Table of Contents
Subject AI Personalized School: Eastern High's Experiment
A Subject AI personalized school opened in Bay City with 106 students setting their own pace. What the personalization evidence says and what to watch for.
An alternative high school with 106 students rebuilt itself around an AI platform this year, and the most interesting part of the plan is not the AI. Eastern High School in Bay City, Michigan adopted the Subject AI learning platform for the 2026–27 school year, letting students customize their own path and pace. It also added career hubs, an academic success center, a regulation room for mental health, a full-time counselor, free meals, and daily transportation. A Subject AI personalized school is a real experiment, and the evidence says the second list is doing as much work as the first.
Key Takeaways
- Bay City Public Schools’ Eastern High School implemented the Subject AI platform starting the 2026–27 school year for its 106 students, replacing Edgenuity.
- The redesign pairs AI-paced online coursework with career training, face-to-face specialized courses via the Bay-Arenac Intermediate School District, an academic success center, enrichment hubs, a regulation room, full-time counseling, free meals, and transportation.
- Superintendent Grant Hegenauer framed the AI as “a powerful instructional co-pilot to enhance the learning experience, not dictate it,” adding that “our human teachers are, and always will be, the heart and soul of Eastern High School.”
- RAND’s 2017 evaluation of personalized learning across 40 schools found effects of about 0.09 in math and 0.07 in reading, roughly three percentile points, with only math statistically significant.
- The AI tutoring result that worked best was human-supervised; the one that failed at scale was a chatbot students rarely opened.
What a Subject AI personalized school looks like in practice
Eastern High School is an alternative high school in Bay City, Michigan serving 106 students. GovTech reported on June 5, 2026 that it would implement the Subject AI learning platform at the start of the 2026–27 school year, giving students the ability to customize their educational path and pace.
The platform side: Subject adapts reading levels, provides instant feedback, and targets specific skill needs, replacing the school’s previous Edgenuity system. Subject describes itself as a K–12 digital curriculum company that pairs state-certified teachers with online courses, accredited through bodies including Cognia and WASC, with NCAA and College Board approval, and reports “1k+ schools across 38 states” and “200k+ students active in 2026.”
The human side is longer than the software side, which is the detail worth noticing. The redesign includes career training and face-to-face specialized courses through a partnership with the Bay-Arenac Intermediate School District, community service project opportunities, an academic success center, enrichment hubs, a coffee lounge, a regulation room for mental-health and social-emotional support, full-time counselor access, multi-tiered support personnel, free breakfast and lunch, and daily transportation.
Superintendent Grant Hegenauer was direct about the division of labor: “Our human teachers are, and always will be, the heart and soul of Eastern High School,” with the AI serving as “a powerful instructional co-pilot to enhance the learning experience, not dictate it.” His stated goal: “This model is intentionally designed to ignite curiosity, maximize daily engagement, and ensure no student is left behind.”
For an alternative high school, that framing matters more than it would elsewhere. These are often students for whom the standard schedule already failed, and for whom attendance and engagement are the binding constraints.
Model by model: what the evidence says about personalization
| Model | Best evidence | Effect | What it suggests |
|---|---|---|---|
| Personalized learning schools (broad) | RAND, Informing Progress, 2017, 40 schools for the Gates Foundation | ~0.09 math, ~0.07 reading; about 3 percentile points; only math significant | Modest and real; hard to implement well |
| Purpose-built AI tutor, college physics | Kestin, Miller et al., Scientific Reports 2025, 194 students | Median learning gains more than double an active-learning class, in less time | Design and subject specificity matter enormously |
| Human-supervised AI tutoring | Eedi trial, 165 UK students ages 13–15, reported Dec 2025 | 66.2% vs 60.7% on new problem types; 0.1% hallucination rate (5 of 3,617 messages) | Supervision of the AI improves outcomes |
| Unsupervised AI chatbot beside content | Oreopoulos et al., 18 schools, Tennessee district | Used on ~1/3 of platform days; “access was nearly universal but engagement was thin” | Availability is not usage |
| Teacher-facing AI (planning) | AmplifyLearn.AI, 12 Washington districts | ~0.025 SD reading per 3 hours; math indistinguishable from zero | Small upstream effect |
| Fully AI-paced school | No published evaluation | Unknown | Watch attendance and completion first |
Read that table in order and a pattern emerges. The interventions with real effects had a human in a specific role: reviewing the AI’s output, or teaching alongside it. The intervention that failed had the AI sitting available and unused. Eastern High School’s design, with a full-time counselor and face-to-face specialized courses, is closer to the first pattern than the second.
RAND’s number deserves a moment. Informing Progress, published in 2017, compared 40 schools (31 charters, 9 traditional) in the Gates Foundation’s Breakthrough School Models program against matched peers. Treatment effects were about 0.09 in mathematics and 0.07 in reading, roughly three percentile points, with only the math estimate statistically significant. Teachers and students reported more personalization than national samples on some dimensions, but the harder-to-implement pieces, including student-teacher conversations about progress and goals, “did not appear to differ from practices in schools nationally.”
That last finding is the one to carry into any conversation about a personalized school: the software part is easy to install and the conversation part is hard, and the conversation part is where the gains hide.
What this means for a parent considering a school like this
Alternative and credit-recovery schools serve a real need, and self-paced models can genuinely help a student who is behind, working, parenting, or dealing with health issues. The pace flexibility is not a gimmick for that population.
The honest risks are three. Self-paced can become self-abandoned without structure, which is why Eastern’s attendance supports (free meals, daily transportation, a coffee lounge that gives students a reason to be in the building) are substantive rather than decorative. AI-paced can drift from grade-level expectations if nobody audits it, which is why accreditation and a teacher of record matter. And engagement is the documented failure mode: the Tennessee study found “access was nearly universal but engagement was thin,” which is what a self-paced model has to beat.
For broader context, PISA 2025, released September 8, 2026, found about 20% of 15-year-olds across the OECD are low performers in all three core domains, up from 16% in 2022. There is a real population that the standard model is not reaching. Experiments aimed at them deserve attention rather than reflexive suspicion, and also measurement.
What to do at home
Ask for completion and attendance data, not enthusiasm
The two numbers that matter in a self-paced model are what share of students complete courses and what share attend. Subject reports a “94% completion rate” across its network. Ask your specific school for its own figure, before and after the change.
Ask who the teacher of record is
Accreditation requires a certified teacher responsible for the course. Ask who that person is for each of your child’s classes, how often they interact, and how a student reaches them when stuck. Our checklist for reading your district’s AI policy covers the related governance questions.
Ask how pace is monitored
Self-paced means someone has to notice when a student stops. Ask what triggers an intervention: days without progress, a missed milestone, a failed assessment. If the answer is “we check in periodically,” ask for the specific threshold.
Look at the wraparound supports as part of the academic plan
Free breakfast, transportation, and a counselor are not extras in a model like this; they are the attendance strategy. A school offering AI-paced coursework without them is running half the experiment. Our piece on AI tutors versus human tutors covers what each part contributes.
What not to do
Do not assume “personalized” means “better”; RAND’s estimate was about three percentile points and only significant in math. Do not assume it means “worse” either; for a student the standard schedule already failed, flexibility can be the difference between a diploma and no diploma. And do not evaluate it on the technology, because the evidence says the human supports are doing the heavy lifting.
What to Watch For Over the Next 3 Months
- Week 4: You know who the teacher of record is, how pace is monitored, and what the school’s own completion rate was last year.
- Month 2 red flags: Your child is “ahead” in a course but cannot explain the material; no adult has contacted you about progress; the regulation room or counselor exists on paper but your child has never been in it.
- Month 3 self-check: Ask your child to teach you one thing from each course, without notes. In a self-paced model, that is the only reliable audit a parent can run.
Frequently Asked Questions
What is Subject AI?
Subject is a K–12 digital curriculum platform that pairs online courses with state-certified teachers, accredited through bodies including Cognia and WASC with NCAA and College Board approval. It reports more than 1,000 schools across 38 states and over 200,000 students active in 2026. Eastern High School adopted it for 2026–27, replacing Edgenuity.
Does a fully personalized school work better?
No published evaluation of this specific model exists. The closest comparison, RAND’s 2017 study of 40 personalized learning schools, found about 0.09 in math and 0.07 in reading, roughly three percentile points, with only math statistically significant. The effects were modest and implementation was hard.
Is my kid’s diploma still valid?
Accreditation is the thing to check, not the technology. Subject lists accreditation through Cognia and WASC plus NCAA and College Board approval, which is what colleges and athletic eligibility bodies look at. Ask your specific school to confirm its accreditation and who the teacher of record is.
What happens if my child falls behind at their own pace?
That is the central question for any self-paced model, and the answer should be a specific trigger and a named person. Eastern High School’s design includes full-time counseling and multi-tiered support personnel. Ask what threshold prompts an intervention and who makes the call.
Why does an alternative school need free meals and transportation?
Because in this model they are academic interventions. Attendance is the binding constraint for the population these schools serve, and the documented failure mode of AI learning tools is thin engagement. Removing the reasons a student cannot show up is how a self-paced model gets used at all.
About the author
Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.
Sources
- GovTech. (2026, June 5). “Bay City Alternative High School to Customize Education With AI.” https://www.govtech.com/education/k-12/bay-city-alternative-high-school-to-customize-education-with-ai
- Pane, J. F., Steiner, E. D., Baird, M. D., Hamilton, L. S., & Pane, J. D. (2017). Informing Progress: Insights on Personalized Learning Implementation and Effects. RAND Corporation. https://www.rand.org/pubs/research_reports/RR2042.html
- Kestin, G., Miller, K., et al. (2025). “AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting.” Scientific Reports, 15, 17458. https://www.nature.com/articles/s41598-025-97652-6
- Chalkbeat. (2026, August 25). “Students rarely engaged with Khan Academy’s AI-powered tutor Khanmigo, study finds.” https://www.chalkbeat.org/2026/08/25/ai-tutoring-students-khanmigo-khan-academy-engagement-study/
- The 74. (2025, December 3). “AI Tutors With a Little Human Help Offer Reliable Instruction, Study Finds.” https://www.the74million.org/article/ai-tutors-with-a-little-human-help-offer-reliable-instruction-study-finds/
- OECD. (2026, September 8). “PISA 2025: Students’ reading and mathematics performance declined sharply across the OECD.” https://www.oecd.org/en/about/news/press-releases/2026/09/pisa-2025-students-reading-and-mathematics-performance-declined-sharply-across-the-oecd.html
- Esbenshade, L., et al. (2026). Generative AI in K-12 Classrooms: A Midyear Implementation Report. AmplifyLearn.AI, University of Washington. https://arxiv.org/abs/2605.16277
- Subject. “K-12 Initial Credit Courses.” https://subject.com/
