AI Can Teach a Lesson. Who Gets the Student to Show Up?
Stephen Hodges, CEO of Efekta Education, on affordable personal tutoring, the work of introducing AI into schools, and the evidence we need before calling it a success.
Stephen Hodges offers a simple scenario. You have a lesson at six in the evening. You are tired, dinner is waiting, and you would rather spend time with your children. If the tutor is a bot, skipping feels easy. If a person is waiting for you, there is a stronger reason to turn up.
In our EdTech Dots conversation, this small example brings a practical problem into view. A system can explain a subject and provide individual exercises. Getting someone to keep learning requires more than access to that system.
Hodges runs Efekta Education, which develops AI teaching technology for use in education systems. His ambition is substantial: personal tutoring at a price comparable to a textbook. Yet he repeatedly returns to teachers, local support and the relationships that make education work. For teams building or buying education software, those details deserve as much attention as the tutor itself.
A personal tutor inside a crowded classroom
Efekta grew out of technology developed at EF Education First. Hodges describes a long progression from early digital learning products to online classrooms connecting a teacher and a student in different parts of the world. Generative AI opened another possibility: lowering the cost of individual instruction without requiring a teacher's time for every interaction.
Consider the classroom he describes. Thirty or forty students are learning English, with widely different starting points. Teaching at one pace leaves some bored and others unable to follow. Even an excellent teacher has limited time for individual speaking practice and feedback.
The proposed role of AI is concrete. Students can practise at their own pace while the teacher spends more time with those who need help. In places with severe shortages of qualified subject teachers, Hodges sees a further use: supporting educators who have been asked to teach a subject outside their training.
These are different situations, and Efekta does not prescribe one fixed balance between teacher and software. A qualified English teacher may mainly want more opportunities for students to practise. A teacher covering English without the relevant qualification may rely more heavily on the system's instruction. The product has to accommodate both.
Giving the teacher something useful to decide
Hodges uses the term agentic teaching for a system that can take on parts of instruction. In practice, the teacher decides which students use the technology, when they use it and how much of a lesson to hand over.
That choice matters when a student remains stuck. More automated explanation is not always the next useful step. Hodges describes the system alerting a teacher when it has tried to teach a concept and the learner is still struggling. The teacher can then intervene directly.
The same boundary applies to suspected cheating. Unusual patterns or sudden changes in performance can prompt a flag. According to Hodges, the system leaves the response to the teacher, who knows the student and has context the software lacks. Where a conversation raises a concern about a child's safety, he describes stopping the AI conversation and alerting the teacher.
This suggests a specific design task: deciding when to hand a situation over, and what information the person receiving it needs. A promise to keep a human involved is thin unless the product makes that involvement workable.
There is also routine work to remove. Hodges describes automated grading, homework tracking, suggested feedback and support for lesson planning. His account of a changing teaching role includes more time with individual students, while teachers retain room to decide how they teach.
The pilot has people around it
A successful pilot can conceal how much support a product needs. With a few thousand students, a vendor may be able to guide teachers closely through the experience. An entire education system needs a different arrangement.
Efekta's approach, as Hodges describes it, centres on ambassadors in school districts. The company trains them, they train local teachers, and Efekta supports them as questions and difficulties arise. The team follows up where training or adoption is falling behind.
This is a practical division of work with the ministry of education. The supplier cannot control everything that happens in a school. Hodges argues that both sides need to be clear, from the start, about what each is responsible for.
He applies the same reasoning to his own developers. Telling them to adopt a new AI coding tool would not, by itself, ensure that everyone used it. Training and incentives would still be needed. Teachers also need an explanation of why a change helps them and their students, followed by support while they learn a different way of working.
For a product team planning expansion, the implication is uncomfortable but useful. The support surrounding a pilot belongs in the rollout plan. If local people are essential to adoption, someone has to recruit, train and support them at the next scale too.
Content has to belong in the classroom
Scaling across countries also means adapting the learning experience. Hodges gives the example of an Egyptian student encountering exercises about going to a mall and buying a burger. Such examples may be distant from that student's everyday life.
He describes much of Efekta's adaptation as making content culturally appropriate and engaging. That does not necessarily require rebuilding the sequence of English learning objectives for every market. It does require attention to the examples through which those objectives are taught.
The distinction is useful for teams estimating the work of localisation. Translating the interface addresses only part of it. Hodges also points to national curricula, integration with existing systems and the information teachers and ministries need. These requirements help explain why a general-purpose AI tool and a product used throughout a school system can be very different things.
Better results, with a question still attached
When we turn to evidence, Hodges describes two kinds of measurement. Efekta tracks proficiency within its system. Ministries also look at external exams, which sit outside the product and have their own public significance.
He reports improvements of around 25% or more in state exam results before and after rollouts. That is a claim made in the interview, rather than an independently assessed result presented here. The qualification that follows it is essential.
Asked how to separate the software's contribution from teacher quality or student motivation, Hodges acknowledges the difficulty. He would like controlled comparisons, but says ministries are reluctant to exclude groups of students from access to the technology. His team therefore also examines relationships between usage and results across classes, schools and districts.
A before-and-after improvement cannot, on its own, establish how much was caused by the software. Nor does the fact that frequent users get better results settle the question: more motivated students may both study more and perform better. Hodges recognises that problem in the conversation.
For buyers, the useful response is to keep asking how an outcome was measured, what was compared and what else could have contributed. Those questions help define what a deployment should actually be evaluated against.
What happens to the school day?
Hodges expects human teachers and schools to remain central. He points to motivation, social interaction, childcare and learning how to function alongside other people. Delivering subject knowledge addresses part of what a school does.
His definition of better education leaves room for different ambitions. Some students could reach a higher level. Others could reach the level they need in less time. Freed-up time might go towards sport, further study or other activities within the school day.
That gives us something more specific to look for than a convincing AI demonstration. Does the student learn more? Can a teacher help someone who would otherwise remain stuck? Is there time for something that used to be squeezed out?
The lesson at six still needs a student who turns up. In a school, it also needs a teacher who knows when to step in and an organisation that makes the new way of working possible.
Based on Krzysztof Kosman's conversation with Stephen Hodges for EdTech Dots. Company practices and results described in the article are attributed to Hodges's account in the interview.
