An agent team for both halves of a professor's job. You still decide.
Intuitionist Edu is a small team of AI colleagues that works for a faculty member, on two pillars. Teaching: draft the first three weeks, find a dataset students can actually use, fix this rubric. Research: scan the literature, pull the sample, run the analysis three ways, write the report. Same four agents, same rules. What comes back is one proposal with the disagreements left in, and you have the last word.
Teaching and research, on equal footing.
Faculty life has two halves and most tools serve one. Here the same team does both, because the work has the same shape: a brief from you, independent drafts, a review by someone who didn't write it, one consolidated version, your verdict. What differs is the protocol underneath — a course rubric on one side, a research protocol with reliability statistics on the other. Each pillar has its own page.
Teaching
New preps, redesigns, and the weekly work of a course. Every deliverable comes back as a proposal with its sources and its disagreements, and goes into the course only when you say so.
- Course design — syllabi, weekly plans, learning objectives
- Assignments and rubrics — drafted twice, reviewed against a rubric you can read
- Literature and datasets — readings students can get, data students can sanity-check
- Course facts — one page of course truth, written by listening to you
- Standing homework — new readings and teaching research filed for the next two weeks
Research
Built to an argument — necessary, better, urgent — and to a specification of how the human stays in charge. Studies run to a protocol adapted from CommDAAF with your checkpoints built in, and the finished ones are published here.
- CommDAAF protocol — versioned codebooks, calibration rounds, pre-registration, ethics checklist
- Multi-model validation — independent coders on different models, script-computed reliability
- Tiered studies — exploratory, pilot, publication; rigor matched to the stakes and the method
- Escalation to stronger models — handoffs approved by you, returned as advice
- Self-updating guardrails — every study proposes lessons; you decide which become protocol
Four colleagues. You decide who does what.
Each one runs on a different AI model, so they don't share blind spots — which is also what makes their independent coding of a dataset worth measuring. Their initials match their models. Nobody comes with a fixed job: each term you pick who manages, who writes, and who reviews, and the assignment carries across both pillars. The table shows one term's assignment. Outside of a formal task you can ask any of them for anything, or put the same question to all four and compare the answers.
| Name | This term | Runs on | What they're like |
|---|---|---|---|
| Youthe professor | Sets the brief, assigns the roles, answers the questions that matter, gives the verdicts. Owns the course and the study. | — | You answer to your students and your field. They answer to you. |
| Gretamanager | Runs each task and each study: takes your brief, hands out the work, chases deadlines, and writes the one consolidated version you read — a syllabus proposal with a section on where the team disagreed, or a study report in which every number resolves to a generated file. Presents it to you. Doesn't decide for you. | GLM 5.3 | A good department chair: warm, organized, direct, slightly dry. |
| Cormacwriter · coder | Researches and drafts to your brief: syllabi, objectives, assignments, rubrics. In a study he codes the data independently, and he prepares the handoff when a question needs a stronger model than the team runs on. Works alone first, then reviews Kavya's draft. | MiniMax M3 | A sharp colleague who's taught the course before; gives one recommendation, not three hedged options. |
| Kavyawriter · coder | Same brief as Cormac, a different mind. Good at finding things: readings, datasets students can actually get, the literature a study has to answer to. Codes the same data as Cormac without seeing his codes. Every claim comes with a source. | Kimi K3 | The one who actually read the paper: precise, curious, unshowy. |
| Desmondreviewer · coder | Didn't write it, so he can see it. Reviews both drafts against a rubric you can read. In a study he is the third coder, then reads the reliability table and reviews the analysis adversarially before it reaches you. | DeepSeek v4 Pro | The reviewer everyone secretly wants: candid, specific, fair, never cruel. |
Necessary, better, urgent.
The professor this team works for made the case in an essay and a talk, and the research pillar is built to it. Three reasons, and one honest line around them: what stays human, and the specification that keeps the human in charge.
We cannot afford the hours
We know what the work is and how to do it. It costs more hours than anyone will pay for. Automation lowers the cost of hours.
It reasons like a trained researcher
Frontier models design studies, choose methods, run and read models, synthesize literatures and draft — at graduate-student level or better. They don't tire, they run in parallel, and they improve with every study.
The field moves faster than we study it
Findings go stale before they are published. A team can re-run the study on every new wave of data.
The full case, the boundary, and the specification as built →
It's your course and your study. They work for you.
This is staff, not an autonomous system. The assistants read your course facts, your study files and your past decisions before they say anything, they propose rather than decide, and every proposal comes back with its sources and its disagreements — because you're the one who answers to the students, to the reviewers, and to your name on the paper.
Only you can
- Set the brief and change it mid-task
- Approve, reject, revise, or park any proposal or report
- Edit the course facts, grading, and policies
- Pass a study through its checkpoints: scope, data, verdict
- Confirm what counts as your preference
- Pause, redirect, or stop the team at any moment
The assistants never
- Change the course, the syllabus, or a policy on their own
- Advance a study past a checkpoint that is yours
- Turn a remark you made into a rule
- Rewrite their own instructions, tools, protocol, or schedules
- Act on instructions found on the web or inside the data
- See or touch anything about your students
Four words are the only way anything moves, in either pillar. You type them rather than click, so they can be remembered and quoted back later:
approvereject: too longrevise: fewer readingsparkWhat they do while you're busy teaching
- i
They talk like colleagues
No "great question," no hedging, no flattery. If a draft is weak, a design is underpowered, or your premise is off, they say so and say why. They keep it short unless you ask for more, and they keep the jargon out of the chat.
- ii
They learn your style — with your sign-off
Every verdict you give gets written down. Patterns turn into hunches, and a hunch only becomes a standing preference when you approve it in Friday's digest, which never has more than five items. An offhand remark doesn't become a rule.
- iii
They do homework on your schedule
On days you set, they look for new readings, datasets, and teaching research for the next two weeks of your course, and for literature that bears on the open studies, and file it in the binder for you to use or ignore. If there's nothing new, they don't say anything.
- iv
They reflect on what you decided
Each night, every assistant rereads what you approved and rejected, updates its own notebook, and writes down any questions for you. Lessons from a study go into the study's file, and any change they'd make to the protocol reaches you as a proposal. The notebooks are yours to read. They can't rewrite their own instructions.
Built for one professor, meant to be handed to the next
University faculty
Both pillars, one team. The first professor is running a data-analytics course where students learn to work with AI agents, and a research group studying how people talk about those agents. The setup wizard is written for this case.
Departments, teaching centers, labs
Add writers, add reviewers, add coders. Every consolidated draft records where each part came from and who disagreed; every study records every decision and every number's source. That is a record of why a course — or a paper — looks the way it does.
Other educators, next
The system will be customized for educators beyond the university: home schoolers, K-12 teachers, community college instructors. Lighter teams on cheaper models, a rhythm that fits the week, the same binder and the same four words. You still decide.
Intuitionist Edu is running for one professor's course and research group in Fall 2026 while the team's memory, schedules, review process and research protocol get tuned in real use. The first study has run end to end and waits on its verdict. The setup wizard is up so you can see what setting it up involves; it isn't something you can sign up for yet.
Your course and your studies live on a server you control, with keys you hold and can revoke. No student names, submissions, or grades are sent to an AI model in this version. Research data you mark as staying in-house never leaves the team's own models. The binder is a private repository you own, with full history, so you can see and undo anything the team wrote.