The Two Decisions Doing The Work
The MSAI program brochure went live in March 2025. It names healthcare, finance, and cybersecurity as the applied areas where AU graduates will use what they learn. It also leans on the three-year OPT extension as a reason international students should choose Atlantis.
Those two decisions are doing a lot of work. One tells faculty what to teach. One tells admissions what to promise. The space between them is where curriculum operations actually live.
Atlantis is data-driven. Same here. The note below is written in that posture.
The Frame
A useful way to think about an AI program is that roughly sixty percent of what ships on "AI" projects is still traditional code and databases. About thirty percent is rule-based logic. Only about ten percent is genuinely a model-in-the-loop problem.
A faculty workshop that teaches students to make that distinction, before they pick a tool, produces graduates who ship working systems rather than demos. The same frame applies to the back office. The document flow behind an I-20 or an OPT filing is mostly rule-based logic and records retrieval. A well-scoped orchestration layer can compress that work without touching the parts that need a human signature.
- 60% Traditional Code and database work. Most of any AI project.
- 30% Rule-based Deterministic logic. Predictable. Testable.
- 10% Model in loop Where a language or vision model is the right tool.
The Case
Flagler College brought in Eduba for a small-cohort faculty engagement. Teaching-focused institution, one week of concentrated workshops, artifacts that faculty took back into their courses the following term. The shape of that engagement fits Atlantis: one cohort of MSAI faculty, plus whoever runs international admissions if the scope opens up.
1,500+ people have been trained across Eduba engagements since May 2025 at companies like Pacific Life and Colgate-Palmolive through Correlation One. Ninety-five percent were still using the tools thirty days later. That adoption number is the thing that matters for a university that has to justify the investment to a board.
For leadership-level audiences, Eduba has also trained 40+ executives at KPMG UK (Big Four) in regulated-industry contexts. Different scale, same frame.
The Paper
Jake's Interpretable Context Methodology paper, submitted to ACM TiiS, gives your faculty a citable artifact and a reproducible framework for organizing agent context.
ICM is MIT-licensed and sits on GitHub, which means it can be assigned as reading in a course. The folder-as-architecture idea works as a lab exercise for students who need to see where structure earns its keep.
Interpretable Context Methodology
Folder structure as agent architecture. A layered filesystem (L0 identity through L4 working artifacts) with measurable interpretability and reproducibility gains.