Tom Osman

Independent proposal // SpaceXAI Grok Bot

Grok Bot Academy: from first task to agent operator.

A beginner-to-advanced learning system for people who want to give Bots real jobs, trust the work they return, and build dependable agent operations.

The teaching model

Every lesson ends in evidence

A product tour can explain where the buttons are. An academy should teach someone to finish work. Each module starts with a job, shows the Bot doing it, exposes the rough edges, and gives the learner a brief they can adapt.

Progress depends on working outputs: a sourced report, a tested skill, a routine that handles missing data, or a handoff another Bot can use. The learner keeps the build, the run evidence, and the revised instructions.

Curriculum // Five levels

Beginner to advanced, one finished job at a time

01 // Foundations

Give your first Bot a real job

Understand the product model, set up a Bot safely, and write a task that produces a useful result.

Modules

  • Bots, tasks, skills, routines, and the persistent computer
  • Choosing one repeatable outcome instead of a vague role
  • Writing the first brief: outcome, sources, rules, boundary, and proof
  • Reviewing a run and correcting the instructions

Capstone

Build a research briefing Bot that returns a sourced report and stops before taking any external action.

02 // Bot Operator

Make one job dependable

Connect the right context, set approval boundaries, and improve a job from evidence rather than guesswork.

Modules

  • Files, browser sessions, apps, connectors, and source rules
  • Authentication handoffs and least-privilege access
  • Approval points for sending, publishing, buying, deleting, and production changes
  • Failure review, source tracing, and useful partial results

Capstone

Build a weekly product-update brief with traceable sources, a missing-data response, and a visible review point.

03 // Workflow Builder

Turn a good run into a reusable system

Convert a proven method into a skill, test it with new inputs, and schedule it only when it is ready.

Modules

  • When instructions are reliable enough to become a skill
  • Teaching a workflow by demonstration, then tightening the written method
  • Testing against second inputs, empty sources, and stale data
  • Routine ownership, schedules, time zones, retries, and failure reports

Capstone

Ship a scheduled workflow that can recover from partial failure and asks for approval at the right moment.

04 // Agent Operations

Run a team of Bots

Design clear ownership and handoffs across several Bots while treating their shared computer as one workspace.

Modules

  • Bot rosters, narrow roles, and unambiguous ownership
  • Passing context and deliverables between Bots
  • Shared files and logged-in sessions as an operating risk
  • Logs, evidence, incident review, and the human escalation path

Capstone

Design a three-Bot operating system with handoff rules, an evidence trail, and a failed-run drill.

05 // Teams & Enterprise

Move from individual wins to team adoption

Plan a controlled rollout with useful policy, internal education, support, and a direct line back to product.

Modules

  • Selecting a pilot team and jobs with visible value
  • SSO, permissions, approved tools, privacy, and usage controls
  • Internal champions, office hours, templates, and support routes
  • Adoption evidence, product feedback, and the next rollout decision

Capstone

Present a 30-day team pilot with a launch kit, risk register, support plan, and adoption scorecard.

The community is part of the curriculum

Teach in public. Learn from the failed runs.

The community should be organised around work people are trying to finish. Good questions become lessons. Reliable builds become templates. Repeated friction becomes product feedback.

Weekly

Build a complete job live

Start from a blank Bot, work through the failures, and finish with the brief, output, and evidence available to copy.

Always on

Use-case labs and build logs

Organize discussion around jobs people are trying to finish. Learners share briefs, failed runs, revisions, and working versions.

Fortnightly

Office hours and workflow clinics

Debug real builds in public, explain the fix, and turn repeated questions into lessons or product feedback.

Monthly

Demo day and field report

Show the best working systems, publish reusable patterns, and send a concise report of friction and requests to the product team.

Measure the work

Completion beats consumption

Video views and community size help with reach. They do not prove that somebody can operate a Bot.

  1. 01Time from signup to the first finished, reviewable job
  2. 02Percentage of learners who complete a capstone
  3. 03Percentage of workflows that succeed with a second input
  4. 04Whether a Bot stops at its stated approval boundary
  5. 05Questions resolved by the community and turned into durable material
  6. 06Recurring product issues reproduced and returned with evidence

First 90 days

Build the academy alongside the product

Days 01–30

Fieldwork and foundations

Run real jobs, interview early users, ship the first learning path and brief template, then open the community around build logs and office hours.

Days 31–60

Operators and workflow builders

Publish the intermediate paths, start weekly live builds, create the capstone review system, and establish the product-feedback report.

Days 61–90

Agent operations and team pilots

Launch the advanced paths, test the enterprise material with a pilot group, publish the pattern library, and review adoption evidence with product.

Independent proposal // Tom Osman

I would like to build this with the Grok Bot team.

This academy concept is my own proposal. It is not an official SpaceXAI course and does not imply an affiliation with SpaceXAI.