A Chatbot Named Ada, and Two Years of Staying Power

TL;DR: A custom calculus tutor that never sleeps, never judges, and never does the graded work, because none of it is graded.

What she’s doing: Abby teaches a calculus course designed specifically for life sciences majors, a population she describes as often anxious about math and reluctant to seek help in person. Roughly two years ago, the College of Engineering offered to build faculty custom course chatbots, and she took them up on it. Crucially, she was not handed a black box. A graduate student taught her to maintain and update the tool herself. The bot, which she has named Ada (after the great mathematician Ada Lovelace), holds her full syllabus, her section-by-section textbook map, and years of her old exams. A student can ask for help with a specific section or request a practice quiz, and Ada knows the upcoming topic and generates one in Abby’s style. Its instructions tell it to scaffold rather than simply answer, and to model her exams down to length and question mix.

Because 100 percent of the course grade comes from in-class assessments, Abby has set aside integrity concerns entirely. Homework is ungraded, so she is comfortable with Ada working problems fully. Her guiding message to students is that using the tool only to extract answers defeats its purpose, since the homework itself is the preparation for the exams they will take without it.

What’s working and what isn’t: Between half and two-thirds of students try Ada, and those who do report positive experiences. Abby credits the narrow scope, calculus alone, for keeping hallucinations low. She also points to tone. She asked the engineering team for something friendly and almost goofy, and students respond to a tutor that occasionally jokes. Most striking, the students most hesitant to attend office hours seem to benefit most, freed to ask the same question repeatedly without embarrassment, and several later begin coming to office hours more. Her one worry, which she is careful to frame as caution rather than observed harm, is over-reliance: that students may reach for Ada before struggling long enough to build durable understanding. So far she has seen no evidence of this

What’s next: Abby is uncertain, and comfortable saying so. She would like AI support in another course, an interactive class built around group work, but she is reluctant to introduce a tool that might pull students toward a screen and away from the peer relationships the course is meant to cultivate. She would rather a stuck student text a classmate than open a chatbot. She is still working out whether, and where, a balance can be struck.

Have you tried something new with AI? How did it go? Send an email to [email protected] to let us know!