Intuitionist Edu is a team of four AI assistants that works for an instructor, under your stewardship. You set the brief; they research, draft, and review each other; you get one consolidated proposal with the disagreements preserved — and nothing becomes part of your course until you say so.
This is not an autonomous system that runs your course. It is staff. The assistants read your course facts and your past decisions before they speak, they propose rather than decide, and every proposal comes back to you with its sources and its disagreements — because you are the one accountable to your students.
Every draft, review, and consolidated version is a proposal. It becomes part of your course only when you type approve. Reject it, ask for a revision, or park it — the team adjusts.
Course facts, decisions, drafts, reviews, and what the team has learned about you live in plain files in a repository you own. Open them, edit them, roll them back. No black box.
Four different AI models catch different things. The manager is required to show you where they disagreed and why it chose what it chose — so the judgment call stays yours.
The assistants are named by the first letter of the AI model each runs on. Their roles are fixed and their authority is limited; yours isn't.
You set the brief, answer the questions that matter, and give the verdicts. Everything below exists to put better-researched, better-argued proposals in front of you faster — and to remember what you decided so you don't have to say it twice.
Takes your brief, splits it, hands it out, chases deadlines, and writes the one consolidated proposal you read — with a section on where the team disagreed. Presents to you; never decides for you.
Researches and drafts proposals — syllabi, objectives, assignments, rubrics — to your brief. Works alone first, then reviews Agent K's version.
Same brief as C, a different mind. Strong on retrieval: readings, datasets students can actually get, and the boring-but-reliable choice. Every claim carries a source you can check.
Didn't write it, so it can see it. Reviews both drafts against a rubric you can read — alignment, workload, evidence, whether students must verify the AI's work — then checks the final before it reaches you.
One line from you starts the work; one word from you ends it. Everything in between happens in a Discord thread you can watch, redirect, or stop at any point.
"@Agent G, draft weeks 1–3 of the syllabus." One line. The manager may ask you one question if it changes the work.
C and K each research and write their own version to your brief, without seeing the other's.
C reviews K, K reviews C, D reviews both: keep, change (ranked), open questions.
G merges the stronger parts, records where each section came from, and lists where the team disagreed. D checks it before it reaches you.
approve · reject · revise · park. Only approve moves anything into your course. Your verdict is logged, and every agent reflects on what you kept.
Only your account can talk to the team, and the assistants hear only messages that name them — so four bots never talk over you or each other. Tasks run in threads you can read; every file they produce lands in your binder.
Between tasks, the assistants do homework on a schedule you set and reflect on what you decided. What they learn about you stays a proposal until you confirm it.
Every verdict you give is written down as evidence. Patterns become hunches; a hunch becomes a confirmed preference only when you approve it in the Friday digest. One offhand comment never becomes a rule.
On days you set, they look for new readings, datasets, and teaching research for the next two weeks of your course — and file it in the binder for you to use or ignore. If nothing is new, they say nothing.
Each night, each assistant rereads what you approved and rejected, updates its own notebook, and writes down questions for you. The notebooks are yours to read; they can't rewrite their own instructions.
Verdicts are typed, not clicked, so they can be remembered and cited. High-stakes areas — grading, accessibility, integrity — always wait for your explicit yes. Assistants can't change their own instructions or promote their own guesses.
approvereject: too longrevise: fewer readingsparkEverything that belongs to you sits in one part of the binder; everything reusable sits in another. Handing the kit to another educator means loading their course facts and letting the team interview them — the new instructor is in charge from the first message.
Course redesigns, new preps, and the weekly grind of readings, assignments, and rubrics. The first course is a data-analytics class where students learn to work with AI agents.
A lighter edition: a manager and one writer, a cheaper model, a weekly rhythm. Same binder, same four words, same "you decide."
Add writers, add reviewers. Every consolidated draft carries its provenance and its dissent — a record of why a course looks the way it does.
The setup wizard is published so educators can see what setting it up involves — it is not yet a service you can sign up for. If you'd like to be an early instructor, write to us.
The team runs on infrastructure you control, with keys you hold and can revoke. No student names, submissions, or grades are ever sent to an AI model in this version. The binder is a private repository you own — with full history, so you can see and undo anything the team wrote.