The AI coach that never writes the paper. It asks the next question.
Kai Coach is a Socratic coach built on your course’s own readings. When a student is stuck, it may supply a term, but never the reasoning. In its first course pilot, papers averaged twice as many developed arguments.
Built and tested at UCLA. Kudu builds every module at no cost to your institution.
KKai CoachBusiness Ethics · price controls
Kai Coach…if a buyer values a unit more than it costs a seller to provide it, both could have gained from that trade. When the ceiling prevents it, what shared benefit is left unrealized?
StudentI don’t know
Kai CoachNo problem. The missing benefit is the gain both sides would have received from a trade worth more to the buyer than it cost the seller to make; the book calls that lost benefit “surplus.” In your own words, does a binding ceiling mean fewer beneficial trades happen, or more?
In this exchange, the coach:Supplied the termLeft the reasoning to the student
First course pilot · History 2C at UCLA, 2025 vs 2026
73%Stated a clear position, on the assigned book, up from 43%
87%Of final exams cited a specific reading, up from 7%
100%Of coach turns asked a question, across 389 sessions
Why it exists
General-purpose AI will write the essay on request. Kai Coach is designed never to.
Your students already have chatbots that will draft a thesis in seconds. Kai Coach gives them what those tools won’t: a coach that makes them do the thinking, inside your course.
General-purpose chatbotKai Coach
General-purpose chatbot:Writes the thesis, the paragraph or the whole essay when asked.
Kai Coach:Never writes a thesis, a sentence or a paragraph. It asks the next question.
General-purpose chatbot:Improvises about a subject it half knows.
Kai Coach:Grounded in the book and readings your course assigns, so it points to real passages.
General-purpose chatbot:Gives generic advice to a generic student.
Kai Coach:Designed around where your previous students stalled.
General-purpose chatbot:Happens in another tab, out of the instructor’s sight.
Kai Coach:Part of the course: you can read every session and download it with scores.
See it work
Watch it coach, one question at a time.
From a first reaction to a real thesis, in three questions
History 2C students were asked to interpret a line from Walter Benjamin: “There is no document of civilization which is not at the same time a document of barbarism.” This student doesn’t start with a draft. They start by admitting the sentence isn’t clear yet.
S
StudentI want to understand the quote before I write anything.
K
Kai CoachGood place to start. Benjamin writes that there is no document of civilization which is not at the same time a document of barbarism. Take the sentence a piece at a time. What do you think he means by a document of civilization? (note 1)
1
S
StudentI don’t know
K
Kai CoachNo problem. A document of civilization is anything a culture keeps and celebrates as proof of its own achievement: an archive, a cathedral, a university, a canon of great books. Pick one we have looked at in this course. Who had to do what, and to whom, for that thing to exist? (note 2)
2
S
Student · first draftBenjamin’s statement shows that many of the achievements celebrated in history were built through inequality and suffering. Cities and cultural institutions represent civilization and progress, but they also show the labor and violence that made them possible. His idea forces us to look underneath the surface of historical events.
K
Kai CoachYou are doing strong application work, but the paragraph is still closer to a reaction than to the claim Benjamin is actually making. You treat a monument of civilization as an achievement or an institution without saying what makes his assertion stronger and more arguable than “sometimes progress involves harm.” What exactly is he claiming is always true — and what would a reasonable person say if they wanted to disagree with him? (note 3)
3
S
Student · revisedBenjamin argues that no document of civilization is not also a document of barbarism, which means that every achievement celebrated as progress rests on suffering or oppression usually hidden from view. Some would argue that civilization can exist without causing harm. Benjamin’s claim is that the exploitation is always part of its history, even when only historical investigation reveals it. (note 4)
4
1
It starts where the student is. The student wants to understand the quote first. The coach breaks the sentence into pieces and asks about one.
2
At the stall: a definition, nothing more. It defines the term, then hands the work back: pick something from this course and say who paid for it.
3
It names the gap, not the fix. The first draft is fluent and impossible to disagree with. The coach asks what Benjamin claims is always true, and what an opponent would say.
4
The student writes the thesis. “Shows” becomes “argues,” “always” makes the claim universal, and an objection is raised and answered.
The coach named none of those three changes. The student made all of them.
The one rule
When a student stalls, Kai Coach may supply the term. It never supplies the reasoning.
Kai Coach may
Name a concept or a term
Give a definition
Point to a section of the reading
Say what the student got right
Kai Coach never
Write a thesis, a sentence or a paragraph
Make the inference for the student
Substitute, rearrange or produce a numerical result
End a turn without a question
In the History pilot, across 389 instrumented discussion and peer-response sessions, every coach turn contained a question.
What the first pilot found
Same course. Same instructor. Same prompts. One year with Kai Coach.
History 2C at UCLA, a six-week survey of medieval and early modern European religious history taught by Prof. Teofilo F. Ruiz, ran in 2025 without Kai Coach and in 2026 with it. Every written component of both years was analyzed. Grades weren’t used, because grading wasn’t blind to the coach.
Two cohorts76 students in 2025, 56 in 2026
Same course designInstructor, prompts word for word, six-week structure
980 pieces of writing745 discussion posts and 235 essays
No grades usedMeasured from the submitted text
Share of student work, 2025 vs 2026
Each row: the 2025 cohort (no coach) and the 2026 cohort (Kai Coach)
2025 · no coach
2026 · Kai Coach
0%25%50%75%100%
Stated a clear position on the assigned bookpapers43% → 73%43% in 2025, 73% in 2026
Cited a specific course-reader sectionfinal exams7% → 87%7% in 2025, 87% in 2026
Traced a mechanism, by Week 5discussion posts7% → 51%7% in 2025, 51% in 2026
Four or more developed argumentspapers5.0% → 14.7%5.0% in 2025, 14.7% in 2026
No developed argument anywherepapers · lower is better34.7% → 15.8%34.7% in 2025, 15.8% in 2026
See every measure as a table
Measured from submitted text. A developed argument paragraph makes a claim, supplies evidence and interprets that evidence.
Measure
2025
2026
Stated a clear position on the assigned book
43%
73%
Developed argument paragraphs per paper
1.0
2.0
Papers with no developed argument anywhere
34.7%
15.8%
Papers with four or more developed arguments
5.0%
14.7%
Mechanism claims per 1,000 words, final exam
3.28
4.99
Final exams citing a specific course-reader section
7%
87%
Machine-writing markers per 1,000 words
0.46
0.18
Discussion posts tracing a mechanism, by Week 5
7%
51%
Developed argument paragraphs per paper
1.0→ in 2025, 2.0 in 2026
Mechanism claims per 1,000 words, final exam
3.28→ in 2025, 4.99 in 2026
Held up under multiple-comparison correction
The gain reached both ends of the class.
Papers with no developed argument fell by more than half. Papers with four or more roughly tripled. The spread barely changed: the whole class shifted, rather than the strongest students pulling away.
5 of 5
identical final-exam questions favored the coached year, across 143 student answers.
The same exam question, one year apart
Why did witch hunts rise as heresy trials declined? The opening lines of two answers:
2025 · No coach
History always has two sides: the main story filled with heroes and victors write what the future will read… Time and time again history repeats itself. Civilization is defined as an advancement of some sort for a group of people…
A truism, a cliché and a definition, before the question is touched.
2026 · Kai Coach
Heresy is defined less by doctrine than by who gets to speak for God; movements become ‘heretical’ when they bypass clerical control.
A causal mechanism, located in institutional authority, in the first sentence.
How to read these results
This is a comparison of two cohorts of one course, not a randomized trial, and the cohorts differ in year and size. Of the content measures, the rise in mechanism reasoning on the final exam is the one that held up after correcting for multiple comparisons. The other figures are observed differences between the two years.
Kai Coach was refined on humanities writing, where the work is a defensible argument rather than a correct final answer. The Physics pilot carries the same habit into STEM: explain the mechanism, and defend your own reasoning.
History
Run complete
History 2C, UCLA · Summer 2026
Discussion posts, Paper 1, peer response and the final exam: four modules across a six-week term, built from the previous year’s student writing.
Physics
Pilot · Fall 2026
Physics 1B, UCLA, with AIMS
Post-exam reflection, pre-exam calibration and weekly probes. Physics adds answers with a marking key, so reasoning and accuracy can be measured together.
Comparative Literature
Winter 2027
Comparative Literature, UCLA
Interpretation, close reading and sustained revision, attached to what the course’s writing requirement already asks for. Its questions come from that course’s own student writing.
Business Ethics
In preparation
Undergraduate pilot module
Coaches one argument about price controls through to the ethical judgment it forces: who goes without when a scarce good is rationed by queue instead of price.
Your course
Every module starts from a course’s own prompts and a previous cohort’s writing, and Kudu builds it at no cost to your institution. Tell us where your students stall.
Kudu builds it. You review it. About one to two hours of your time.
01
Show us where students stall
On a mapping call, we look at your prompts and a previous cohort’s writing to find where your students get stuck.
02
We build it on your readings
Kudu builds the module at no cost to your institution, from the book and readings you assign: a Kudu textbook, or the edition you already teach from.
03
You review it and release it
Preview it as a student, set the dates and release it. In all, it takes about one to two hours of your time.
You stay in control
In a Kudu course, Kai Coach is a module listed with your chapters. It also works in courses that don’t use Kudu.
Private instructions the coach follows in every session
Open and due dates for each session
Focus areas students choose from
Credit for taking part, or for reaching your goal
Preview as a student before you release it
Every conversation, downloadable with scores
For academic leaders
Integrity by design. Evidence of learning.
A response to generative AI that strengthens student writing instead of policing it, measured on real coursework, and built by Kudu at no cost to your institution.
01
Integrity by design
Kai Coach is built not to do students’ work. In the History pilot, every coach turn across 389 sessions asked a question.
02
Evidence you can show
A two-cohort comparison on real coursework, measured from submitted writing rather than grades. The full pilot report is available on request.
03
Gains across the whole class
Papers with no developed argument fell by more than half, while papers with four or more roughly tripled.
04
The same privacy as KAI
Student messages are anonymized before they reach a model provider, providers work under contractual no-training guarantees, and student data is never sold or shared. How KAI handles data
Chief AI Architect, Kudu Humanities · UCLA Humanities
Every Kai Coach module is designed against the ways a previous cohort’s writing went wrong, not a general theory of what students get wrong. That is why a new module begins by reading the course’s own student writing.
Selected as a 2025 ASU+GSV AI Innovator.
Questions
What faculty and deans ask first
How is Kai Coach different from KAI?
KAI is the course-aware AI in every Kudu course: students ask, and it guides them. Kai Coach is a structured module an instructor adds for a specific piece of work, such as discussion posts, a paper, peer response or exam preparation. It has its own sessions, dates and credit, and a stricter rule: every turn is a question.
Will it write an essay if a student insists?
No. It is designed never to write a thesis, a sentence or a paragraph. When a student is stuck it may supply a term or a definition, or point to a section of the reading. Then it asks the next question.
What does it cost, and how much of my time does it take?
Nothing for your institution. Kudu builds every module, and faculty spend about one to two hours reviewing and releasing it.
Do I need a Kudu textbook, or Kudu at all?
No. Kai Coach is built on whatever your course assigns: a Kudu textbook where you use one, or the trade or university-press edition you already teach from. It also works as a standalone module in courses that don’t use Kudu.
Can I see what my students and the coach said?
Yes. You can read your students’ sessions with the coach and download them with scores, so the coaching is part of the course rather than something happening beside it.
How is student data handled?
The same way as in KAI: student messages are anonymized before they reach a model provider, providers work under contractual no-training guarantees, and we never sell or share student data.
Start with one course
Bring Kai Coach to your course, or your campus.
Kudu builds the module at no cost to your institution. Your part is a mapping call and about one to two hours of review.
Course transcripts are paraphrased. The Physics exchange is a design target, not a recorded session.
Results: History 2C, UCLA, 2025 (76 students, no coach) and 2026 (56 students, Kai Coach), measured from submitted writing. See How to read these results.
Kai Coach was designed and built by Elizabeth Landers, Chief AI Architect, Kudu Humanities.