Multiverse uses AI to score instructors’ lessons throughout the day

From monthly checks to hours of monitoring

Multiverse was founded by Euan Blair, the former UK prime minister’s eldest son. The company employs more than 800 people and trains workers at the National Health Service, local councils, universities and private companies. Its courses are funded largely through the government apprenticeship levy.

The new agent, introduced in recent weeks, analyzes lesson transcripts and assigns each instructor a “risk status” and “confidence percentage.” It can flag a teacher for leaving a connection problem unresolved for about a minute, using too many filler words, sounding vague or uncertain, or focusing on one learner for too long while others wait.

Before the system, managers observed each instructor once a month. Now AI can monitor them for several hours a day. The instructions tell the system to look for patterns across a transcript rather than isolated slips. But they also say that if an instructor visibly gets confused or corrects themselves while explaining an exercise, that counts as evidence the explanation was unclear the first time—even if the instructor fixes it.

Several instructors said the system has left them tense, sleeping poorly and seeking therapy. One, speaking anonymously, said the stress had become “just awful” and had begun waking him in the night.

“Instead of teaching the lesson and listening to the student, I’m thinking: am I saying this correctly now, will AI lower my score because of some stupid thing I said by accident?”

Multiverse says AI does not replace human judgment. Its role, the company says, is to point managers toward lessons that may need attention; only managers write performance feedback. The most serious action the system can take is to recommend that a person review a lesson. The company says instructors were told about the rollout in advance and will be able to access the data collected.

The gap between support and surveillance

The rollout follows a broader turn toward AI-based workplace monitoring. Burger King began using AI earlier this year to analyze restaurant workers’ conversations with customers. Meta paused keystroke tracking on employees’ computers in the summer after privacy concerns and staff dissatisfaction.

Critics argue that this kind of monitoring can reduce workers’ autonomy. One study found it can provoke resistance. A new study due this week says UK technology workers believe AI is reducing job satisfaction. The CWU’s United Tech and Allied Workers branch has described “cognitive surrender”: workers lose skills and confidence, while work becomes more intense rather than easier.

Euan Blair has argued for wider AI adoption in the UK economy. Last week, he said “every worker will need retraining” and that employers want their staff to “embrace AI as much as possible.” Multiverse uses models from Anthropic and OpenAI. The company is valued at £1.6 billion, against £118 million in revenue and £28 million in losses last year. Blair owns a large minority stake, and his personal wealth is estimated at £350 million.

In February, Ofsted found Multiverse fell short of expected standards in several areas, while praising its instructors as “qualified and effective.” That contrast matters: the people doing the teaching are being scored by a system whose judgments can prompt managerial scrutiny, even when an instructor’s apparent deviation is an attempt to respond to a student.

What the system cannot see

One instructor’s account, published alongside the reporting, describes managers questioning teachers over supposed departures from the lesson plan when they had actually been answering students’ questions. The instructor says staff cannot see how the AI analyzes lessons; so far, only managers have access to the dashboard. The company says instructors will eventually get access.

The instructor’s concern is not only that the system makes mistakes. Constant monitoring can shift attention away from the learner and toward the score: whether a phrase might be misread, or an adaptation might look like noncompliance. The instructor also fears recordings could be used to train AI that might eventually replace mentors and teachers, though the article provides no evidence that Multiverse is doing this.

I think the unanswered issue is what the risk rating is for beyond deciding which lesson a manager should review. Multiverse describes a quality-control tool, but instructors experience an opaque system that can shape how they teach before any manager steps in. If the score changes the lesson itself, the boundary between support and surveillance is already doing less work than the company’s description suggests.