Grok Bot Official Use Guides Published by SpaceXAI

SpaceXAI released five official Grok Bot use guides on August 28, 2026, covering meta mode, engineering, design, GTM, and product directions. The guides detail six core components: job descriptions written like hiring JDs, connections to real accounts (Jira, Figma, Salesforce, GitHub, Meta Ads), a 24/7 cloud computer per bot, scheduled routines, skills via recorded UI automation, and direct bot-to-bot handoffs. The meta mode guide describes a PM bot coordinating multi-team projects with channels, Notion boards, and human approval for new bots. The engineering guide shows six bots running a mobile game studio, claiming CPI reduction from $15 to $1 and 4x D7 retention improvement through autonomous read-hypothesize-verify loops. The design guide uses four bots for exploration, reducing idea-to-prototype cost to near zero and using real production assets with Figma MCP for precision. The GTM guide provides a full sales bot roster with style profiles from email scans, anti-spam guardrails, honesty constraints, and real-time slide generation during calls. The product guide introduces the 'attention list' as a new work primitive, generated by observing actual behavior, and argues for multiple specialized bots over one universal agent for citability, parallelism, and scope memory. All guides emphasize human final review for external emails, spending, and deletions.
Key insights
Grok Bot's six core components define a new agent architecture
The guides establish six building blocks: job descriptions as system prompts, connections to real SaaS accounts, a persistent cloud computer per bot, time-based routines, skills recorded via UI automation (used as a workaround when APIs are unavailable), and direct bot-to-bot handoffs. This architecture enables bots to operate continuously and collaboratively without human triggering.
Meta mode replicates human project management for bot teams
When multiple bots cause noise and sync issues, the solution is to copy human PM structures: a project equals a channel plus a Notion board plus a roster, with a PM bot handling creation, staffing, and blocked-task escalation. The guide's author notes the irony that the system increasingly resembles one originally designed for humans.
Six bots autonomously run a mobile game studio with 15x CPI reduction
The engineering guide uses six specialized bots (orchestration, analytics, creative, client engineering, ops, bug fix) to operate the game Rank'em. The bots handle 85% of the work (user acquisition, creative, QA, ops) while the game itself is only 15%. Claimed results: CPI dropped from $15 to $1, D7 retention improved 4x, achieved through autonomous data-to-action loops. One bot proactively added a user request as a daily routine.
Design exploration cost drops to near zero with bot teams
Using four bots, a designer built three ambient invocation prototypes that were never shipped but provided real judgment data. The approach uses real production assets (animation spec files, Figma MCP for exact coordinates) rather than approximations, and combines deterministic controls with natural language tuning. The conclusion: bots tighten the iteration loop without replacing human judgment.
GTM guide provides a full sales bot roster with practical anti-spam and honesty safeguards
The sales bot team includes Chief of Staff, Prospecting, per-customer experts, a product expert connected to codebase, forecasting, slides, and coaching bots. Key techniques: scanning 15-30 sent emails to build a writing style profile, checking 90-day contact history to avoid re-pitching, an explicit prompt rule to admit 'no verifiable updates found' rather than fabricate, and real-time slide generation from call transcripts shown at wrap-up.
Attention list emerges as a new work primitive for agent coordination
Instead of static priority lists, a bot continuously observes a PM's actual activity across Slack, email, and meetings to generate an 'attention list' of what they really focus on. This serves as both an information filter for agents and a mirror to compare claimed vs. actual priorities. The guide argues for multiple specialized bots over one universal agent for citability, parallelism, and non-cross-contaminating scope memory.
Human final review is maintained for high-stakes actions
Across all guides, external emails, spending money, and delete operations remain subject to human approval. The product guide explicitly states that people want to know the other party has thought seriously before making a request, setting a clear boundary for agent autonomy.