Creating lessons with a local agent
One shot
Local Agent
Research & plan: Claude reads the NL curriculum and writes a series plan covering objectives, vocab, and misconceptions.
Build lessons: it drafts each 50-minute lesson with quizzes, images, and teacher notes, saved as source files.
Push to LessonUp: MCP delivers the lessons straight into LessonUp as editable drafts with preview links.
VS
Simple
Predictable
Persistent workspace
Consistent
house style
Tool integration
Progressive knowledge
Human-in-
the-loop
Reusable
Creating lessons with a local agent
One shot
Local Agent
Research & plan: Claude reads the NL curriculum and writes a series plan covering objectives, vocab, and misconceptions.
Build lessons: it drafts each 50-minute lesson with quizzes, images, and teacher notes, saved as source files.
Push to LessonUp: MCP delivers the lessons straight into LessonUp as editable drafts with preview links.
VS
Simple
Predictable
Persistent workspace
Consistent
house style
Tool integration
Progressive knowledge
Human-in-
the-loop
Reusable
one shot:
Simple
Predictable
Local Agent:
Persistent workspace
Progressive knowledge
Tool integration
Consistent house style
Human-in-the-loop
Reusable
What did We/JW accomplish?
42 consistent practice exams produced in a fraction of the usual time
Based on the methodology of the real exams
Newly created sources and data
First-time local agent for JW
Proves the model, every run after raises quality
Next steps still TBD
Deze slide heeft geen instructies