# First AI pilot in education Fill in one card before your institution starts its first AI workflow. Fill in what you know; every blank line gets an owner and a date. Any blank line means: not ready to start. | Question | Answer | Owner / by when | | --- | --- | --- | | Which adult-side workflow is the first pilot: back-office paperwork, teacher planning and grading, family communication, advising, admissions or research? | | | | The workflow itself: who does it, how often, and how deep is it piled? | | | | What does a wrong draft cost, and who catches it? | | | | Does it run in a walled garden the institution owns, where no model trains on our data? (yes / no) | ☐ Yes ☐ No | | | Who uses it: teachers and staff only, or students too? Student access only through walled-garden tools? | | | | What sanctioned studio may staff use today, so nothing goes into public tools? | | | | Boundary test: does any output enter a grade, a student's record, a decision about a student, or a conversation with a child or family? (yes / no) | ☐ Yes ☐ No | | | If yes: the named owner, and the bar the output must clear first | | | | Are our use cases sorted Explore Now, Explore Later and Explore Never, with anything that would replace the teacher in Never? (yes / no) | ☐ Yes ☐ No | | | Which routine questions go to AI, and who keeps the worried calls? | | | | What already runs that nobody calls AI (adaptive practice, plagiarism detection, early warning, scheduling), and is it counted before anything new is bought? | | | | Is every claim of classroom-ready AI tested on our own hardest unit or messiest handbook, and what came back? | | | | Which number is named before the pilot (attendance minutes, prep hours, subscription dollars), and who measures it now? | | | | What bar is set in advance, and when is the same number measured again? | | | | Where does our data live, and how long before the data house is in order? | | | | Decision and date | ☐ EXPLORE NOW ☐ EXPLORE LATER ☐ EXPLORE NEVER Date ______ Signed ______ | | ## Example (fictional): translated family newsletters in a mid-sized district Fictional example. | Question | Answer | Owner / by when | | --- | --- | --- | | Which adult-side workflow is the first pilot: back-office paperwork, teacher planning and grading, family communication, advising, admissions or research? | Family communication: the weekly classroom newsletter, translated into the district's three main home languages | n/a | | The workflow itself: who does it, how often, and how deep is it piled? | About 40 classroom teachers, one newsletter each week; about 45 minutes to write, then two days waiting for the district translator | n/a | | What does a wrong draft cost, and who catches it? | A wrong date or a clumsy sentence reaches every family; a bilingual aide can read every draft before it goes | n/a | | Does it run in a walled garden the institution owns, where no model trains on our data? (yes / no) | Yes: the district's own studio; the data stays ours and no model trains on it | n/a | | Who uses it: teachers and staff only, or students too? Student access only through walled-garden tools? | Teachers and front-office staff only; no student-facing use | n/a | | What sanctioned studio may staff use today, so nothing goes into public tools? | The district studio, opened to all staff this term | n/a | | Boundary test: does any output enter a grade, a student's record, a decision about a student, or a conversation with a child or family? (yes / no) | Yes: the translated newsletter goes home to families | n/a | | If yes: the named owner, and the bar the output must clear first | **Blank: not ready** | Director of communications, by 4 Dec | | Are our use cases sorted Explore Now, Explore Later and Explore Never, with anything that would replace the teacher in Never? (yes / no) | Yes: sorted by the leadership team in October; newsletter translation in Explore Now, anything that replaces a teacher in Explore Never | n/a | | Which routine questions go to AI, and who keeps the worried calls? | Drafts and due dates to the assistant; calls about a child's progress stay with the teacher | n/a | | What already runs that nobody calls AI (adaptive practice, plagiarism detection, early warning, scheduling), and is it counted before anything new is bought? | Adaptive-practice licenses in 6 schools and plagiarism detection in both high schools, counted this term; nothing new bought | n/a | | Is every claim of classroom-ready AI tested on our own hardest unit or messiest handbook, and what came back? | A parent-communication assistant ran on our messiest newsletter, the one with three events and two date changes; it kept both dates, and talks continue | n/a | | Which number is named before the pilot (attendance minutes, prep hours, subscription dollars), and who measures it now? | Teacher minutes per newsletter: about 45 today, target 15, measured over four weeks by the two lead principals | n/a | | What bar is set in advance, and when is the same number measured again? | Bar: at most 1 wrong date or mistranslation in 20 spot-checked newsletters, checked monthly. First spot-check: 3 in 20 → **Fails the bar** | Director of communications, re-test by 26 Jan | | Where does our data live, and how long before the data house is in order? | Contact records and language preferences about 80% complete in the student information system | Data lead, by 30 Jan | | Decision and date | ☐ EXPLORE NOW ☒ EXPLORE LATER ☐ EXPLORE NEVER · Re-test 26 Jan · Date 20 Nov · "No named owner yet for family-facing output, and the first spot-check failed the bar." | Superintendent | From the course AI in education, Dr. Tony Hoang · v1 · Oct 2026