An Autism Center Replaced Its Supervisor With AI. Here Is What AI Can't Take Over in RBT Supervision
Informational only; not clinical, legal or certification advice. Details come from the sources listed below as of Nov. 18, 2026. Check the current BACB RBT Handbook for your own requirements.
This week KrASIA published a story (adapted from 36Kr) worth reading. A director of an autism center in Pingdingshan, China, started using an AI system in January. Six months later, he said, he let the center's outside part-time supervisor go, because the system could take on some of that work at lower cost.
That center isn't working under BACB rules. But the question will show up here: if software can plan lessons and draft reports, do we need as much supervision? For RBTs, the answer is already written down.
What the story actually says
The details matter more than the headline. According to the article:
- The system was built by Dami & Xiaomi with Amazon Web Services and trained on the company's own therapy records. The company says it is used at nearly 100 of its centers and more than 300 partner institutions for assessments, scheduling and lesson planning.
- Therapists record responses on a tablet during an assessment, and the system produces a draft report for a supervisor to review and refine.
- The company gave reporters a comparison of 65 children at three outside institutions showing more progress in the AI-assisted group after three months. That is the company's own data, not an independent study.
- Staff review AI-generated material before using it. In one case, the system judged that a child had mastered a task and could move on. Her therapist thought she needed more practice.
- A senior supervisor found that many staff at smaller centers had never recorded every exchange, which left the system working from an incomplete picture of the child.
So even inside the story, a person caught the mastery call the AI got wrong, and the AI was only as good as the data people collected.
What BACB rules require for RBT supervision
The BACB's RBT Handbook (updated October 2026) says RBTs must practice under the "close, ongoing supervision" of a qualified RBT Supervisor or RBT Requirements Coordinator who is responsible for the work they perform. In practical terms, the handbook says:
- Supervision must be at least 5% of the hours an RBT spends providing behavior-analytic services each calendar month.
- It must include at least two face-to-face, real-time contacts per month. Phone calls and email don't count.
- The supervisor must observe the RBT providing services in at least one of those contacts. Observation by video conference is allowed.
- At least one contact must be individual. The other can be a small group of 2 to 10 RBTs.
- Video monitoring on its own, without real-time interaction or feedback, may not count as supervision.
The handbook's list of supervision activities includes observing, behavioral skills training, performance feedback, modeling professional and ethical behavior, building problem-solving skills, reviewing written materials like progress notes and data sheets, and evaluating the effects of services. An AI tool can help with one item on that list: reviewing paperwork. It can't watch an RBT run a program and coach them through it in real time.
Where AI can actually help supervisors
I want AI to give BCBAs more time with clients and staff, not less. Used that way, it can help:
- Prep before you observe. Have your platform pull trends, missed targets and note gaps so you walk in knowing what to watch for.
- Draft the paperwork after. Let it turn your observation notes into a feedback summary that you edit and sign.
- Track the 5%. The BACB's supervision checklist asks supervisors to have a system for tracking each RBT's monthly service hours against supervision hours, and for keeping supervision records accessible for up to 7 years in case of an audit. Software is good at that kind of counting.
- Flag data quality. If the tool notices missing trials or copy-paste notes, that's a coaching topic for your next contact.
Questions to ask before you cut supervision time
- Is this tool replacing paperwork time or observation time? Only the first is a real saving.
- Who signs off on mastery and program changes? It should be a qualified person who has seen the client.
- Are your RBTs recording data completely enough for any AI reading of it to mean something?
The bottom line
Supervision is how we protect quality and outcomes, not a cost line to optimize away. In the story, it was a therapist sitting across from a child who said "not yet." Let the software handle the counting and the drafting. Keep people doing the watching, the coaching and the clinical decisions.
Sources
- https://kr-asia.com/an-autism-center-in-china-let-its-supervisor-go-after-adopting-ai
- https://www.bacb.com/wp-content/uploads/2026/10/RBTHandbook_261028-a.pdf
- https://www.bacb.com/wp-content/uploads/2025/09/RBT-Supervisor-RBT-Requirements-Coordinator-Supervision-Checklist-260617-a.pdf
- https://www.bacb.com/ongoing-supervision-assessment-and-training/
More from the blog
- SOC 2, BAAs and Badges: How to Read an AI Vendor's Security Claims Before Your ABA Clinic Signs
- AI Translation for ABA Families: What Still Needs a Qualified Human Translator
- AI Trend Readings on ABA Graphs: What the Math Sees and What Visual Analysis Still Catches
- AI Claim Scrubbing for ABA Billing: What It Catches and What It Can't Fix
Informational only; not legal, clinical or compliance advice. Vendor HIPAA/BAA statements are as published by the vendors and are not verified by us. Sponsorship never changes our rankings. Corrections: admin@gtbehaviorbolts.com.