Product 01 on the Learneta platform
Agentic LXD
Agentic LXD is product one on the Learneta platform. It is focused solely on the LXD process, the design and build of learning experiences, with the online-programme operating model as its archetypal home and enterprise equally in scope. It takes an expert's intent and approved content through to built learning in seconds.
Built in public under the working name Learneta.AI; shipping under Agentic LXD.
Two working engines
From a briefing to a validated design. From a storyboard to a built module.
The nine-agent design engine
Takes a subject-matter expert's briefing and produces a validated learning design and assessment plan. Nine agents, each holding one part of the design discipline, checking each other's work before anything reaches you.
Out validated design · assessment plan
The non-generative builder
Takes a signed-off storyboard and renders it into an interactive module. The source text is preserved verbatim; the AI decides component choice, sequence, interactivity, and pacing, and every section carries a verification state.
Out interactive module · verification per section
The principle
The AI never writes content. It makes pedagogical decisions about how to render content.
Most AI authoring tools treat the source as something to be rewritten. That puts the model in the wrong job. The pedagogical job, deciding how to render approved content into components a learner can use, is the one the model is good at. The writing job, deciding what words to use, is the one your SME, your legal reviewer, and your counterparty have already done.
In regulated and enterprise learning, content is signed off once. Every paraphrase restarts the review cycle. Agentic LXD makes the pedagogical decisions, leaves the words alone, and lets you ship the module that was already approved.
The verification mechanic
Every section carries one of three states, computed from the source.
Verified
Every span in the rendered section traces to the source storyboard. Coverage is 100%. No paraphrase, no addition.
Drawn
Most of the section traces to source spans. A bounded set of fields (titles, prompts, button labels) are drawn from approved patterns and flagged for SME review.
Drift
A span has appeared that does not trace to the source. Flagged immediately. The SME must approve, replace, or remove it before the module can ship.
The state is computed from the source, not asserted by the model.
The component library
Eight pedagogically distinct components. The choice is the pedagogical decision.
TextBlock
Source narrative, rendered for readability.
KeyPointCallout
Pull the line worth surfacing out of the source.
AccordionReveal
Drill-down for hierarchical content.
SortableCards
Active recall through sequencing.
ScenarioBranch
Decision under constraint, with consequence.
KnowledgeCheck
Formative check, source-aligned.
ComparisonTable
Set things next to each other.
ReflectionPrompt
Open question, no graded response.
Status
Two working prototypes, in active build.
Agentic LXD is in R&D, not on general release. The architecture and the thinking are published as the build goes on; the writing is the best window into both. If the problem it solves is your problem, say hello.