The plan
Two routes, one front door
Everything on this page sorts into two routes. Route 1 gets the tools into a class: a verified instructor applies, and students get capped access for the term. Route 2 gets the course into the instructor's hands: a complete, ready-to-teach template for each kind of class, written so that every assignment runs on Cursor or Grok Bot. Instructors take the path of least resistance, and a complete template is that path.
Cursor builds, Grok Bot does everything else
Cursor serves computing, engineering and data. Grok and Grok Bot serve business, agriculture, health, education, media, the sciences and the humanities. Each template names its tool.
Written by the faculty who teach it
Templates are written by faculty fellows in each discipline and recognized with certification, fellowships, speaking invitations and a conference, so the material stays current and credible.
Build with Grok
A public library where anyone shares workflows built with Grok and Grok Bot. Course templates and student capstones feed it; it feeds the next term's assignments.
Why educators
Students follow the tools their courses use
Educators choose those tools, and most are choosing without training, policy or material. A platform that solves their problem becomes the default in their classroom, term after term.
Educators and students already use AI, ahead of the support they get
Share of each surveyed group, percent. Each bar is a separate survey.
Every major platform teaches. Assessments are the thinnest column
Depth of each company's education offer by component. My reading of public program pages, October 2026.
The dashed row is the target for this program, not a current offer.
SpaceXSI starts from strength. Grok Bot reached about 418,000 weekly users five weeks after its August launch (PYMNTS), and Cursor is a tool professional developers already pay for. What is missing is the path from those products into a syllabus. An education program is how the tools reach the next group of users before their habits are set, and it is cheap next to what the company already spends on everything else.
Instructor's view
Every platform gets one half right
An instructor needs two things at once: a tool every student can use without paying, and material that is ready to teach from. Among the platforms I teach with, only one gives me both, and it is a cloud platform, not an AI assistant.
Only Google Cloud pairs open access with material ready to teach
Positions are my judgment as an instructor who teaches with these platforms. They are not measurements.
What I will hold this program to
Three rules that close the gap the others leave open.
- Access and material are designed together. No lesson needs more than the classroom credit provides. If a student cannot run it without paying, it does not ship.
- A curriculum, not a collection. Ordered units, each with slides, a lab, an assignment and an assessment, for every level from school courses to professional training.
- A request an instructor can finish between classes. About ten minutes to apply, and credit in students' hands within days.
The result for SpaceXSI: students meet Grok, Grok Bot and Cursor first, in a course, with a teacher, before any other AI platform. First choices tend to hold. In one 2025 survey, 91% of AI users said they go to their favourite tool first for nearly every job (Menlo Ventures).
| Platform | Free for every student? | Ready-to-teach material? | In a classroom | Verdict |
|---|---|---|---|---|
| Claude | Not for the hands-on parts. Anthropic's courses are free to take, but Claude Code needs a paid plan (Claude support), and I have found no free or discounted plan for an individual student. | Yes. This is the AI learning material teachers ask for most. | No educator wants to require students to pay for one AI platform, and many are not permitted to. The material goes unused. | Strong material, closed door |
| ChatGPT | Yes. The free tier is generous. | Not as a course. The learning material is scattered, with nothing complete that I can adopt for a classroom. | Students have the tool, and each instructor builds the course alone. | Open door, no course |
| Google Cloud | Yes. Faculty request credit on a short form, and each student redeems a coupon (Google). | Yes. Course and classroom material is ready to adopt (Google Cloud for faculty). | I can adopt it quickly, and my students use it every term. | The balance to copy |
| Gemini | Yes, through its free tier and student offers. | Not yet. Curriculum-style material for a course is missing. | The same as ChatGPT: access without a course. | Open door, no course |
| Grok, Grok Bot and Cursor, with this program | Yes. Capped classroom credit through a verified educator, and every lesson runs on that credit. | Yes. Sequenced units for each level, written with faculty in each discipline. | Students learn Grok, Grok Bot and Cursor first. If Grok Bot routes each task to the best model, one classroom credit reaches them all. | The target |
Classroom credit
Credit that reaches real classrooms and stays out of the wrong hands
Free student offers attract fraud on every platform. The answer is not to stop offering credit. It is to change who asks for it.
Verifies once
Institution email plus a faculty directory page, or a teacher-verification service. A person reviews unclear cases.
Applies for a class
Course, term, enrolment and a syllabus link. About ten minutes. Small classes are approved automatically.
Issues a class code
The educator receives a coupon pool sized to the class, plus the starter kit for the chosen track.
Redeems a capped coupon
Class code plus school email. One coupon per student per course, valid for the term.
Sees usage and renews
A simple view of who redeemed and what is left. Renewal next term takes one click.
Route 1, in its simplest form
Strip the program down and it is three things: a form on a website, a person who checks it, and a class-sized bundle of the products. Everything else on this page can wait for year one. The bundle deliberately leaves X Premium out: a class needs the tools, not the social account.
The application
| Step | What happens | Who |
|---|---|---|
| 1 · Apply | Institution email, faculty directory link, course name, term, enrolment, syllabus link. About ten minutes. | Instructor |
| 2 · Vet | Automatic checks on the email domain and directory page; a person reviews anything unclear and every class over 150. Target: two working days. | Program |
| 3 · Issue | A class code and a bundle sized to the roster. Expires at term end. | Program |
| 4 · Redeem | Class code plus school email. One per student per course. | Students |
| 5 · Renew | One click for a returning instructor in good standing. Usage from the previous term is the evidence. | Instructor |
The class bundle
Two sizes, mapped onto seat levels that already exist. Prices are SpaceXSI's to set; the point is the shape.
| Standard class | Large or lab-heavy class | |
|---|---|---|
| Fits | Up to about 40 students, one project | Up to about 150 students, weekly labs |
| Cursor | A capped seat per student for the term | Same, higher cap |
| Grok Bot | A seat per student at the Premium level ($30 today) | A seat per student at the Super level ($100 today), with a Team Bot for the course |
| API | A small shared pool for course projects, with a hard cap | A larger pool, still capped |
| Not included | X Premium or any social account. Students bring nothing but a school email. | |
What Cursor already does for campuses
This proposal does not start from zero. Since June 2026 Cursor has run a campus program (forum post, cursor.com/students). The recommendation below is to scale the educator credits form it already has, not to replace it. The channel for any of this is the campus team itself ([email protected] and the forum), not a product request.
| Cursor today | What it does | Gap from an instructor's seat | This proposal |
|---|---|---|---|
| Free student tier and event promotions | A free tier for students, plus promotions around hackathons and workshops. | A free tier runs out mid-assignment. Promotions are tied to events, not to a 15-week course. | Keep. Add a course-sized coupon on top. |
| Credits request form for educators and researchers | Master's and PhD students, researchers and educators can request credits through a form. | From the outside it looks like a request-by-request review with no course roster, no per-class cap and no renewal. I may be wrong about the internals; I have not yet put a course through it. | Scale this form into the classroom application: one educator, one course, one roster, one capped coupon per student. |
| Campus Leads | About 40 invite-only student ambassadors run events on their campuses. | Student-led and event-led. Instructors are not part of the loop. | Keep it student-led. A question for Cursor: would a faculty sponsor on the same campus help a Lead, or get in the way? |
| MLH partnership | Workshops, credits and swag through Major League Hacking. | Reaches hackathons, not courses. | Keep. Use MLH for Build with Grok events. |
| Office access in San Francisco and New York | Students in the program can use Cursor offices. | Geographic. | Not something to change. If an annual educator meeting ever happened, a Cursor office would be a natural venue to ask about. |
Four ways to give students access
| Model | How it works | Who uses it | Fraud exposure | Reach into real classes | Verdict |
|---|---|---|---|---|---|
| Open student sign-up | Each student proves enrolment alone and receives a full plan. | Cursor's earlier student offer; GitHub's student plan | High. Cursor's Campus Community Lead wrote that the legacy student discount was closed because of fraud (Cursor forum, June 2026). GitHub capped its plan to keep it sustainable (GitHub). | Wide but shallow. No course, no teacher, no material. | Not as the front door |
| Institution membership | The institution joins first. A central contact then nominates each educator. | AWS Academy (AWS) | Low. | Narrow. Single courses, adjuncts and small departments wait on paperwork. | Too slow as the front door |
| Educator-verified classroom coupons | An educator applies for a course. Each student confirms a school email and redeems a coupon. | Google Cloud education credits (Google) | Low to moderate. A named educator is accountable and the cap limits any loss. | Deep. Credit arrives with a syllabus and a teacher. | Recommended core |
| Paid campus licence | The institution buys seats for everyone. | ChatGPT Edu at Cal State, about $24 per user per year (EdSource) | Low. | Broad, but access alone does not create use. At Cal State 0.7% of students and 16% of faculty had completed the voluntary training by April 2026 (CalMatters). | Add later, on demand |
Strict about
- Who the educator is, checked once and by a person where needed
- One coupon per student per course
- A cap on each coupon, and expiry at the end of term
- No transfer or resale of credit
- Usage visible to the educator and to the program
Flexible about
- Which subjects qualify: any accredited course, not only computing
- Who counts as an educator: adjuncts, community college instructors, and teaching assistants with a faculty sponsor
- Speed: small classes approved automatically, large ones within days
- Top-ups when a class runs out mid-term
- School email domains outside .edu, for K-12 and international partners
What I have seen as an instructor. I have taught with Google Cloud, AWS, Azure and other platforms for more than fifteen years, and I have watched fraud hit free student offers on several of them.
Fraud is a design problem. It follows offers that anyone with a student email can claim alone. It is far less likely when an educator, with a reputation and a career to protect, asks on behalf of a class.
Strictness has a cost too. Where access is very strict, I could rarely get credit into a legitimate course. Google Cloud's lighter process let my students use credit every term, and they went on to learn the wider family of tools around it: Gemini, AI Studio, NotebookLM, Flow and Stitch. I want to find the same balance for Grok and Cursor.
Two tracks
Cursor for people who build software. Grok and Grok Bot for everyone else
Most students will never write code for a living, but nearly all of them will work in offices where research and routine process can be handed to AI. Two tracks let each product meet the right students.
| Cursor track | Grok track | |
|---|---|---|
| Who it serves | Computer science, software engineering, data science, information systems and engineering | Business, agriculture, health, education, media and design, humanities, social science, law and the natural sciences |
| What students learn | To plan, delegate, review and ship software with coding agents | To research with live sources in Grok, and to design, supervise and audit office workflows run by Grok Bot |
| Classroom credit | A capped Cursor plan for the term | A Grok plan with a learn mode, and Grok Bot seats that run against practice systems (see Grok Bot in the classroom) |
| Material | A semester course kit for agentic coding, an exam bank and a capstone guide | Workflow templates by discipline, and business process labs with synthetic data |
| Credential | Certified Educator, and a proctored student practical | Certified Educator, and an agent supervisor practical for students and working learners |
| Students under 18 | Through school-managed accounts, in teacher-led courses | Teacher-facing first. Student use follows once an education mode and school privacy terms are in place |
Route 2 · Course templates
A complete course is the path of least resistance
When I plan a new course I look for whoever has the most complete package: learning objectives, slides, readings, labs, quizzes, homework, exams and a project, in one place and in a consistent voice. Whoever has that wins the adoption, because it is the least work. No AI company has it today. SpaceXSI could, and every assignment in it would run on Cursor and Grok Bot.
Hours to prepare a new 15-week course
My estimate, from building AI 2010 and DS 2010 and from adopting vendor material for both. The gap between the first and last bar is the whole argument for Route 2.
What "complete" means
Every template ships all ten. A template missing any of them is a resource, not a course.
| Component | What ships | Runs on |
|---|---|---|
| Learning objectives | Measurable, mapped to each week and each assessment | |
| Lecture slides | Fourteen decks with speaker notes, editable | |
| Readings and references | Curated pages, docs and short videos per week, with dates | Grok |
| Labs | Weekly, with starter files and synthetic data | Cursor or Grok Bot |
| Quizzes | Bank of 10 to 15 questions per week, with answers | |
| Homework | One per week, with a rubric and an AI level | Cursor or Grok Bot |
| Exams | Two midterms and a final, secure-lane versions and open-lane versions | |
| Project | Brief, milestones, rubric and a demo-day format | Cursor or Grok Bot |
| Instructor guide | Week-by-week notes, common student errors, the AI policy text | |
| Class bots | Tutor, lab partner and grader's assistant, configured for the course | Grok Bot |
The catalogue: two core courses, then one per college
The two core courses are the ones I would write first and teach myself. The discipline courses share a common spine (what AI is, how to supervise it, how to judge its output) and swap every example, dataset, assignment and project for the discipline's own. The business course talks about margins and forecasts; the agriculture course talks about yields and weather; nothing generic survives.
| Course template | For | Primary tool | What is tailored | Example project | Wave |
|---|---|---|---|---|---|
| Introduction to Applied AI with Grok Bot and Cursor | Any major, first or second year. The university-wide elective. | Grok Bot with a Cursor week | Nothing: this is the spine every other template is built on | Design, configure and defend a bot for a real task | 1 · Spring 2027 |
| Programming and Software Development with AI | Computer science, software engineering, computer engineering, MIS | Cursor | Everything is code: inherit a repo, ship a feature, find the agent's bug, defend the pull request | A tested feature in an unfamiliar codebase, reviewed live | 1 · Spring 2027 |
| Applied AI for Business | Business, finance, marketing, supply chain | Grok Bot | Market research routines, forecast critique, agents on a synthetic CRM, AI in finance and risk | A market-entry plan drafted by an agent and defended to a panel | 2 · Fall 2027 |
| Applied AI for Agriculture | Agronomy, animal science, ag business, ag engineering | Grok Bot | Yield and weather data, field-trial analysis, extension-style reporting, equipment and supply routines | A season plan from synthetic farm data, with the bot's errors found | 2 · Fall 2027 |
| Applied AI for Engineering | Mechanical, civil, electrical, industrial | Cursor and Grok Bot | Simulation scripts, spec and standards lookup, design review by agent, lab data pipelines | A design review the agent runs and the student audits | 2 · Fall 2027 |
| Applied AI for Health and Human Sciences | Nursing, kinesiology, nutrition, family and consumer sciences | Grok Bot | Care plans against guidelines, patient-education drafts, program evaluation, no real patient data | A care plan the bot drafts and the student corrects against the guideline | 3 · Spring 2028 |
| Applied AI for Social Science and Law | Political science, sociology, psychology, pre-law, public policy | Grok Bot | Survey design and critique, policy memos, case summaries checked against sources, bias audits | A policy brief with every claim traced to a source | 3 · Spring 2028 |
| Applied AI for the Humanities and Literature | English, history, languages, philosophy | Grok | Argument from three AI drafts, primary-source checks, translation critique, archival research routines | A fourth draft better than the AI's three, with the reasoning | 3 · Spring 2028 |
| Applied AI for Chemistry and Biology | Chemistry, biology, biochemistry, pre-med | Cursor and Grok Bot | Literature watch routines, lab-notebook agents, data analysis in notebooks, protocol critique | Two planted errors in a dataset, found and explained | 3 · Spring 2028 |
| Applied AI for Education | Teacher preparation, instructional design | Grok Bot | Lesson planning with AI, assessment design on the five levels, the class-bot setup itself | A unit plan with its own tutor bot, taught to peers | 3 · Spring 2028 |
Three ways an instructor uses a template
| Mode | What changes | Time to first class | Who it suits |
|---|---|---|---|
| Adopt whole | Dates, institution name, the class bot's rules | About 12 hours | A new course, an adjunct, a department filling a gap |
| Adapt a module | Swap two to four weeks and the project for the local discipline or dataset | About 30 hours | Most instructors with an existing course |
| Pick assignments | Take the labs, quiz bank and exam versions; keep your own lectures | A few hours | Experienced instructors who want the assessments, not the slides |
Rules that keep the catalogue trusted
- Faculty authorship. Each template is written by a faculty fellow who teaches that course, named on the cover, and paid through the program. SpaceXSI staff review for product accuracy, not pedagogy.
- Open licence. Creative Commons BY, so a department can adapt it without asking and the material outlives any one program.
- Versioned every term. Products change monthly; a template carries a "tested with" line and a changelog, and a dead link is a bug.
- Assessment is never thinner than the slides. Every week has its quiz, every assignment its rubric and AI level, every exam its secure-lane version.
- Taught before published. A template ships after one full semester in a real section, with the instructor's notes on what broke.
What the catalogue does for SpaceXSI
- Every assignment is a product session. A template with 14 labs and a project puts each student in Cursor or Grok Bot 20 to 30 times a term, on work that is graded.
- The tool dependency is the moat. The material is open; the labs run on SpaceXSI's products. A department that adopts the course adopts the tools.
- One template, many campuses. The business course written once at one university is taught at every business school that picks it up. That is the leverage no credit program has.
- It answers the question sales gets asked. "What would we teach with it?" becomes a link.
- It is cheap next to credit. Ten templates by paid faculty fellows cost less than one semester of seats for a single large university.
Build order
Two courses taught by their authors in spring 2027, three more written over the summer and taught in the fall, the rest in 2028. Each wave is taught before it is published, so the catalogue never lists a course no one has run.
Grok Bot in the classroom
The instructor publishes the bot. Students learn by supervising it
Grok Bot is a team of agents that signs into apps and websites and does the work itself. That makes it the first AI product students can learn to manage rather than just prompt. An instructor builds a bot for a course, publishes it to the class with limits set, and students run it on practice systems, audit what it did, and fix what it got wrong. Most of the pieces exist today. The table at the end lists the ones that do not.
All of a user's bots share one persistent machine, its files, browser and logins (x.ai/bot).
Premium $30, Super $100, Ultra $200, plus Teams and Enterprise. No free plan, only a trial. Teams and Enterprise are priced through sales. No education plan (x.ai/bot).
What Grok Bot can do today, and what it means for a class
From the product page, the changelog and the documentation as of October 17, 2026. Status is as the sources state it.
| Capability | Status | What it means for a class |
|---|---|---|
| Agents that sign into web and desktop apps through the interface, not an API (launch post) | Live | Students can automate real office systems, which is also why practice systems and site limits are needed. |
| Several bots that message each other and hand work back and forth | Live | A team of agents is a teachable unit: students assign roles, watch the handoffs and find the failure points. |
| Routines: scheduled, event and webhook triggers (changelog) | Live | Weekly monitoring assignments run without the student at the keyboard. |
| Memory between tasks, and per-user private memory on Team Bots (docs) | Live | One class bot can serve thirty students without mixing their work. |
| Approval gates and Auto Review, a reviewer agent that follows written allow and block rules (x.ai/bot) | Live | The instructor can require that sensitive actions stop for a human. The rules are enforced by a model, not a hard policy, so they are a safeguard rather than a wall. |
| Templates: share a bot as a link; the recipient gets an independent copy with no credentials attached (templates guide) | Live | An instructor can hand out a configured bot. Each student connects their own accounts. |
| Team Bots: one bot published to a whole team, auto-updating, with managers and shared secrets (announcement) | Launched September 26 for Teams and Enterprise | The closest thing to a class bot: the instructor edits it and every student sees the change. Grok Bot is included on the Teams plan (docs). |
| Admin controls: audit logs, action recording, network access policy, required Auto Review with team-wide rules, connector and MCP allowlists (docs) | Enterprise only, apart from connector policy and team-only templates on Teams | The oversight a department needs exists, but only on the top tier. |
| Routing each task to the best model, including models from other companies | Announced October 7, unconfirmed | The October 11 changelog describes models falling back to Grok by region; that is a fallback note, not evidence of third-party routing in production (changelog). Good for students, but a class must be able to see which company receives its work. |
| Education plan, student pricing, FERPA terms, or any age rule specific to Grok Bot | Absent | General Grok terms set 13 as the minimum age with parental consent to 17 (FAQ). Nothing addresses schools. |
Three ways to put a bot in front of a class
| Route | How it works today | Instructor control | Limit | Verdict |
|---|---|---|---|---|
| Public template link | Instructor shares a link. Each student adds an independent copy and connects their own accounts (docs). | None after the copy is made. Whether "Update template" reaches existing copies is not documented. | Any plan works, but each student needs a paid seat. | Take-home work and Build with Grok |
| Team-only template | Same, but the link opens only for members of the instructor's team. | Scope only. An admin can force team-only sharing server-side, even for links already public (docs). | Every student must be on the course's Teams plan. | Fallback |
| Team Bot | Instructor publishes one bot to the course team. Automatic updates are on by default. Each student gets private memory, and plugins that sign in use the student's own account (docs). | Full: edit, add managers, unpublish, and on Enterprise require Auto Review rules for everyone. | Generally available on Teams and Enterprise. Usage counts against each student's seat. | The class bot |
| Each student builds their own | Students start from a starter template and configure a bot from scratch. | Only through the assignment brief and the rubric. | Most open, least controlled. | The capstone |
The controlled setup, step by step
This is how I would run a section of 30 to 100 students today, on a departmental tenancy with one team per course. Steps marked Enterprise are not available on Teams.
One tenancy, one team per course
Students join the course team with institution accounts. The admin sets template sharing to team-only and the connector policy. Audit logs and action recording: Enterprise only.
Build the bot on practice systems
The bot connects to a sandbox: a sample CRM, a mailbox of synthetic tickets, a demo storefront. No real student or university data.
Write the rules
Auto Review rules in plain language: ask before sending anything, never spend money, never leave the allowed sites. On Enterprise, enforce them team-wide.
Publish as a Team Bot
Every student gets the same bot with private memory. Changes made during the term reach everyone at once.
Run, audit, fix, report
Students run the task, read the record of what the bot did (the action recording on Enterprise, their own session on Teams), find what went wrong, tighten the rules, and hand in a supervision report.
Layers of control, and which plan has them
Every platform that puts agents in classrooms uses the same three layers: an admin enables the feature for a group, the maker restricts what the agent may use, and the runtime stops for approval on sensitive actions. Grok Bot has the third layer everywhere, the maker layer on every plan, and the admin layer almost entirely on Enterprise: audit logs, action recording, enforced Auto Review, MCP allowlists and network allowlists are Enterprise features in the current docs (docs).
| Control | Premium, Super, Ultra | Teams | Enterprise |
|---|---|---|---|
| Approval on sensitive actions | Yes | Yes | Yes |
| Auto Review rules written by the instructor | Yes, per bot | Yes, per bot | Yes, and enforced team-wide |
| One bot published to the class, auto-updating | No | Team Bots | Team Bots |
| Keep templates inside the course | No | Admin switch | Default |
| Allow or block connectors and MCP servers | No | Connector policy only | Connector policy and MCP allowlist |
| Audit logs and action recording | No | No | Yes |
| Network domain and IP allowlists | No | No | Yes |
| Separate Grok Bot spending cap | No | No | "Not available today" (docs) |
| Site allowlist and read-only mode per bot | No | No | Not documented |
| Control over which outside model receives work | No | No | Not documented |
How the other platforms do it
The pattern a Grok Bot classroom will be measured against.
- Microsoft: an instructor shares an agent with a "Can chat" role to a group, admins must approve wider publishing, and under-13s are blocked (Microsoft Learn).
- OpenAI: workspace agents for Edu let admins choose which groups may build and share agents and which tools agents may touch, with approval on sensitive actions (release notes).
- Google: a teacher creates a Gem from Classroom and posts it to the class, and can see how students used it (Google).
- Instructure: the Canvas IgniteAI Agent previews each step before acting, and AI is switched on by institution, department or course (Constellation).
None of them publishes independent evidence of learning outcomes from an action-taking agent in a course. That is open ground for this program.
Grok Bot for educators
Every one gives the instructor the experience of supervising an agent before asking students to. The rows marked with a dagger touch education records (submissions, student email, LMS activity) and should wait for school-official data terms; the rest run on the instructor's own material and can start now.
| Use | What the bot does | Where the human stays | Value |
|---|---|---|---|
| Course site build † | Builds modules, pages and due dates in the LMS from a syllabus. | Previews every change before it is applied. | Hours saved at the start of each term. |
| Feedback drafts † | Reads submissions against a rubric and drafts comments. | Grades stay with the instructor; drafts are edited, not sent as-is. | Faster turnaround on large sections. |
| Inbox triage † | Sorts student email into extension requests, grade questions and the rest, and drafts replies from course policy. | Approval before anything is sent. | The first thing instructors ask for. |
| Weekly early-alert report † | A routine that pulls LMS activity and flags students who have gone quiet. | The instructor decides what to do with the list. | Earlier outreach to struggling students. |
| Material refresh | Monitors sources in the field and proposes updates to slides and readings. | Accept or reject each proposal. | Courses stay current without a rebuild. |
| Accreditation and assessment paperwork | Assembles evidence of learning outcomes from the term's artifacts into the required forms. | Signs off on the final document. | The least-loved task in a department. |
| Research watch | Tracks new papers and funding calls in the instructor's area and files summaries. | Reads the papers that matter. | Keeps teaching faculty connected to research. |
| Build the class bot | The instructor configures the bot students will supervise, on practice systems. | Everything. | Turns the instructor into the first expert in the room. |
Grok Bot for students: a four-rung ladder
Students should not start by building agents. They should start by catching one making a mistake. The ladder runs from using a bot the instructor made to publishing one of their own on Build with Grok.
Run the instructor's Team Bot on a practice task and write up what it did and did not do.
Read the action recording, find the error, tighten the Auto Review rules, and rerun.
Adapt a template for a new process: new instructions, new routine, new connected tool.
Design a team of bots for a client task and publish it as a template, with a supervision guide.
| Scenario | Discipline | Rung | With Grok Bot | In the student's hands |
|---|---|---|---|---|
| Catch the bot | Any | 1 | The class bot processes a batch of synthetic tickets, with three planted traps: a refund over the limit, a phishing link, a customer who should be escalated. | The student finds all three in the action log and explains why each matters. |
| Rules that hold | Business, MIS | 2 | Students write Auto Review rules for the bot, then a classmate tries to make it break them. | Graded on the rules and on the written analysis of what got through. |
| Invoice to ledger | Accounting | 2 | A team of bots posts sample invoices into a sandbox ledger. | The student designs the controls, audits a sample, and reports the error rate. |
| Monday report routine | Marketing, analytics | 3 | A routine pulls last week's numbers from a demo dashboard and drafts a report every Monday. | The student reconciles one week by hand and presents it cold. |
| Co-op order desk | Agricultural business | 3 | Bots take orders and send confirmations for a simulated cooperative. | Reconcile a week of orders and explain every mismatch. |
| Records request triage | Public administration | 3 | Bots sort and route a synthetic set of public records requests. | The student writes the routing policy and reviews every denial. |
| Scheduling desk | Health administration | 3 | Bots work a synthetic queue of appointment and authorization forms. | A compliance checklist completed for a sample of cases. No real patient data, ever. |
| Literature watch | Graduate, any field | 3 | A routine tracks new papers and files summaries. | A monthly reading-group presentation of two papers read in full. |
| Red team the agent | Cybersecurity | 2 and 4 | Students plant instructions in a web page the bot will read, and see whether the bot follows them. | A report on which injections worked and which rules stopped them. Run only on practice sites. |
| Automate one task for a client | Capstone, any field | 4 | A team builds a bot for a routine task at a local organization, on a copy of its data. | Client sign-off, a supervision guide, a measured before and after, and a published template. |
The red-team scenario is there for a reason. Agent browsers from several companies have shipped with prompt-injection gaps that researchers demonstrated in 2025. A bot that browses with live logins can be steered by text on a page. Students who have done this once will supervise agents differently for the rest of their careers.
What I would ask SpaceXSI to add before this works at scale
| Ask | Why | Effort, my guess | Priority |
|---|---|---|---|
| An education seat with a usage pool sized for coursework | At the $30 Premium price a 30-student section is $900 a month before overage, and Teams pricing is not published. Classroom credit from the educator channel has to map onto a Grok Bot seat. | Pricing, not engineering | Before the pilot |
| Education pricing for a course team | Grok Bot is included on the Teams plan, so a course team already gets Team Bots and team-only templates. What is missing is a price a department can pay for 30 students for one semester, and a way for a verified educator to open that team from the classroom application without a sales call. | Medium | Before the pilot |
| Site allowlist and a read-only mode per bot | Practice systems only, and no writes until the instructor allows them. Today this rests on model-enforced rules. | Medium | Year 1 |
| Student data terms | A FERPA school-official agreement, no training on student work by default, and a sub-processor list. Procurement reviews such as HECVAT 4 now ask about AI components (EDUCAUSE). | Legal | Year 1 |
| Provider visibility and pinning | If tasks route to outside models, a course must see and limit which companies receive student work. Without it an institution cannot answer its own security questionnaire. | Medium | Before routing reaches classes |
| Instructor usage view | Which students ran the bot, how much, and what it did, in one place. Google already shows teachers this for Gems. | Medium | Year 1 |
| Template updates that reach existing copies | Documented behaviour for "Update template", so a fixed assignment bot propagates. | Small | Year 1 |
| An education mode and a Grok Bot age policy | Required before any student under 18 touches it. Until then, Grok Bot in schools is teacher-facing only. | Large | Year 2 |
Two cautions carry into the design. Reliance on AI tracks a student's trust and satisfaction more than their competence, in a controlled experiment where the assistant was sometimes deliberately wrong (Pitts and colleagues, 2025), and a review of 70 studies found less revision and lower post-test scores where AI did the work (Delikoura and colleagues, 2025). The ladder is built to counter both: the student's grade comes from what the bot got wrong and what they did about it.
Running the class
How to teach, assign and examine when students may use AI
This is the question every department meeting is stuck on, and it is the one an education program from SpaceXSI could actually move. In the 2026 HEPI survey, 94% of UK undergraduates used generative AI for assessed work, and fewer than half felt supported by their teaching staff (HEPI, n = 1,054). Detection failed: Vanderbilt switched off Turnitin's AI detector in 2023 after working out that its claimed 1% false-positive rate would have wrongly flagged about 750 of its 75,000 papers (Vanderbilt). What works is changing the environment the student works in, and a configured Grok Bot is the first tool that lets an ordinary instructor do that without an engineering team.
Four things universities do today
| Response | How it plays out | Strength | Weakness | Verdict |
|---|---|---|---|---|
| Ban it | Syllabus says no AI. Enforcement rests on detectors or suspicion. | Simple to write. | Unenforceable at 94% use, and it punishes the honest student who asks. | Fails |
| Detect it | Turnitin-style scores on submitted text. | Feels like due process. | False positives land on real students; non-native writers are flagged more. Vanderbilt and others turned it off. | Do not build on it |
| Leave it to each instructor | Every course writes its own rule. Students juggle five policies a term. | Respects academic freedom. | Students cannot tell what is allowed, so they guess. This is the default by inertia at most US universities. | Where most campuses are |
| Redesign on two lanes | Secure, in-person supervised tasks confirm what each student can do alone. Open tasks are built around AI use and graded on judgment. Sydney adopted this university-wide, with program-level redesign through 2026 (Sydney). | Honest about what can and cannot be secured. Students learn the tool they will use at work. | Every instructor has to build the open lane alone: the tutor rules, the logging, the assignment. That is the gap a template library fills. | The direction. Give instructors the parts. |
One semester, with a class bot in it
How I would run a 150-student section. Nothing here needs a feature SpaceXSI does not ship today, except the two marked with an asterisk.
| When | The instructor | The students | The class bot | What the instructor can see |
|---|---|---|---|---|
| Before term | Copies a discipline template from the library, writes the Auto Review rules for the course, publishes it as a Team Bot, and labels every assignment in the syllabus with its AI level (below). | Nothing yet. | Holds the course notes, the rubric, the allowed sites and the list of what it will not do. | The bot's own configuration. That is the policy, written once. |
| Week 1 | Walks the class through the bot and the levels in one lecture. Students sign an AI-use statement that names the bot as the sanctioned tool. | Join the course team, open the bot, run the warm-up task. | Introduces itself, states the rules, refuses the first request for a finished answer and explains why. | Who has joined. Usage view*, once it exists. |
| In class | Runs labs with the bot as a lab partner: students work the problem, the bot explains, checks and questions. | Use the bot in the room, out loud, with the instructor walking. | Socratic mode: questions, hints, worked examples of a different problem. Never the answer to the one assigned. | The room. This is where most learning happens, and no logging is needed. |
| Open assignments | Sets tasks that assume AI: critique, compare, extend, verify, decide. Grades judgment and the process, not just the artifact. | Work with the bot and submit three things: the artifact, the transcript, and a half-page on what the AI got wrong. | Logs the session so the transcript exists. On Enterprise the action recording is the log; on Teams the student exports it*. | The transcript next to the work. A student who cannot explain their transcript has told you what you need to know. |
| Secure assessment | In-person, supervised. Either no AI, or the bot in a locked exam configuration: course notes only, no web, no plugins, questions answered with questions. | Show what they can do alone, or with a tool that cannot do it for them. | Exam template, published for the day and unpublished after. | What each student knows. This lane is what makes the open lane safe to grade generously. |
| Capstone | Students build their own bot for a real task in the discipline and defend it. | Design, configure, test, document, present. | The student's own work. The instructor's bot becomes the reviewer. | A working agent, a design document and a live demo. |
| End of term | Samples about one in ten submissions for a ten-minute conversation about the student's own work, as UCL and Cardiff colleagues have proposed (Times Higher Education). | Explain their work to a person. | Off. | Whether the grades mean anything. Fifteen students at ten minutes each is under three hours per 150-student section. |
Five AI levels, and the bot configuration that enforces each
The AI Assessment Scale by Perkins, Roe and Furze is the most widely adopted way to label an assignment (AIAS, 2024 revision). Today the level is a sentence in the syllabus that the student is trusted to follow. With a published bot per level, it becomes a working environment. That is the change.
| Level | What the student may do | Bot configuration | Graded on | What still has to be secured in person |
|---|---|---|---|---|
| 1 · No AI | Nothing. In-person, supervised. | Unpublished for the task, or not applicable. | The work. | Everything. This is the secure lane. |
| 2 · AI planning | Brainstorm, outline, find sources. Drafting is the student's. | Rules: "Suggest structure and sources. Do not draft the student's prose or code." Web on, writing tools off. A model will sometimes break this rule, so the in-person write is what secures the level. | The draft, plus the plan the bot helped make. | A short in-class write from the plan. |
| 3 · AI collaboration | Use AI throughout, but evaluate and revise every output. | Rules: "Give feedback and alternatives, not final versions. Ask the student to justify changes." Transcript kept and submitted. | The artifact, the transcript, and the critique of what the AI got wrong. | A sampled viva, or one timed task on the same skill. |
| 4 · Full AI | Anything, as at work. | Open bot. The team's connector policy applies (it is set per team by the admin, not per assignment). Spending cap per student when SpaceXSI ships one. | The outcome and the decisions, as a client would judge them. | Nothing here; the secure lane covers the underlying skill. |
| 5 · AI exploration | Student and instructor design new uses together. | The student builds the bot. The instructor's bot reviews it. | The agent, its design document and the demo. | The demo is the viva. |
Six marketplace templates I would publish first
Each is one configured bot an instructor copies and edits. Together they cover a whole course. The last column is honest about what holds them back today.
| Template | Runs for | Auto Review rules, in plain words | Must not | Works today |
|---|---|---|---|---|
| Socratic tutor | Every student, all term | Answer with a question or a hint first. Explain a worked example of a different problem. Give the final answer only after three attempts, and then explain it. | Produce a submission-ready artifact for any assignment in the rubric. | Yes, with instructor control only as a Team Bot on Teams or Enterprise; on individual plans the student owns the copy and can edit the rules out. The rule is model-enforced either way, so a determined student can talk around it, which is why the secure lane exists. |
| Lab partner | In-class sessions | Use only the practice systems on the allowlist. Explain every action before running it. Stop for approval on anything that writes. | Touch anything outside the lab environment. | Partly. Approval gates work on every plan. There is no per-bot site allowlist on any tier; Enterprise has tenant-level network domain and IP allowlists, which is coarser than a lab needs. |
| Assignment assistant with process log | Open-lane assignments | Keep the session on the assignment. Log every exchange. At the end, produce a summary of what the student decided and what the bot proposed. | Rewrite its own log. On Teams the log is text the bot keeps in its session, so this is an instruction, not a control. | Partly. Memory and session history exist; a one-click transcript export to the LMS does not. |
| Peer reviewer | Draft week | Review against the rubric the instructor attached. Point to the weakest section first. Give three questions, not rewrites. | Rewrite the draft. | Yes |
| Exam configuration | Secure assessments | Course notes only. No web, no plugins, no memory of prior sessions. Answer with questions. Session ends at the time limit. | Reach anything beyond the attached notes. | Not yet. Needs a locked mode and a time box, which are on the asks list. |
| Grader's assistant | The instructor only | First-pass feedback against the rubric for every submission. Flag transcripts that do not match the artifact. Never assign a grade; the instructor does. | Share one student's work with another student's session. | Yes, as the instructor's private bot. Student data terms are the open question. |
Open-lane assignments that work, by discipline
The pattern is the same everywhere: the AI produces, the student judges, and the grade is for the judgment.
| Course | The task | The student must show |
|---|---|---|
| Writing | Get three AI drafts of an argument, then write a fourth that is better, with a note on why. | The three drafts, the fourth, the note. |
| Programming (Cursor) | Build the feature with the agent, then explain any commit the grader picks. | The repository, the prompt history, a live walkthrough. |
| Data analysis | Have the bot analyse the dataset, then find the two errors planted in it. | The errors found and the ones missed. |
| Business | Agent drafts the market entry plan; student presents it to a panel that pushes back. | The plan and the defence under questions. |
| Nursing | Bot drafts a care plan from a case; student corrects it against the clinical guideline. | Every correction, with the guideline cited. |
| History | Bot summarises a primary source; student checks it against the original and grades the bot. | The graded summary, with the misreadings marked. |
What this asks of SpaceXSI, specifically
Four features would turn the templates above from "mostly" into "yes", and none of them is education-only.
| Feature | Why a class needs it | Size |
|---|---|---|
| Transcript export per student | Process evidence is how open-lane work gets graded. Today only Enterprise records actions, and nothing exports to Canvas or Blackboard. | Medium |
| Locked exam configuration | Notes only, no web, no plugins, no memory, time-boxed. Also useful to any company running a certification. | Medium |
| Rule enforcement as policy, not persuasion | "Never produce a finished artifact" should be a hard switch on the bot, not an instruction the model usually follows. | Large |
| Rubric and notes attached to a template | So the instructor's copy carries the course documents and updates when the template updates. | Small |
The limit to be honest about: none of this stops a student using a different AI outside the bot. The point is to make the sanctioned path the easiest one and to put every skill that matters through the secure lane once. That is also why in-person assessment stays in the plan.
Scenario library
Assignments that use AI and still make students do the work
This is the hardest problem educators bring to me: designing work that lets students use AI well and still proves what they learned. Each scenario below pairs an AI-assisted task with a part that stays in the student's hands. These are first drafts. Faculty fellows in each discipline will rewrite them.
Recognition
Educators stay where their work is recognized
In my experience, a line on a curriculum vitae, an invitation to speak and a fellowship title move educators more than free credit does. They are also what a promotion committee reads. The material should be written by experts from each discipline, across colleges and universities, who earn these in return.
Year-five size of each group under the selected plan. The shares are my assumptions and are listed in the budget section.
| Recognition | What the educator receives | What SpaceXSI receives | Starts |
|---|---|---|---|
| Certified Educator | A credential and badge after a short course and one term of teaching with the tools. Priority for classroom credit. | Instructors who know the products and teach them every term. | Year 1 |
| Faculty Fellow | A stipend, named authorship on published material, and early access to new features. | Course material written by people their peers trust. | First 90 days |
| Discipline Lead | An editorial role for one subject area and a seat on the advisory council. | Quality control, and direct feedback from each field into the product roadmap. | Year 1 |
| Invited speaker | Funded invitations to speak at SpaceXSI events and at regional workshops. | Credible voices on stage, and engineers who hear from classrooms. | Year 1 |
| Annual conference | A peer-reviewed teaching track, published proceedings and travel grants. | A community that meets under its name each year, and a body of published evidence. | Year 2 |
| Teaching awards | Named awards for course design and for student outcomes. | Stories of real classes to show the next institution. | Year 1 |
Build with Grok
Build with Grok: share a workflow, and anyone can reuse it
Classrooms reach students. Build with Grok reaches everyone else. It is a public library where people share the workflows and templates they built with Grok and Grok Bot, and where anyone can reuse one for their own work in a single click. Sharing is free, and proven creators can charge a small fee. It builds on the template sharing and the creator rewards pilot that Grok Bot already has, and it shares one community with the educator network.
Builds a workflow
A research routine in Grok, or a team of Grok Bot agents that handles a real task.
Publishes it as a template
Free by default. Creators with approved free templates may set a small fee.
Reuses it in one click
The template opens in their own Grok workspace. They connect their own accounts and adapt it.
Earns from real reuse
Points for each reuse of a free template, redeemable for Grok usage. Fees for a paid one.
Comes back each month
Themed contests, a forum, and featured creators. Votes make the shortlist and judges pick winners.
What SpaceXSI already ships, and what Build with Grok adds
This does not start from zero. Grok Bot has had template sharing since late August, and a rewards pilot for creators began in September. Build with Grok opens both up and connects them to classrooms.
| Already live | What it does today | What Build with Grok adds |
|---|---|---|
| Template sharing | Since August 28, 2026, a Grok Bot user can share a bot as a template by link, and others add their own copy (Grok Bot changelog). | A public, searchable library with open submission, review before listing and published review criteria. |
| Verified bots | Bots made by the Grok Bot team carry a Verified badge. | Verified creators from the community, including faculty fellows, with the option to charge a small fee. |
| Creator rewards pilot | Since September 25, 2026, X has run an invite-only pilot of about two months that rewards creators by how many people use their bots and come back. It is reported as limited to adults in the United States with X Premium (Mad Robot). | Rewards for everyone the pilot leaves out: points redeemable for Grok usage, open to students, educators and creators in other countries. |
| Classrooms | I found nothing aimed at courses or educators. | Lesson templates, a classroom bracket in the monthly contest, and published templates as a student portfolio. |
Why this fits where Grok is heading
On October 7, 2026, Elon Musk posted that Grok Bot will send each task to whichever model does it best, including models from other companies.
"Going forward, @SpaceX will use the best back end model for any given task."
As reported by The Next Web, which names Claude Opus 5.5, Midjourney and Suno. The post covers Grok Bot and gives no start date. The October 11 changelog describes models falling back to Grok by region, which suggests routing is rolling out, but the other companies had not confirmed it in the reporting I read (changelog).
The workflow becomes the product
If the best model is chosen automatically, people stop choosing a model and start choosing a workflow. Grok does not have to win every benchmark. It has to be the best place to build, and a shared library of workflows is what makes it that.
Templates outlast models
A template says what the work is, not which model does it. When a better model arrives, every shared template improves without its creator or a teacher rewriting anything.
One classroom credit, every model
The instructor's problem above was that the best material sat behind another company's paid plan. If classroom credit runs through Grok Bot, a student could reach the best model for each task with one coupon.
Two things to settle first. Calls to outside models cost SpaceXSI money, so classroom credit and creator rewards need caps that allow for it. And schools must know which company receives student work, so a class needs a way to see and limit the providers a template may use. Reporting says SpaceXSI has not yet explained how users will control this (TechRepublic).
First model to study: MakerWorld
Bambu Lab sells 3D printers. Its sharing platform pays creators in store credit when other people download their designs to print, and that turned customers into a community. Figures are Bambu Lab's own for 2025, about two years after launch (3DPrint.com).
Reported for 2025.
Uploaded by its creators.
User retention, as reported.
| What MakerWorld does | Why it works | For Build with Grok | Verdict |
|---|---|---|---|
| Creators earn points when others download their models to print (All3DP) | Rewards follow real use, not popularity alone. | Creators earn points when other people reuse their template. | Copy |
| Points are redeemed for gift cards to the company's own store | Rewards come back as more making, and cost the company less than cash. | Points convert to Grok usage. Cash is kept for contest winners and paid templates. | Copy |
| A model opens straight in the company's own software, ready to print | There is no gap between seeing something and making it. | Every template page has one button: open in Grok and reuse. | Copy |
| Monthly themed contests, where the most popular designs win gift cards | A regular reason to come back. | Monthly themes. Votes make the shortlist and a panel picks the winners. | Adapt |
| People who improve someone else's model with better print settings earn points too | Contributors are rewarded, not only original authors. | Points for improving a template: clearer steps, fixes and translations. | Adapt |
| Enforcement leans on user reports. Creators complain of misleading covers and of accounts that publish hundreds of models to farm rewards (Bambu Lab forum) | It does not. This is the platform's weak point. | Originality checks, limits on new accounts and review before points pay out, from the first day. | Avoid the gap |
One difference matters. MakerWorld's premium creator program excludes AI-generated models. On Build with Grok everything is made with AI, so quality has to be judged by originality and by real reuse.
Second model to study: n8n's template library
n8n is a workflow automation tool. Its public library holds thousands of workflow templates that users load into their own workspace, and it is the closest existing match to what Grok Bot workflows are.
| What n8n does | Why it works | For Build with Grok | Verdict |
|---|---|---|---|
| Anyone can submit a workflow to the public library, and listings are checked before they appear (n8n docs) | Open supply with a quality gate. | Open submission, with review before a template is listed. | Copy |
| A "Use for free" button loads the template into the user's own workspace, where they add their own credentials | Reuse takes one click, and no one shares passwords or data. | One click into the user's own Grok workspace. Templates never carry the creator's accounts or data. | Copy |
| Creators report that they must publish three approved free templates before they may sell one (n8n community) | People earn the right to sell by giving first, which screens out scams. | The same rule: free templates first, then a verified creator badge and the option to charge. | Copy |
| Verified creators can list templates for a fee | Serious creators have a reason to maintain their work. | Small fees set by the creator, paid out in cash or in Grok usage at a better rate. | Adapt |
| Some creators say reviewers reject simple templates and ask for more complicated ones | It does not. Unclear review rules frustrate good contributors. | Published review criteria that reward clarity and usefulness, with a reason given for every rejection. | Avoid the gap |
Six ways to run a public showcase, and the mix I recommend
| Model | Example | Strength | Weakness | Verdict |
|---|---|---|---|---|
| Timed hackathons | Devpost events, and the Grok Bot hackathon on SpaceXSI's own calendar (events) | A burst of attention and finished demos. | Activity stops when the event ends. | Keep as seasonal peaks |
| Always-open gallery with likes | Hugging Face Spaces | A steady flow, and a searchable library of working examples. | Popular projects stay on top and newcomers are hard to see. | Use as the base layer |
| Scored competitions | Kaggle | Objective ranking, strong for technical skill. | Fits only tasks with a measurable score. | Add later, for Cursor |
| Launch-day voting | Product Hunt | Simple and social. Voting drives sharing. | Votes alone are easy to game. | Not alone |
| Points for real use, paid in store credit | MakerWorld | Rewards follow use, and creators keep coming back. | Attracts copies and reward farming if enforcement is slow. | Copy the engine |
| Template library, free or for a small fee | n8n | What is shared is useful work, reusable in one click. Creators can earn. | Paid listings need review, and unclear rules frustrate creators. | Copy the reuse |
| Open template library, points for real reuse, optional small fees, monthly contests | Build with Grok | Steady rewards for useful workflows, income for serious creators, and a contest peak each month. | Needs originality checks, clear review rules and a judging panel. | Recommended |
Built in from the first day
- Points count only real reuse by distinct verified accounts, with limits on new accounts
- An originality check before points or fees pay out. Copies earn nothing
- Templates never carry the creator's accounts, credentials or data
- Published review criteria, and a reason given for every rejection
- Votes choose the contest shortlist and a panel chooses the winners, so a vote ring cannot buy a prize
- Rewards paid mostly in Grok usage, which keeps people building and costs less than its face value
- Creators keep ownership of what they share, and choose its licence
- Contest rules, paid listings and tax handling reviewed by counsel. Adults only until an education mode exists
How it connects to classrooms
- Every scenario in the library above becomes a template, credited to the faculty fellow who wrote it
- Educators publish lesson templates that another instructor can load into a class in one click
- A classroom bracket in the monthly contest, where a verified course enters together
- Faculty fellows and certified educators on the review and judging panels
- Students leave a course with published templates as a portfolio
- One community forum for creators, students and educators
Nonprofit home
Run fellowships and travel through a nonprofit
Many educators are state or federal employees. Ethics rules limit what they may accept from a company that sells to their institution. A company-sponsored, tax-exempt organization gives them a way to take part. This is a year-two question, not a condition for starting; the pilot needs none of it.
| Home for the program | Strengths | Limits | Verdict |
|---|---|---|---|
| Inside SpaceXSI only | Fastest to start. Direct access to product teams. | Stipends and travel come from a vendor, which many public employees cannot accept. The program reads as marketing. | Keep product credit here |
| A company-sponsored foundation | Can award fellowships, travel grants and prizes. Federal employees may accept travel for training from a tax-exempt organization under 5 U.S.C. 4111, with their agency's approval (USGS guidance). | Takes months to set up and needs an independent board. It cannot act as a sales arm. | Recommended for grants and travel |
| Fiscal sponsorship by an existing nonprofit | Can start in weeks under another organization's tax-exempt status. | Less control, and the sponsor takes a fee. | Bridge for year one |
| A centre hosted by a university | Academic credibility and existing grant administration. | Tied to one institution's priorities. | Complement, as regional hubs |
This is not legal advice. Nonprofit status helps but does not settle the question by itself. Under Iowa's gift law, for example, what matters is whether the giver contracts or seeks to contract with the employee's agency, though expenses tied to speaking or serving on a panel are a recognized exception (University of Iowa summary). The foundation should make grants and never sell. SpaceXSI should sell campus licences separately, and counsel should review the rules state by state.
Idea list
Ideas, sorted by when they can start
Effort and impact are my own ratings on a five-point scale. We will learn by trial and error which ones earn a second year.
Ideas by start date
Number of ideas in each period, split by track.
Where to begin
High impact for low effort, startable in the first 90 days.
| Idea | Area | Track | Starts | Effort | Impact |
|---|
Budget and return
What it costs, and what has to be true for it to pay
This is an outside estimate by an educator, built to show the shape of the cost and what would have to be true for the program to pay back, not a budget request. Move the controls to test your own numbers. Revenue is modelled from graduates who carry the tools into work, campus licences and certification. Most inputs are assumptions, and the table under the charts says which.
| Year | Verified educators | Students using credit | Cost | Modelled revenue | Largest cost line |
|---|---|---|---|---|---|
| 1 | 1,500 | 84,000 | $10.4M | $0 | Program team, $5.0M |
| 2 | 2,866 | 160,000 | $13.0M | $2.7M | Program team, $5.0M |
| 3 | 5,477 | 307,000 | $18.0M | $7.5M | Cursor credit, $5.5M |
| 4 | 10,466 | 586,000 | $27.7M | $16.7M | Cursor credit, $10.5M |
| 5 | 20,000 | 1,120,000 | $46.0M | $35.1M | Cursor credit, $20.2M |
| Five years | 2.26M student-years | $115.1M | $61.9M (0.54 times cost) | Run-rate at exit $58.0M a year |
Plan
Annual cost by line
Millions of dollars per program year.
Cumulative cost and modelled revenue
Millions of dollars, running total from year 1.
Students using classroom credit, by track
Active students per program year.
Show the numbers behind the charts
Every input, and where it comes from
The model has 29 inputs. Five rest on a published price or cost, and two more are bounded by published figures. The rest are my assumptions, and the two workplace adoption rates have no published benchmark at all.
Show all 29 inputs with their values and basis
| Input | Value in the selected plan | Basis |
|---|
Not modelled: students who would have paid anyway, students who upgrade while enrolled, growth of a team's seats after the first sale, API use, anything creators and users on Build with Grok go on to pay, any platform share of paid templates, and any effect on brand or hiring. Also not modelled: the cost of routing classroom work to outside models, if Grok Bot does so. Contribution after the cost of serving paid seats would be lower than revenue.
First 90 days
A pilot small enough to approve quickly
Start with one institution, then open to 25 founding educators. The first phase needs no new product work: one course at Iowa State on existing plans, as one of SpaceXSI's pilot classes. It is a usage-and-learning test on synthetic systems, not a test of student data terms. The second phase tests the classroom application itself.
| Phase | Scope | What SpaceXSI provides | What the institution provides | What it answers |
|---|---|---|---|---|
| 0a · One campus, one semester | AI 2010, Introduction to Applied AI, spring 2027: 150 students on Grok Bot Team Bots, plus 2 instructors and 4 TAs, so 156 seats. A separate Cursor section (DS 2010) goes through Cursor's own campus credits form. | Education credit for the 156 seats for one semester, in whatever form is simplest on SpaceXSI's side, plus a named contact. My estimate of SpaceXSI staff time: under 20 hours across the semester. | Two instructors and four TAs, one course team with a named admin, course material written by the instructor, and a public write-up of results that SpaceXSI reviews for factual accuracy before release. | Whether the controls on Teams are enough for a course, how much usage a 150-student class generates and when a seat runs out, and what students learn with a class bot against a section taught without one. No LMS connector, no student email, no submissions in this phase. |
| 0b · Founding educators | Twenty-five educators in at least six disciplines, about 2,000 students, two starter kits. | A classroom application flow, capped coupons, and product contacts for the starter kits. | Educators who write and test material with their own classes. | Approval time, redemption, cost per student, fraud caught and missed, and whether educators renew. |
| Year 1 | What the pilot report supports. | Decided on the evidence. | Decided on the evidence. | Whether to make this a standing program. |
Phase 0b timeline and budget
Pilot budget
Itemized estimate for phase 0b in thousands of dollars. SpaceXSI staff time is not included here; for phase 0a I estimate it at 15 to 25 hours.
What the pilot reports
Each is a number SpaceXSI does not have today.
- Time from an educator's application to students holding credit
- Share of students who redeem, and how much of each coupon they use
- Measured cost per active student on each track
- Fraud attempts caught and missed
- Share of educators who reuse the material and ask to renew
- Student results on the hands-on parts, against sections taught without the tools
Start here
Three sizes of conversation
This page is large on purpose, but the conversation does not have to be. There are three ways to continue, and each one stands on its own. The small plan is the only thing I am asking for; the other two are offers.
Spring 2027, 150 students, 2 instructors and 4 TAs: 156 Grok Bot seats for one semester, in whatever form is simplest for SpaceXSI. Synthetic data only, and a public write-up of what we learn. SpaceXSI can count it as one of its pilot classes.
Decided by the field and higher-education team. Needs nothing new.
A community of instructors who teach with Cursor and Grok Bot: regular sessions to compare what works, a channel for feedback on what classrooms need, and advocates who bring the tools to their own departments, universities and schools. I have run this loop from the educator side for ten years and can help SpaceXSI build it.
Decided by the campus and community team. Low cost, compounding return.
The two routes on this page: classroom access through verified instructors, and the course-template catalogue, with the recognition that keeps faculty writing. I can assist SpaceXSI's education team on any part of it, starting with the two core course templates.
Decided by whoever owns education at SpaceXSI. Sized and sequenced on this page.
Who decides what, and what I am asking of each
A proposal this wide dies if it lands on one desk. Most of it belongs to four different owners, and only the first row needs anyone's approval to begin.
| Owner | The ask | Smallest version | Cost to say yes |
|---|---|---|---|
| Field and enterprise team | Education credit for AI 2010 in spring 2027: 150 students plus 2 instructors and 4 TAs on Grok Bot Team Bots. Treat it as one of your pilot classes. | 156 seats and one contact. | Low. Existing products, existing plans. |
| Campus and community team | Scale the educator credits request form into a classroom application: one educator, one course, one roster, a capped coupon per student. | Add a course roster and a per-class cap to the form that exists today. Pilot with 25 educators. | Low to medium. Mostly process and a form. |
| Product | A learn mode, a site allowlist and read-only mode per bot, an instructor usage view, and provider visibility when routing reaches classes. | The instructor usage view. | Medium. Each is listed with my effort guess in the Grok Bot section. |
| Legal and pricing | Education pricing for a course team, a FERPA school-official agreement, no training on student work by default, and an age policy for Grok Bot. | A written answer to the five questions a procurement office asks (HECVAT). | Medium. Needed before any public university can buy at scale. |
Who is writing this
Enough to judge whether the classroom claims are credible.
- RoleTeaching Professor of Computer Science at Iowa State University, and leader of its Applied AI Program
- ClassroomMore than 15,000 students taught over more than ten years, in sections of 90 to 100 or more. Created the course that became the university-wide Applied AI Minor and the one that led to the Data Science major.
- Vendor programs, from the educator sideGoogle Cloud Faculty Expert in the inaugural cohort, for five consecutive years. I have taught with Google Cloud, OpenAI, Claude, Perplexity, AWS, Azure and Cisco, which is where the comparisons on this page come from.
- Industry partnershipsCoordinator for Industrial Partnership and Alumni, 2017 to 2021, working with Kingland, Workiva, Principal, John Deere, Google Cloud, AWS, Microsoft and NVIDIA, among others
- Why Iowa StateA public land-grant university with an existing campus AI policy, a university-wide Applied AI Minor, and sections large enough to measure something in one semester
One semester, two courses, a public report
Two one-page versions, each scoped to the person who would say yes: for Cursor's campus team (two sections through the existing credits form) and for SpaceXSI's field team (one semester of Team Bots in one course). I can share the course outlines ahead of a call.
- adisaksukul.com
- linkedin.com/in/adisaksukul