Monthly Meetup
October 2026
Monthly Meetup
October 2026
Deze slide heeft geen instructies
Product update
Phase out Classic Lesson Editor
Share with class
Traffic light
Draw on slide
Activation
Guide users
Improve PowerPoint import
Request user information
LessonUp Studio
Deze slide heeft geen instructies
Product update: performance metrics NLE - October 2026
Creating users
Teaching users
Disclaimer: It's only October 8
Deze slide heeft geen instructies
Product update: lessons created
60% of lessons that are created from scratch are created in the NLE
40% of lessons that are created from scratch are created in the CLE
Measured since the beginning of the current school year
Deze slide heeft geen instructies
Product update: lessons saved from the library
66% of lessons that are saved from our library are NLE lessons
33% of lessons that are saved from the library are CLE lessons
Measured since the beginning of the current school year
Deze slide heeft geen instructies
LESSONUP · AGENTIC HARNESS UPDATE
Prototype in hours,
ship in small steps
~45s. Where we are. With the harness, Claude builds a working prototype of a roadmap feature in a fraction of the hours a classic project takes: traffic light and draw on slide are both merged, and go live on 13 October. And it's not just developers: Olaf, our designer, now prototypes his own ideas. Honest part: the code quality is not good enough yet to merge a one-shot build as it is. So the plan has steps. First we explore how prototypes fit into our workflow. Then we learn to break a prototype down into small PRs, because one big build is hard to review and still contains errors. The goal: each of those small PRs gets through with a single approval. The stretch goal: a PR approved the first time.
Feedback loop: the harness improves with every run
Build, blind
Claude, from the brief
Human review
The answer key
Score
Rebuild against the review
Trace
Each miss to a rule
Improve
Rules, hooks, skills, reviewers
GOAL
Small PRs, one approval each
THE WHOLE TEAM
Every improvement lands in the shared harness
~60s. This is the harness and its feedback loop. Claude builds a feature, a human reviews the PR, and that review becomes the answer key. We then rebuild the same feature blind: the run is barred from the PR, the tickets and the project doc, and a leak scan checks it stayed blind. A fresh agent scores the rebuild against the review, thread by thread. Every miss is traced to a rule that was wrong, missing or never read, and the harness improves. And that improvement isn't just for us: it lands in the shared harness, so every developer, designer, PM and QA person who builds with Claude gets the better version.
How we work: two goals that feed each other
GOAL 1
Improve the harness
GOAL 2
Deliver features
every feature teaches the harness
a better harness builds the next feature faster
Yjs rule, fit-reviewer, the Opus 5.5 rewrite
Traffic light and draw on slide, live 13 October
~30s. We work with two goals at once. One: improve the harness. Two: deliver real features while we do it. They feed each other: every feature we build teaches the harness something, and a better harness builds the next feature faster. We never train the harness on toy problems, and we never build a feature without learning from it.
Results so far: less and less is left for the reviewer
TRAFFIC LIGHT
46
9
needed a human on the real PR
still needed after a follow-up run
DRAW ON SLIDE
30
5
needed a human on the real PR
still needed after a follow-up run
~40s. What's left for the reviewer. On the real traffic light PR, the human review left 46 comments. A follow-up run, blind, got 37 of them right on its own: 9 left. On draw on slide we made 30 fixes by hand. The follow-up run left 5.
Why this project: how far can Claude carry our roadmap?
01 · THE PROBLEM
Engineering time is our tightest constraint
Every roadmap item competes for the same few engineers.
02 · THE BET
Let Claude build, steered by a harness
Guided by everything we know about our codebase, so it builds features the way we would.
03 · THE PAYOFF
Ideas become working prototypes, fast
Anyone on the team can try an idea for real before we commit engineering time to it.
~40s. For anyone new to the project: why are we doing this? The problem: engineering time is our tightest constraint, and every roadmap item competes for the same few engineers. The bet: let Claude build, steered by a harness. A harness is everything around Claude Code that makes it build our way: instructions and rules, hooks that check its work, and reviewer agents. The payoff: ideas become working prototypes fast, and anyone on the team can try an idea for real before we commit engineering time to it. It started as LessonUp Labs: three people, six focused weeks.
Traffic light
Goal: let teachers use a visual indicator when it's time for silence, group work, collaboration, etc.
One of the few components that is still missing in the NLE
Deze slide heeft geen instructies
Draw on slide
Goal: Let teachers draw or write anything on top of their slide, while teaching. Especially useful when using a smartboard with touchscreen.
One of the few missing classic features while in teach mode
Deze slide heeft geen instructies
Share with class
Goal: Let teachers share a lesson directly with their students, managed via a class in LessonUp
Most requested feature in the NLE
First ~150 users got early access this week
Coming release includes UX updates on the modal
Deze slide heeft geen instructies
Product update: Activation insights
Achieve the "aha" moment in the first 7 days
The sooner a teacher experiences the value of LessonUp, the higher their lifetime retention. When this happens in the first 7 days, the activation rate is roughly 40%. Delaying past 90 days drops it steeply to 27%.
More lessons created, means a higher chance of activation
Users who create 5+ lessons reach an activation rate of 46.8%.
Deze slide heeft geen instructies
Guide users
Goal: Nudge users directly to creating or saving their first lesson after registration
A/B test:
Control group: Land in explorer
Variant: Nudge with modal
Deze slide heeft geen instructies
Improve PowerPoint import
Added a feedback form to the import flow
Picked up several improvements
skip placeholder text
import formula as formula component
cropped images
custom shapes
Developed a better way to test and analyze powerpoint imports with the help of AI for future improvements.
Deze slide heeft geen instructies
Request user information
Initial goal: get more information about users that register for our platform.
Next step: Use this information to
Better segment our data
Use it to send relevant lesson material
Use it to pre-fill things like filters in the lesson library or AI tools
Set up as an A/B test
Aim is to launch next week
Deze slide heeft geen instructies
Deze slide heeft geen instructies