Giving Teaching Back to Teachers

The course runs me

Last year, I caught myself saying something out loud that I had been feeling for years: the course runs me, I do not run the course. I teach in the MBA classroom at Kellogg. I love teaching and I am reasonably good at it. But I hate the “pajama time” that comes with teaching — the time spent outside the classroom, away from the students, developing lectures, designing assignments, doing grading.

Let me walk you through the process I go through in teaching an MBA elective, to give a better sense of the pain points. I teach AI-Native Marketing and AI-First Product Strategy, which means my course starts aging the day I finish building it. Frameworks that felt current in the spring feel dated by fall. The reading that anchored a session has been overtaken by events. So the summer goes to hunting fresh readings and cases, rebuilding lecture decks, and redesigning assignments and exams, partly because the field moved and partly because last year’s versions have already found their way into answer banks.

Then the quarter begins, and the running costs take over. Published studies of course workload put grading and feedback alone at roughly half of teaching time, and grading by hand is inconsistent: bias, fatigue, and drift across weeks mean two students can submit comparable work and receive different grades. Office hours become a queue in which I answer the same four questions eleven times, while the student with the truly interesting confusion never signs up. And at the end of it all, I have no systematic picture of which sessions landed and which did not. Like most professors, I am flying blind in terms of linking instruction to outcomes, for each student.

Every stage of that lifecycle takes more time than it should and produces worse results than it should. Designing the course, refreshing the content, building the assessments, giving feedback, meeting students, improving the next iteration — all of it is straining under the same shortage: my attention. And this is where AI, used judiciously, can transform pedagogy.

Band-aids and blind spots

AI has been widely embraced by instructors and institutions. But perhaps not in the way it should be. The instinctive institutional response has been to bolt AI onto whatever software the school already owns. But bolting on AI is a band-aid. A chatbot attached to an LMS can answer a logistics question. But it does not know your learning outcomes, it has not read your rubric, and it has no view of how this week’s discussion went. Consequently, it cannot help you design, assess, or improve the content and the pedagogy. The band-aid covers the wound without treating it.

The deeper blind spot is what our core systems were built to do in the first place. Canvas and the other learning management systems are exactly what their name says: management systems. They are very good at course administration — rosters, submissions, gradebooks, announcements, and deadlines. But course administration is the half that software has already solved. The LMS does not design a course, it stores one. It does not evaluate student thinking, it collects it. And it does not personalize learning, it broadcasts. Course design, evaluation, and personalization are where the pain points are, and where the opportunity to transform pedagogy lies.

Four audiences for AI in pedagogy

I have sat in enough campus technology decisions to know why most of them don’t work. The professor who champions a tool, and in this story that professor is me, cares about hours returned and control retained; nothing should reach my students without my approval. MBA students, who are paying a premium and know it, care about feedback that is fast and fair, and guidance that builds their thinking rather than handing them answers. The dean of curriculum and teaching cares about consistency across sections of the core, analytics credible enough for accreditation, and a governed answer to the shadow AI already running loose in every study group. And the IT and security team cares about native integration with the LMS the school already runs, compliance with FERPA and GDPR, and a guarantee that student work never trains anyone’s models. Any AI platform for pedagogical support must speak to the pain points and desired outcomes of all these stakeholders.

Grading and feedback alone consume half of an instructor’s teaching time.

Reimagining pedagogy with an AI-native operating system

The alternative to the band-aid is to treat the course itself as the system and run its full life, from the first design decision to the last grades turned in. The insight is to build an operating system in which AI does the heavy lifting at every stage and the instructor sets the standard and approves every consequential decision.

Now let’s walk through the lifecycle with such an AI-native pedagogy OS in place. Before the quarter, the system drafts the syllabus, suggests current readings and cases, updates the lecture notes and readings based on new developments, and generates assessments matched to my learning outcomes; I shape and approve, and course design becomes a conversation rather than a blank page in August. The course runs inside Canvas, so nothing changes for the registrar or for IT; the operating system works above the LMS rather than replacing it, doing the work the LMS was never built to do. When I give a midterm on Friday, the platform grades against my rubric within hours, steadily and consistently, and every grade waits for my approval before a student sees it. I review and sign off over coffee; my students have feedback while the material is still alive; the class-level diagnostic tells me which concept to reteach on Monday. All term long, every student has a Socratic tutor grounded in my course materials rather than the open internet, one that asks before it tells. That is personalization the broadcast model of the LMS can never deliver: the strong student gets stretched, the struggling student gets scaffolding, and the students who come to my office hours have interesting questions that go beyond the basics the tutor has helped them with. When the quarter ends, the evidence of what landed flows into the next iteration of the course.

Notice the compounding logic in this system. Faster grading alone saves hours, and hours matter. But when assessment evidence feeds the tutor, the tutor’s conversations inform course improvement, and improvement reshapes next term’s design, the course becomes a closed-loop learning system. That loop cannot be assembled from point tools, and it is what separates an operating system from a collection of features. Two convictions keep it AI-native rather than AI-flavored. The professor stays in command: my judgment is the product, not a bottleneck to be automated away, and nothing reaches my students unapproved. And the system cares about process, not just answers: consumer AI chases the answer, while a teacher cares how a student reasons, where they get stuck, and what they do next.

Getting started

For faculty and academic leaders who want to move, here is my advice, based on what I do myself:

• Start with the assessment bottleneck. Grading is where the hours are and where consistency gains show up first. It is the lowest-hanging fruit, and the easiest to judge against your own rubric.
• Keep approval human. Adopt tools where the professor signs off on everything consequential, and nothing is sent to students without a human in the loop. Students are deservedly skeptical about “AI grading,” so you must be clear that AI assists and enhances the graders. It does not replace the grader.
• Ground tutoring in the course, not the internet. A tutor that draws on your syllabus, readings, and standards reinforces your teaching. One that draws on the open web competes with it.
• Redesign assessment for the AI era, not against it. Shift weight toward work that shows reasoning: staged drafts, oral defenses, decision memos, simulations. Every assignment should be AI-resistant in a way that encourages AI use but does not let AI alone give the answer. Process is harder to fake and better preparation for the jobs your MBAs will hold.
• Pilot one course, then scale by evidence. A single MBA course in a single term generates the proof a school needs. Resist legislating campus-wide before you have run the experiment.

Built by a teacher, piloted in my own classroom

Pedagogix OS is our implementation of everything above: five modules — Design, Administer, Assess, Tutor, and Improve — covering the full life of a course with a person in the loop at every step. The software drafts; the professor approves. It is LMS-native, working inside Canvas and its peers rather than asking a school to replace them, FERPA and GDPR compliant with regional data residency, and student work never trains models. It has been piloted at the Kellogg School of Management, in the classrooms where I have taught MBAs and executives for more than three decades, and it is now ready for other business schools and universities.

We are equally clear about what we refuse to be: not a chatbot bolted onto legacy software, not a black box that grades in the dark, not a company that treats student data as a product. Adoption starts the way it should, with a free instructor tier, then a department pilot with white-glove onboarding, then an institutional license as the evidence accumulates. AI created this new workload. Software should be the thing that carries it, across the whole life of the course and not just the grading weekend, so that professors can do the work only professors can do. That is what giving teaching back to teachers means.