aher.dev

October 2026 / Course Co-Pilot

3,405 Duke courses, out from behind the login

DukeHub lists 3,405 open courses for Spring 2027, in 6,560 sections across 220 departments. You need a Duke NetID to see any of them.

That's fine for registering. It's a problem for everyone else: a high school senior comparing Duke with three other schools, an admitted student picking a program, or a Duke student who has only ever searched their own department. They're all making decisions about a course list they can't read.

So I took a snapshot of every open course on 22 October 2026 and put it on one page anyone can browse and filter. It's unofficial, it leaves out instructor emails and class numbers, and it links back to DukeHub for anything that matters.

Open the explorer

The best courses hide

589 sections have their own topic, and 245of those sit under a generic title like "Special Topics". FINANCE 894 reads "Special Topics" in the list. Section 301 is Campbell Harvey teaching stablecoins and tokenization. PUBPOL 890 has 10 topics, from food policy to security in the Western Hemisphere with a former US Ambassador to Venezuela. Each topic is its own row in the explorer, because each one is a different course.

There's just as much in the half-credit house courses. There are 17 this spring, from NeetCode 75 to Pink Floyd and the Concept LP to an astrobiology course taught by Duke's Executive Vice Provost.

DukeHubNetID only2,855 entrieswritinglab workAI 3/5shownreview list1,758 rows
Figure 1. The snapshot becomes one row per course. Rules read each description, and anything they aren't sure about goes to review instead of the page: first a model that has to quote its evidence, then me.

Filters DukeHub doesn't have

The explorer can filter by what a course teaches. Sixteen skill tags (negotiation, writing, programming, lab work, fieldwork and more) come from rules that read every description. A tag only shows when the rules find it in the title, in the department, or more than once in the description. Anything weaker goes to review instead of onto the page.

The first reviewer is a model. OpenAI's GPT-6 Luna read the 1,812 weaker descriptions and could add a tag only by quoting the line that supports it, word for word. That pass cost about 8 cents. It added 816 tags and removed 232 leadership tags the rules had mostly guessed from the word "team". Every tag it added shows its quote on the course page, and the 1,758 rows that are still unsure stay on a review list for me.

That's the same design as Course Co-Pilot, the transfer-credit agent I scoped earlier: read the catalog, grade your own confidence, and hand the hard cases to review. This is the first piece of it running on real data.

How the AI-resilient score works

The filter I wanted most is the one I trust least, so it shows its working. "AI-resilient" scores each course from 0 to 5, one point each for physical, lab or field work; live work with people; judgment someone is accountable for; a practiced craft; and a small, in-person format.

It leans on Microsoft Research's 2025 study of where generative AI applies to real occupations. It found the least applicability in work with a physical component and hands-on work with people, and its authors are careful to say applicability isn't job loss. So this is a signal, not a promise, and every course page shows which checks passed and the word that triggered each one.

ARTSVIS 368SThe View CameraOperate a viewcamera ... developfilm in thedarkroom ... makecontact prints.seminar · 9 seatsPhysical, lab or field workLive work with peopleAccountable judgmentPracticed craftSmall, in person3 / 5
Figure 2. One real course, scored. Each matched phrase ticks a check and fills one segment of the meter.

Notable faculty works the same way. A professor gets the label only when I could find a source for the credential, and the source is linked. 63courses have one. Several claims I started with didn't survive that check, so they aren't there.

Taught by people who shaped their field

Every credential here links to its source.

Hidden under "Special Topics"

DukeHub shows a generic title. The section number is a different course.

Hard to automate

High AI-resilient scores: hands, people, judgment, craft, small rooms.

Half credit, very specific

House courses and electives that sound made up and aren't.

What it can't tell you

It only has classes that were open on 22 October, so anything already full is missing. Seats move every day. Special-topic sections often carry only the generic course description. And the tags come from words, not from sitting in the room.

If you're choosing a school, filter by the skill you want to leave with and see who teaches it. If you're already at Duke, try no prerequisites plus no Friday.

Open the explorer