How Does Multiverse Use AI to Develop Its Coaches?

In its September 25, 2026 post, Multiverse says Compass reviews instructors’ coaching sessions against its Compass Framework and combines findings with learner feedback. It flags points for review and suggests development areas; Delivery Managers check outputs, discuss goals with instructors and decide what to do. The company says Compass is not a performance verdict and does not make employment decisions. The post describes staged internal use, not a commercial launch.

Turning session observations into coaching prompts

Compass is meant to surface moments for discussion across coaching sessions, not replace manager-instructor conversations. Multiverse’s May 19 coaching-model page offers older company context for its framework. The September post says Compass combines session observations with learner feedback, but describes no score that settles teaching quality.

That differs from learner-facing AI. Our coverage of OpenAI’s AI Academy learning paths for workers looks at skills programmes; Compass is presented as instructor-development support. The University of Maryland’s M-Powering Teachers project is a separate example of AI-assisted educator feedback. It is a University of Maryland project, not Stanford’s.

People retain the consequential decisions

Multiverse says Delivery Managers review outputs, check concerns, agree development goals and make decisions. It says employment decisions remain human and performance management relies only on human observation. Its AI principles from January 2025 are older company policy, not independent evidence that Compass works as intended.

Human oversight also matters in schools. Our report on the AFT, UFT and Microsoft school AI standard covers a different initiative but raises the question of educator involvement. Google’s classroom-interactive research is another learner-facing use, not coach review.

The October step was still a plan

The September post says managers tested reports in August and introduced instructors to Compass in September while reports remained manager-visible. Moving reports into an internal coaching platform was planned for October. That step was future-tense in the source; this article cannot confirm completion.

Because Compass reviews workers’ sessions, data handling merits scrutiny without assuming wrongdoing. The UK Information Commissioner’s Office updated worker-monitoring guidance on September 28, 2026. It is UK data-protection context, not a finding about Multiverse. Details such as purpose, access and retention still need confirmation. Our reporting on AI literacy in Finnish primary schools covers a separate effort.

Evidence is not the same as a company claim

The University of Maryland project page links a 2025 peer-reviewed study about automated feedback and teachers’ questioning. It is not an evaluation of Compass and cannot establish that the tool improves coaching. Multiverse’s July 2026 impact report and July 2026 impact hub are company-published background, not independent validation of this rollout.

Our report on the NBER Khanmigo school trial covers a different intervention, not coach development. Multiverse’s September post also reports a coach-satisfaction percentage without sample size or independent evaluation method. That is a company-reported figure, not proof that Compass caused an outcome.

What a credible evaluation would need to show

Useful evidence would show how often managers correct flags, whether instructors find feedback relevant, and whether development goals lead to observable change. It should explain how learner feedback combines with session review and how differences between instructors or contexts are handled. The 2023 UNESCO guidance on generative AI in education emphasizes human agency and ethical validation; the 2023 NIST AI Risk Management Framework offers broader, voluntary risk context. Neither assesses Compass. The practical test is useful, contestable feedback while humans retain authority over consequential choices.