thcThought Control

Scrimba

Interactive coding screencasts: pause the video, edit the teacher's code.

scrimba.com ↗docs ↗Format announcement (2017) ↗Launch HN (YC S20) ↗MDN partnership ↗Imba (the language Scrimba is built in, MIT) ↗

language
Imba (compiles to JavaScript), front end and back end
backing
Scrimba AS (Oslo), co-founded by Per Harald Borgen (CEO) and Sindre Aarsæther (CTO, creator of Imba); Y Combinator S20; earlier backers listed on v1.scrimba.com/about: StartupLab, Nordic Makers, Alliance Venture, Amasia. Funding totals differ between aggregators (GetLatka/Caplight ~$0.38M; Crunchbase page shows an implausible figure), so not stated.
borrow

The best proof of the teaching model scene plans, but closed and web-only, so take the interaction design, not the product.

Borrow from it: take a specific idea, API or format.

the steelman: the best honest case for it

Scrimba proved, years before anyone said 'agent tours', that the right primitive for teaching software is a recording of events rather than pixels. Because the screencast is a replay of real editor state, the learner can stop at any second and be inside the instructor's actual code, editing and running it, then resume the narration. That one move removes the biggest failure of video tutorials (copying code line by line, setup breakage) and it scaled: a paying multi-path coding school with over a million learners and MDN's endorsement. Its authoring story is also humane: press record and talk while you code; no editing suite. If you want to know whether 'narrate, then let the learner take over' works pedagogically and commercially, Scrimba is the existence proof.

scores
datadriveseeedittimeratetermpxteachmatureopen
UI as data2 scene 5The recording is an event log internally, but proprietary and not exposed as a value anyone can store or diff.
Agent can drive it0 scene 5
Agent can see it0 scene 5
Small, targeted edits0 scene 5
Time travel3 scene 0Learner can scrub, pause and fork-by-editing; human-facing only.
Live data rate0 scene 5
Terminal native0 scene 5
Pixel graphics4 scene 4
Teaching5 scene 3Built around exactly record, narrate, take over, rewind.
Maturity4 scene 1Years in production, large learner base, MDN partner.
Openness0 scene 4Fully proprietary.

Scrimba scene now scene planned

how agents use it

How

None today. No MCP server, API or documented format; an agent could only drive scrimba.com through a browser like any user.

Wiring it in

Not feasible beyond browser automation; recording/authoring is a human-in-browser flow.

Seeing the result

Only via screenshots/DOM scraping of the web app.

Small edits

No addressable edits for an agent; a human edits code in the paused scrim.

History

Human scrubbing and rewind of the recording; nothing exposed to an agent.

In short

The best model of 'teacher narrates, learner takes over', but a sealed product with no agent surface.

architecture

A scrim is a recording of editor and browser events plus the instructor's audio, not a pixel video. Playback re-creates the instructor's actions in a live in-browser IDE in the same environment, so at any moment the learner can pause, edit, run and copy the code. The IDE and the player are one tool, built in Imba. Execution is client-side (browser), which is why backend topics were hard to support early on.

performance

No published numbers. The 2017 post argues file-size and searchability advantages qualitatively.

adoption and upkeep

Adoption

CEO's LinkedIn describes a 1.5M+ learner platform; the 2020 Launch HN reported 3,000 paid enrolments in the first career path. MDN chose Scrimba as its recommended course partner in 2024.

Used by: MDN Web Docs (recommended course partner for the MDN Curriculum)

Maintainability

A going commercial concern with steady content output, but nothing an outsider can maintain or fork.

Releases: Continuous SaaS deploys; no public versioning. A v2 platform replaced v1 (v1.scrimba.com still reachable for legacy recording). · Recent: Active: 2026 articles, AI Engineer path, AI tutor built into Pro (per scrimba.com articles). · Bus factor: medium: small private company; the platform rests on a niche in-house language (Imba) whose author is the CTO.

weaknesses
and thc-scene

Overlap

The teaching-tour concept: a narrated, replayable sequence where the learner can take over mid-stream and rewind.

What scene would be reinventing

The record-events-not-pixels format and the pause-and-take-over interaction; Scrimba already shipped and validated both, for code in a browser.

The gap it leaves

An agent as the live narrator in a terminal, with a documented, open state/event format the agent can read, seek, fork and replay; Scrimba has no agent surface, no terminal and no open format.

What to borrow

  • Make 'take over' a fork of the recorded state at the playhead, and let 'resume' either discard the fork or diff it against the instructor's path.
  • Audio/narration track keyed to the same timeline as UI events.
  • Authoring is just 'record while you talk'; derive the tour from a session trace rather than hand-writing steps.
sources
  1. Why we're creating a new video format for code (Per Harald Borgen, 2017) medium.com
  2. Launch HN: Scrimba (YC S20) news.ycombinator.com
  3. How Scrimba came to be scrimba.com
  4. MDN partners with Scrimba developer.mozilla.org
  5. About Scrimba (v1) v1.scrimba.com
  6. Per Borgen LinkedIn (1.5M+ learners) no.linkedin.com
  7. Best AI tools and courses for learning to code (Scrimba, 2026) scrimba.com