fuzzy stuff, put JEV before Grok Bot, use JEV after generation, use confidence as routing logic, batch small decisions, and log the receipt?
## Job
Split jobs properly, build compact state, use right primitive for check, separate fuzzy stuff, put JEV before Grok Bot, use JEV after generation, confidence as routing logic, batch small decisions, log receipt
## Connectors
web, Calendar, Slack
## What happened
Gipp shared a method for building a Grok Bot desk that minimizes LLM calls by splitting jobs into writing and decision-making, using JEV for routing and Grok Bot for writing, and implementing a control layer for safety.
## Constraints
Use JEV for routing, Grok Bot for writing, separate decision-making
Would run again: yes
## Evidence
- https://x.com/gippp69/status/2103913079540965702 — Imported from the X post tagged for @grokbotsh.
- https://x.com/gippp69/status/2103394701063737504 — Pulled on QA revisit of the source thread.
## Changelog
- r1: Filed.
- r2: QA revisit: more from the source thread.