Opinion AI at work · Learning and development
Will AI replace classroom trainers? Two views
Publisher's interest: our publisher, WeLearn2Profit, runs team-building and training programmes.
In short
AI tutors have already beaten good classes on some topics. But results depend on how the AI is designed, and evidence from workplace training is thin. My view: AI will replace much of the lecturing, not the trainer.
Generative AI can now explain a topic, quiz you and give feedback at any hour. So HR teams are asking a fair question: do we still need classroom trainers?
Here are the two strongest answers, the evidence behind each, and where I land.
What the research shows so far
In a Harvard trial published in 2025, physics students learned more from a purpose-built AI tutor than from an active-learning class. The researchers found that “students learn significantly more in less time when using the AI tutor”.
A 2025 study of nearly 1,000 students at a high school in Turkey found a risk. Students who practised with a plain ChatGPT-style tool did better while they had it. Once it was taken away, they scored 17% lower than students who never had it.
A version of the tool with teaching safeguards largely avoided that harm. How the AI was designed made a big difference.
At work, a study of more than 5,000 customer support agents found that AI help raised productivity by 15% on average. Less experienced and lower-skilled workers gained the most.
Older reviews point the same way. Intelligent tutoring systems, computer tutors that guide learners step by step, were “nearly as effective as human tutoring” (VanLehn, 2011).
Web-based training taught facts slightly better than classrooms and skills equally well. When both used the same teaching methods, the difference disappeared (Sitzmann and colleagues, 2006).
Note the limits. Most of these studies took place in schools and universities. The closest workplace evidence we found is two small US Navy trials of an older AI tutor, covered below.
We found no study that tested generative AI replacing trainers in corporate classroom training.
View 1: AI will take over much of classroom training
- It already teaches some topics better. A well-designed AI tutor beat a strong active-learning class at Harvard.
- It has beaten classroom courses at work before. In two small US Navy trials, an older AI tutor trained IT staff in under half the time of the classroom course. Its trainees also scored much higher on end-of-course tests. The trials had 40 and 24 trainees, and the reviewers called these results outliers.
- It speeds up learning on the job. In the customer support study, newer staff gained the most from AI help.
- Planning work is exposed. An International Labour Organization (ILO) study rated primary school teachers’ lesson planning as having medium exposure to generative AI.
- Employers expect change. In the World Economic Forum’s Future of Jobs Report 2025, 40% of employers expected to reduce their workforce where AI can automate tasks.
View 2: AI changes the trainer’s job, not the need for one
- The AI tutor’s own authors say so. The Harvard team wrote that “an AI tutor should not replace in-person teaching”.
- Unguided AI can hurt learning. In the Turkish study, students leaned on the plain chatbot and learned less. Someone has to design how AI is used.
- The human parts of teaching are less exposed. The same ILO study rated instructing as low exposure, and keeping discipline as very low.
- Humans gave better feedback in one test. In a 2024 study of feedback on school students’ essays, human raters did slightly better than ChatGPT (GPT-3.5) on all but one quality measure.
- AI can make trainers better. In a US trial, students whose tutors had an AI assistant were 4 percentage points more likely to master maths topics. Students of lower-rated tutors gained the most. (This is a working paper; we found no peer-reviewed version.)
- The need for training is large. The World Economic Forum says 59 in 100 workers worldwide would need training by 2030. In Budget 2027, Malaysia’s government said HRD Corp will offer 3 million skills training opportunities with RM3 billion in funding.
My view
I think AI will replace much of the lecturing in classroom training. Explaining facts and checking understanding are where AI tutors already do well.
But I don’t think it replaces the trainer. The parts that change behaviour, such as practice, feedback, real work problems and follow-up, still need someone to design and run them.
The trainers who do well will use AI for preparation and practice. They will spend their own time where people need a human most.
What HR teams can do now
- Ask providers how they use AI before, during and after a session.
- Use AI for knowledge checks and practice, not as the whole programme.
- Check what data an AI tool stores about your staff, and where.
- Measure what people do differently at work, not just quiz scores.
Sources
- Kestin and others (2025). AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting · Scientific Reports 15, 17458
- Bastani and others (2025). Generative AI without guardrails can harm learning: evidence from high school mathematics · PNAS 122(26), e2422633122
- Brynjolfsson, Li and Raymond (2025). Generative AI at work · Quarterly Journal of Economics 140(2), 889–942
- VanLehn (2011). The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems · Educational Psychologist 46(4), 197–221
- Sitzmann, Kraiger, Stewart and Wisher (2006). The comparative effectiveness of web-based and classroom instruction: a meta-analysis · Personnel Psychology 59(3), 623–664
- Kulik and Fletcher (2016). Effectiveness of intelligent tutoring systems: a meta-analytic review · Review of Educational Research 86(1), 42–78
- Gmyrek, Berg and Bescond (2023). Generative AI and jobs: a global analysis of potential effects on job quantity and quality · ILO Working Paper 96
- Steiss and others (2024). Comparing the quality of human and ChatGPT feedback of students' writing · Learning and Instruction 91, 101894
- Wang and others (2024, revised 2025). Tutor CoPilot: a human-AI approach for scaling real-time expertise · EdWorkingPaper 24-1054, Annenberg Institute at Brown University
- The Future of Jobs Report 2025 · World Economic Forum, 7 January 2025
- Budget 2027 speech · Ministry of Finance, 9 October 2026