Giving AI 70 Percent of the Teaching Revealed What Humans Do Better

Last updated: July 3, 2026

A Bloomberg opinion piece published July 1 profiles Per Scholas, a no-cost tech-training nonprofit operating in 24 cities, as a model for what AI in the classroom actually looks like when it’s working. The headline claim: AI handles up to 70 percent of instruction. That is a real change worth understanding. But the more useful story comes from Per Scholas’s own five-year research partnership with the American Institutes for Research, which found that when the nonprofit introduced an AI tutoring platform for its learners, nearly half never used it. The researchers’ conclusion: adoption of a learning tool is a design challenge, and the real design problem is not the software. It is the human relationship that makes a learner willing to try something unfamiliar. Parents deciding whether to hand their child an AI learning app should read this research first.

TL;DR

Bloomberg (July 1, 2026) profiled Per Scholas as a model for AI-integrated learning, with AI potentially handling 70 percent of instruction.

Per Scholas and AIR's five-year research found that when AI tutoring was optional, nearly half of learners never used it - the technology worked, but adoption was the human-relationship challenge.

Per Scholas's results - 84 percent job placement, doubled and tripled incomes - are built on human wraparound: counseling, zero-interest loans, instructors who know their students as individuals.

AI learning apps for children can be effective tools when a child engages - but the human presence that creates that engagement is the prerequisite, not a nice-to-have.

Per Scholas operates in 24 cities with no-cost 15-week courses combining live instructor hours with AI-led practice sessions via its Azari tutoring platform.

A Bloomberg analysis of Per Scholas, one of the country’s most sophisticated AI-integrated training programs, describes a future where AI handles 70 percent of instruction. Their own five-year research partnership with the American Institutes for Research describes the challenge that got them there: when the AI tutoring tool was introduced, nearly half of learners never used it. What bridged that gap tells parents everything about how to use AI learning tools effectively at home.

Common questions

Are AI tutoring apps worth using for my struggling child?

They can be. Per Scholas’s research suggests AI tutoring works when learners engage with it, and some tools are genuinely well-designed for targeted skill-building. The question isn’t the app; it’s whether your child has enough belief in their own progress to keep trying when it gets hard. A screener is a starting point, not a diagnosis. If your child might need formal accommodations (an IEP or 504 plan), or you suspect a vision, hearing, or medical cause, pursue a professional evaluation too – that’s the only route to those supports.

If AI can handle 70 percent of instruction, what do human teachers actually do?

Per Scholas found that instructors do the part that doesn’t fit into the 70 percent. They read student discomfort as a predictor of success. They adjust based on what they observe in real time. And they answer the question every struggling learner is actually asking: is getting better possible for someone like me? AI handles content efficiently. Humans handle the belief that the content is worth tackling.

My child has an AI learning app. Should I sit with them or let them work alone?

The Per Scholas research suggests it depends on what your child needs to keep going when the task gets hard. For children, the choice to push through or give up is often shaped by whether someone who matters is present and expects them to continue. Sitting nearby – not hovering, just being in the room – signals that you expect success. That signal matters even when you say nothing.

How is workforce training research relevant to my K-12 child?

The mechanism is the same at any age: learning involves challenge, and persisting through challenge requires believing it’s worth it. Carol Dweck’s decades of identity-based motivation research shows this holds from elementary school through adulthood. The difference with younger children is that the human relationship shaping their beliefs is primarily a parent – not an instructor they chose.

Per Scholas gave AI 70% of instruction. Their own 5-year research: nearly half of students never touched the AI tool. What that tells every parent about learning apps. Tweet Now Follow Learning Success

What Per Scholas actually built – and what the research actually found

Per Scholas runs 15-week, no-cost technology training programs across 24 U.S. cities, helping adults transition into IT support, cybersecurity, and software engineering. Its St. Louis location has graduated more than 400 learners with an 84 percent job-placement rate; most arrive earning roughly $19,000 a year and double or triple that figure after graduation. The nonprofit is backed by Google, the Lilly Endowment, and the New York Times Communities Fund, and it offers wraparound services – zero-interest loans (which boosted graduation rates by 13 percent in one internal study), counseling, equipment loans, transportation vouchers, and career coaching – alongside its coursework.

The AI integration is real. Per Scholas uses Azari, a custom AI tutoring system available 24 hours a day that adapts to each learner’s pace, providing targeted practice, real-time help with coursework, and interview preparation. A Bloomberg analysis published July 1 describes the trajectory: AI handling up to 70 percent of instruction, with human instructors focused entirely on mentorship and relationship-building.

But the joint research Per Scholas and the American Institutes for Research published in May 2026 – drawn from five years of collaboration – offers the necessary context: when an earlier AI tutoring platform was introduced in the program, nearly half of participants never engaged with it. The tool worked for learners who used it. The gap was not the technology.

Author Quote "

I appreciate worrywarts: Their discomfort drives them to be successful. I’m looking very forward to seeing what she’s able to do.

What the coverage gets wrong

Most coverage of AI in education frames the question as whether AI will replace teachers, treating any AI-ed adoption story as either a salvation or a threat. The Per Scholas and AIR research asks a more useful question: why do learners engage with tools that work for them when they do - and what prevents engagement when they don't? The answer from five years of real program data is relational, not technological. Coverage that frames AI adoption in education as an innovation story misses what Per Scholas's own data shows: the human relationship is not a nice-to-have alongside the AI. It is the condition under which any learning tool gets used by the learners who most need it.

Why the AI worked for some and sat unused for others

Per Scholas and AIR concluded that adoption of an AI learning tool is a design challenge, not a technology challenge. Before implementing any AI solution, organizations need a clear theory of change: how is this tool expected to create value, what assumptions are being made about how learners will behave, and what does success actually look like over time? When those questions are answered in advance – and when the human infrastructure supports engagement – AI learning tools can be powerful. When they are not, learners who most need the tool are often the least likely to open it.

The identity research makes the mechanism clear. Carol Dweck’s work on motivation shows that whether a learner engages with a challenge depends not primarily on the difficulty of the task but on whether they believe the challenge is for someone like them. An AI that presents a quiz does not answer that question. A human who has watched them struggle and stuck around anyway does. Per Scholas instructor Zell Davis described one anxious learner in the February 2026 IT Support graduation ceremony: “I appreciate worrywarts: Their discomfort drives them to be successful. I’m looking very forward to seeing what she’s able to do.” That observation – reading a student’s anxiety as a predictor of success rather than a sign of weakness – is what changes what a learner believes about herself.

The AIR researchers found that organizations best positioned for AI-powered learning are not those that adopt AI first. They are those that “build internal capacity to continuously learn, adapt, and improve.” At Per Scholas, that capacity was built through human relationships: instructors who know their students, counselors who help them through the hard stretches, and a belief in the brain’s ability to change with the right practice. The AI tool worked at scale only after the human scaffolding was already in place.

Key Takeaways:

AI tools work for learners who engage: Per Scholas's five-year research with AIR found AI tutoring was effective for learners who used it - but nearly half never engaged, revealing a human-relationship gap the technology could not fill on its own.

Human relationship is the prerequisite, not the supplement: What got Per Scholas learners to persist through 15-week programs was counseling, zero-interest loans, and instructors who read student anxiety as a predictor of success - not the AI platform.

Parents are the wraparound their child's AI tools need: The supports Per Scholas artificially builds for adult learners - someone who believes in you, stays through the hard part, and notices before you quit - are what a present parent provides. That presence makes any learning tool more effective.

What this tells parents about AI learning apps at home

This is not an argument against AI tutoring tools for children. The evidence from Per Scholas says they work when learners engage with them – and some AI practice tools are genuinely well-designed for targeted skill-building. But Per Scholas instructor Chris McCain put the human element plainly in March 2026, when asked about AI and the future of tech-support roles: “You’re going to have to have the oversight, so I’m not worried about robots taking over. Nope.” She estimates that half of what she teaches in IT support could be automated within a few years. The human in the loop remains essential not because AI is weak but because the judgment about whether the tool is working in real life – and the adjustment when it is not – requires someone who actually knows the learner.

For a parent of a child who struggles with reading, math, or attention: you are that oversight. Not in a supervisory sense – in a relational one. The question is not whether the app is any good. It is whether your child trusts that getting better is possible, and whether you are present enough to notice the moments when they stop believing it. An AI tutor is patient, available, and never frustrated. It cannot tell you that your child sat with the app for three minutes and gave up because they believed they were the kind of person who fails at this. You can see that. You can respond to it. That difference is what the Per Scholas data is actually measuring.

The sequence matters: belief first, tool second. Per Scholas builds belief through human relationships, then layers the AI practice on top. Parents using any AI learning technology with a struggling child should take the same approach. Stay in the room. Not to supervise – to signal that you expect success.

Author Quote "

You’re going to have to have the oversight, so I’m not worried about robots taking over. Nope.

AI learning tools are real and getting more capable. For a child who struggles to read or calculate, a patient, always-available practice partner sounds like exactly what’s needed. And the Per Scholas evidence says it can be. But the research also shows that handing a struggling learner a good tool and stepping back produced a 50 percent non-engagement rate among adults who chose to be there. The human presence that signals “I expect you to succeed” is not a supplement to the technology. It is what makes the technology worth opening. The Learning Success AI Assessment identifies the specific cognitive gaps where your child needs targeted practice – so you know what you’re working toward together, not just which app to open next.