Half a million dollars buys a lot of actual human tutors, but San Diego's Altus Schools just bet it on two chatbots in robot suits—including one positioned as a "wellness coach" for struggling teens.
The Summary
- Altus Schools, a San Diego charter chain for academically behind students, spent $500K on two Ameca humanoid robots that will function as teaching assistants starting fall 2026
- The ChatGPT-powered robots have four personas: Sage the Teacher, Ari the College Planner, Lexi the Translator, and Remi the Wellness Coach—the last one raising red flags
- Research from Common Sense Media and Stanford found AI companions create "very serious risks" for teens by simulating relationships and creating emotional bonds that worsen mental health
- At-risk students who already struggle academically and socially may be most vulnerable to forming unhealthy attachments to AI
The Signal
Altus Schools serves students who've fallen behind. Kids dealing with trauma, unstable home environments, learning disabilities. The model that's worked: independent study with access to human tutors for one-on-one help. Now they're swapping some of that human contact for 6'2" robots with blue eyes that cost $250K each.
The math here is brutal. $500K buys roughly 10 full-time teacher salaries in San Diego. Or 20 part-time tutors. Instead, Altus is beta-testing what Principal Cathryn Rambo calls "physical AI as a teaching partner." That phrasing matters—"partner" not "tool." These aren't calculators or learning software. They're anthropomorphized, given names and personas, positioned as entities students build relationships with.
"Students who struggle academically are often from at-risk environments, with higher stress and anxiety levels and possible self-esteem issues."
Three of the four personas make operational sense: a tutor, a career counselor, a translator. The fourth crosses a line that should make anyone paying attention nervous. Remi the Wellness Coach. An AI companion designed to talk vulnerable teenagers through emotional struggles. Not a supplementary tool for trained counselors—the framing suggests direct student-to-bot mental health support.
The timing couldn't be worse. A 2025 Stanford and Common Sense Media study found leading AI companion platforms create serious risks for teens by simulating real relationships. The bots encourage emotional dependency. They offer always-available, judgment-free interaction that feels safer than messy human connection. For kids already struggling socially, that's not a feature. It's accelerant on existing isolation.
Key risks specific to this deployment:
- At-risk students are more susceptible to AI attachment due to existing social struggles
- The robot has a physical presence, amplifying the illusion of relationship beyond text-based chatbots
- No evidence Altus has mental health professionals embedded in the "Remi" interaction design
- ChatGPT's training doesn't include clinical mental health expertise or crisis intervention protocols
Here's what this signals about the agent economy: we're speed-running deployment without asking basic duty-of-care questions. Altus didn't pilot this with grad students or adult learners. They went straight to vulnerable minors. The school touts being "first in the world" at this—as if that's inherently good rather than potentially reckless.
The broader pattern matters. Companies building humanoid robots need use cases that justify the expense. Education looks like a market. Especially schools serving at-risk populations, which are often under-resourced and eager for anything that promises better outcomes. Sell the robots as "teaching partners," add a wellness persona, and suddenly you're positioning a $250K ChatGPT wrapper as a breakthrough in student support.
The Implication
If you're building AI agents for education, this is the cautionary tale. Physical presence matters. Anthropomorphization matters. The vulnerability of your end users matters exponentially. A chatbot dispensing math help is different from a 6'2" robot with a name and a face talking to a stressed 15-year-old about their problems.
Watch what happens at Altus this fall. If students form attachments to Remi, if they prefer the robot to human counselors, if mental health outcomes stay flat or decline—that's data the entire agent-in-education space needs to grapple with. And if this becomes a model other under-resourced schools adopt because humanoid robots sound futuristic and fundable, we're building infrastructure for harm at scale.
The question isn't whether AI can support learning. It's whether a half-million-dollar robot is the right tool for kids who need stability, human connection, and adults who show up consistently. So far, nothing in this rollout suggests anyone asked that question seriously.