Anthropic's Scion is heavily inspired by Mononoke
Anthropic’s new Scion version-control system is being criticized for closely mirroring Meta’s open-source Mononoke backend for Sapling, with some arguing the data model and terminology amount to a near one-to-one copy. Commenters stress that reimplementing open-source ideas is legally acceptable and historically common, but contend that Anthropic’s marketing overstates novelty and fails to adequately credit specific prior work. The episode is cited as an early example of how AI-assisted development can obscure provenance, weaken attribution norms, and potentially discourage open sharing of infrastructure designs.
Perceived Plagiarism vs. Acceptable Inspiration
- Many see Scion as extremely close to Meta’s Mononoke, beyond normal “inspiration,” with similar data model and terminology.
- The original article’s author explicitly says copying the design is fine; the complaint is about weak attribution and presentation.
- Some argue Anthropic did credit Sapling (of which Mononoke is a component) and that further citation isn’t typically expected.
- Others counter that Scion’s blog post oversells novelty and underplays how directly it mirrors Mononoke.
Attribution, Norms, and Law
- Several participants distinguish legal requirements (licenses allow reimplementation) from social norms (basic decency to credit prior work).
- Others push back: open source was meant to enable reimplementation; demanding extensive credit is seen as personal morality, not obligation.
- There’s mention that patent and copyright risks often lead big companies to discourage looking at prior art, which conflicts with attribution norms.
Role of LLMs and “IP Laundering”
- Concern that LLMs make it trivial to absorb open designs and reproduce them without clear provenance, eroding incentives to share ideas.
- One comment claims Anthropic explicitly had its model analyze the Mononoke codebase and re-engineer it, undermining any “clean room” framing.
- Participants note that agent workflows optimize for outcomes, not literature review or credit, so responsibility must rest with human operators.
Marketing, IPO, and Narrative Framing
- Some see the Scion post as IPO-oriented marketing: “agents scale so fast we had to invent new infrastructure,” boosting perceived valuation.
- Others argue this is also a genuine scaling issue; multiple large companies are hitting VCS limits with AI-driven code volume.
Technical Discussion: VCS and Monorepos
- Mononoke is identified as the server-side of Sapling, combined with EdenFS (virtual FS) and a client (hg-derived).
- Several argue Git is poorly suited for truly large monorepos and high merge rates; Meta, Google, and others created custom systems years ago.
- Some question why Anthropic didn’t just adopt Sapling/Mononoke instead of building a very similar system in-house.
Broader Reflections on Git and Alternatives
- Multiple commenters praise Sapling-style workflows and criticize Git/GitHub as a decade-long local maximum that stalled VCS innovation.
- There is interest in new VCS startups and alternatives (e.g., jj-based systems), with calls for better tooling and Windows support.