This is not a conventional product comparison, because these products do not compete: Medly AI tutors GCSE, A-level and IB students; iatroX serves medical students and doctors. What makes the pairing worth examining is that they are two applications of the same technological shift, the verticalisation of AI tutoring, and comparing them shows precisely what changes, and what does not, when the subject matter becomes medicine. For readers weighing actual product decisions in either market, this is background; for anyone thinking about where AI education is heading, it is the interesting version of the question.
Two very different learners, one underlying problem
Personal tuition has always worked for the same reason across every subject: it reacts to the individual, their attempt, their misconception, their next step. And it has always failed to scale for the same reason: reacting to individuals is expensive. Generative AI changes the economics of exactly that reaction, which is why the first durable AI education companies are not content libraries but feedback systems, and why they are emerging simultaneously in markets as different as school revision and postgraduate medicine.
Medly's model
Medly's implementation, per its public materials and coverage (last checked August 2026): an AI tutor constrained to the learner's curriculum and exam board; marking of the learner's own typed or handwritten answers; feedback that identifies where marks were lost rather than supplying model answers; and weakness identification feeding what gets practised. Reportedly backed by close to £6 million and used by more than 400,000 UK students, it is the strongest current evidence that curriculum-aware, feedback-first tutoring is what the market wanted all along; the fuller analysis of its design is at /blog/medly-ai-lessons-for-medical-education.
What changes in medicine
Three things, each structural. Scale of curriculum: a specification becomes a profession; the blueprint spans thousands of presentations, dozens of exams, multiple jurisdictions, and the tutoring system must know which of the 40+ assessments this learner faces. Nature of the material: school questions have answers; clinical questions have best answers, surrounded by plausible alternatives, atypical presentations and guidance that updates, the same disease genuinely presents differently, and the marking problem becomes probabilistic. And nature of the errors: the school learner's mistake is usually a knowledge or technique gap; the medical learner's is often a reasoning failure, the discriminating feature under-weighted, the competing diagnosis never generated, and feedback that cannot see reasoning cannot fix it.
iatroX's approach
The iatroX architecture is what Medly's model looks like after those three changes are taken seriously. The exam-specific question bank identifies the gap, adaptively, against the learner's actual blueprint. The Socratic Tutor interrogates understanding rather than presenting explanations: what did you weight, what did you exclude, what would change your mind, until the misconception is named against cited guidance. Spaced repetition revisits the gap on a forgetting-curve schedule, because a corrected error unretested is a corrected error on loan. The Study Planner converts performance into priorities, deciding what tomorrow should contain. And askiatroX sits alongside for the moments the context shifts from examination to practice, the same knowledge, needed clinically, with citations into the guidance itself, a component the school version of this problem simply does not need.
Why vertical tutors may beat generic AI for structured learning
The pattern both products embody: generic models know everything and know nothing about you. A vertical tutor holds the three things a frontier chatbot structurally lacks, the assessment blueprint, the learner's longitudinal performance, and the pedagogy wired into the interaction loop, and those three compound: blueprint plus performance yields diagnosis, diagnosis plus pedagogy yields the right next activity. The generic model can be prompted toward any one of these in a single session; it cannot hold them across months, which is what learning takes. Our direct test of that claim against the strongest generic offering is at /blog/can-chatgpt-study-mode-replace-a-medical-ai-tutor.
Conclusion
Medly and iatroX are the same bet placed in different markets: that the value of AI in education lies not in generating answers but in orchestrating learning, and that orchestration requires verticality, the curriculum, the learner model, the loop. The competition that matters over the next five years is probably not Medly versus tutoring agencies or iatroX versus question banks; it is domain-specific learning systems versus generic chatbots, and both companies are evidence for the same side.
Frequently asked questions
Could Medly expand into medical education?
Its founders are medically trained, so the domain knowledge exists; the barrier is the one this article describes, medicine's curriculum plurality and reasoning-centred errors make the build categorically larger. Any entrant discovers that the exam-by-exam construction is the product.
Which should a medical student's younger sibling use?
Medly, without hesitation, for GCSEs and A-levels; the medical student should use the medical version of the same idea. The household analogy is exact, which is rather the point of this article.