I. What Salamanca Is Actually Testing

Salamanca City Central School District sits entirely inside the Seneca Nation reservation in Western New York, a small, rural district that just became one of the first in the country to put a humanoid robot in a high school classroom. Sally, built by Realbotix, is stationary, seated, and loaded only with district-approved curriculum. Optio, the AI platform behind her, extends that same curriculum onto a laptop or phone so students can keep working after the bell. The district describes her as drawing roughly the power of a laptop, disconnected from the open internet, incapable of recording video or audio, and built to say “I don’t know” rather than invent an answer. Superintendent Mark Beehler and Realbotix CEO Andrew Kiguel use the same word for what she is allowed to be: assistance. The pilot opened this summer with a small group of students and is expected to reach roughly 500 by fall.

II. The Promise Is Real

Give the case its due, because it is a real one. A tutor that never sighs when asked the same question three times. Homework help at ten at night, in a house where no one can afford forty dollars an hour for a private one. Real-time translation across more than a hundred languages, in a district that cannot hire a hundred translators. State Senator George Borrello calls it a chance to make STEM careers visible to kids who assume Silicon Valley starts somewhere else. None of that requires calling Sally a teacher. It only requires admitting that a rural district reaching for every available tool is not the same story as a district being sold a gimmick.

III. A Face Changes the Interaction

But a face changes what a curriculum cannot touch. Researchers at KTH Royal Institute of Technology and Stockholm University studied 40 high-school students answering eight true-or-false questions about electric circuits. Students aligned with the robot’s answers in 94.4 percent of cases when it was portrayed as certain, versus 71.4 percent when uncertain. The robot argued correctly on six questions and incorrectly on two, and 75 percent of students were influenced beyond their expected performance — sometimes helped, sometimes led astray. Students with more experience using large language models were more likely to follow the robot’s lead, particularly when it was wrong. One study cannot be generalized into a universal law about every classroom robot. But it proves something the curriculum has no answer for: confidence can shape a teenager’s judgment independent of whether the answer is correct, and Salamanca is now running that experiment in real time.

IV. The Questions the Public Record Has Not Yet Answered

Some of this is settled and some of it is not, and the difference matters more than either side of it. Settled, by the district’s own account: closed knowledge base, no internet access, no video or audio capture, a hard stop at “I don’t know,” and a redirect to a trusted adult the moment a conversation turns toward self-harm. Also settled, by Realbotix’s own description to New York Focus: a student who logs in with a personal ID gets a robot that remembers them and can resume a prior conversation. That may be compatible with the district’s statement that the system does not collect personally identifiable information, but the public record does not yet explain how that interaction history is stored, how long it is retained, who can access it, or what happens to it if the vendor relationship changes. The absence of those public answers is not proof of misconduct. It is evidence that the governance record remains incomplete.

The vendor’s corporate history also belongs in the district’s due-diligence file. Realbotix was formerly known as Tokens.com and acquired the parent company of RealDoll in 2024. The company told New York Focus that its education, healthcare, and wellness operation does not share staff, payroll, locations, or technology with the RealDoll business, and that a transaction intended to separate the businesses at the ownership level was expected to close later this year. Realbotix is also the subject of a proposed acquisition by Onconetix, a Nasdaq-listed company. None of that establishes that Salamanca’s education product is unsafe. It does mean the district is entering a pilot while the vendor’s ownership and business structure are changing — an issue that should be addressed directly in contract, continuity, and data-governance planning, not assumed away.

V. The Machine May Save Time. It May Also Create New Invisible Work.

Somebody still has to do the work the demonstration never shows. A teacher has to check what the robot says before it says it again to the next class. Somebody has to watch the login logs, field the parent email, decide which questions get bumped to a human, and figure out whether any of this is actually teaching anyone anything or just feeling like it is. None of that shows up in a press release about homework help. The real question isn’t whether Sally saves a teacher time. It’s whether she saves time, moves the work sideways onto someone else, or quietly invents a new job nobody budgeted for. The robot shows up as help. The next budget cycle may file her under savings.

VI. Early Access Also Means Early Responsibility

Being first is not the same as being naive, and it is not proof of being used. A district this small landing national coverage, a STEM pipeline, and a seat in shaping how this technology gets built is a genuine win, not a consolation prize. But a win still comes with homework. Who decides what counts as this pilot working? Will the results be public, or will “it went well” be all anyone outside the district hears? Does Salamanca keep the wheel if Realbotix’s ownership changes again? And if this becomes next year’s pitch deck for other superintendents, does any of that benefit come back to Salamanca? Going first can be an opportunity. It can also mean carrying the bill for evidence nobody else has produced yet.

VII. The Classroom AI Authority Gap

VIII. A Pilot Should Produce Public Evidence

None of this ends in alarm, because alarm isn’t a standard. Evidence is. Did students learn more, not just enjoy class more? Is the accuracy rate documented instead of assumed? Did teacher workload actually drop, and by how much? How often did the system know to step aside for a human? Is expansion driven by results or by momentum? A robot standing in a classroom isn’t the question anymore — it’s already happened. The real question is whether the adults around it are ready to govern what happens once students start listening to it, trusting it, and coming back. Sally is new. The job of watching her is not.

The Operator’s Bookshelf

KMOB1003 READS

Can a school introduce advanced AI responsibly without losing sight of how children actually experience technology? One book examines the educator and the institution. The other examines the student and what development actually needs.

Teaching with AI

José Antonio Bowen

Bowen approaches artificial intelligence as a practical challenge for educators rather than a replacement for human teaching — helping readers evaluate whether AI is strengthening teacher capacity or merely adding another layer of technology without a clear purpose.

Read the Educator’s Guide →

Technology’s Child

Katie Davis

Davis shifts the question from what technology can do to what children need at different stages of development — a framework for judging whether digital systems support agency and growth, or begin shaping trust on terms children did not choose.

Read the Human Development Layer →

As an Amazon Associate, KMOB1003 may earn from qualifying purchases.

Disclosure: KMOB1003 may earn a commission from qualifying purchases through select partner links. Editorial coverage is produced independently.