Why we built the LCP before the LMS.

Order of operations is architecture. The decisions you make about what to build first shape everything you build afterward. ArthurAI was built in an order that looks backward from the perspective of the EdTech market, and the order is the most architecturally consequential decision the company has made.

Most EdTech ships the Learning Management System (LMS) first. The LMS is the container: rosters, courses, content, assessments, gradebook, reports. With the container shipped and customers using it, the company then adds "personalisation" as a feature — usually a recommendation engine that suggests which static lesson a student should see next, based on telemetry from the gradebook.

That order produces a recognisable kind of product. Personalisation is bolted on. The recommendation engine works at the lesson-selection level. The lessons themselves are static. The system can personalise which lesson is played; it cannot personalise the lesson itself. That ceiling is built into the architecture by the order in which it was constructed.

We started with the Learning Cognitive Profile. Thirty questions across five categories. Four cognitive dimensions (visual-verbal, active-reflective, sensing-intuitive, sequential-global). The instrument was the first thing the platform asks every learner to complete, because the lesson generator that comes next needs the four-coordinate signature as input on every generation call. The LCP was not a feature. The LCP was the precondition.

Then we built the lesson generator. Not a content library with a recommendation engine on top, but a generation pipeline that produces a fresh lesson per learner per concept per day. The generator reads the LCP signature on every call and adapts the lesson's block composition to the learner's cognitive style. Visual-strong learners get more diagrams, hierarchy trees, comparison tables. Verbal-strong learners get more definitions and accordions. Same concept; structurally different lesson per learner.

Then we built the live tutor. Not as a chat surface bolted on, but as a streaming tutoring channel that inherits the same LCP signature, the same cognitive posture, the same disciplinary reasoning. The tutor and the lesson are the same product surface seen from two angles, not a tutor app glued onto an LMS.

Then we built the assessment surface, the certificate system, the dashboards, the reports, the bulk import. These are the LMS-like surfaces. They came last, because they need to be downstream of the cognitive instrument, the generation engine, and the tutor — not upstream of them.

Closed-loop adaptive learning, at the lesson level, not the lesson-selection level. When a learner crosses 50% completion of the day's lessons, the entire next day's lessons are queued for generation, with today's practice accuracy passed in as mastery signal. The system is not picking which static lesson to show tomorrow; it is generating tomorrow's lessons, calibrated to who the learner is and what they just demonstrated they had or had not mastered.

That capability is unavailable to any platform that built the LMS first. A content library with a recommendation engine on top cannot, structurally, do this. The ceiling is the static content. The generation engine sees no such ceiling because the generation engine is the content layer.

Time and complexity. The LMS-first companies in market were faster to ship. Their first version was a complete container; ours did not have a gradebook for a long time. We invested in the LCP, the generator, and the tutor before we shipped the surfaces a procurement officer recognises from prior EdTech evaluations. Customers had to understand that what they were buying was architecturally different. Some did. Some did not.

The cost was worth it, because the order is the architecture. We could have built the LMS first and added personalisation. We would have ended up with a product the market already had several of. We chose to invent the instrument, the engine, and the tutor first, and let the LMS surfaces follow from what they enabled. The platform that exists today is the artefact of that order.

How the LCP actually works →