Challenge
Crash Course wanted weekly educational video at the cadence of late-night television and the production cost of a YouTube creator. Most edu-pipelines could do one or the other. Sustaining weekly across 60+ subjects, each with distinct visual languages, demanded a pipeline that scaled instead of one that bottlenecked at the senior animator.
Approach
A modular vector-asset library, eventually 50K+ assets, organized by reusable scene primitives. A custom branding system that let each subject look distinct without rebuilding from zero. Workflow automation between writers, animators, and shipping queues so handoffs stopped being meetings.
This is the kind of work that, today, would carry the Summit tier label on postreality.ca.
Instrumentation
| KPI | Method | Frequency |
|---|---|---|
| Weekly release cadence | Publishing queue | Weekly |
| Subscriber count | YouTube Analytics | Continuous |
| Asset-library reuse rate | Internal asset DB | Quarterly |
Result
| The Number | 16M+ subscribers |
| Tier | REPORTED |
| Source | YouTube Analytics, public counts |
Plus 2B+ views and the YouTube Diamond Play Button. The pipeline ran continuously from 2012, and stayed in production after the 2024 handoff until the studio’s own exit from the series in 2025.
Impact
The pipeline outlasted the role. It stayed in production for about a year after the 2024 handoff, until the studio’s own exit from the series in 2025, which is the cleanest measure that the system worked: a successor team could run it without rebuilding it.
Updated August 2026. Thought Café’s farewell puts the studio’s run on the series at 2011 to 2025 and says the show continues without them. That bounds this claim rather than extending it. Whether the incoming team still runs this pipeline is not something we can source, so the claim stops at 2025.
What we’d do differently
Today the same pipeline would carry an embedded knowledge graph for asset semantics and an LLM-assisted scene-composer. The 2014 asset library was tagged by hand and stayed accurate until volume outran the taggers. A graph-backed semantic index would have bought another five years of asset reuse before the next architecture pass.
