Gitlab Duo

Original Article ↗ Hacker News Discussion ↗

Pricing & Adoption

  • Many self-hosted users feel pressured by repeated tier changes and price hikes (e.g., needing Premium, then Duo Pro on top), calling GitLab “unpredictable.”
  • Some argue enterprises likely requested AI features and that they represent a new revenue stream.
  • Others say pricing now exceeds GitHub for comparable tiers, while differentiation has eroded.

Model Quality & Technical Concerns

  • Several point out Duo currently uses older models (e.g., Claude 2); for some this is a “hard pass” vs direct use of newer APIs.
  • GitLab staff link to an internal epic showing migration plans to Claude 3.
  • Comments note weak code completions and “junior dev”‑level comments (e.g., trivial explanations), and problems in JetBrains integrations.

Perceived Value of Integration

  • Some trial users on self‑hosted Ultimate report Duo adds little compared to standalone AI tools, and creates extra admin overhead.
  • MR/issue summarization and “explain this code” features are described as often inaccurate or fluff-heavy.

Self‑Hosting & Alternatives

  • Multiple users mention switching or considering switches to Forgejo, Gitea, Sourcehut, OneDev, Soft‑serve, or plain git+ssh/gitolite.
  • A question about using local LLMs with GitLab CE gets the answer that AI code lives in the EE tree, so CE support seems unlikely for now.
  • Some see GitLab’s remaining edge as “enterprise self-hostable Git with CI,” but say that gap is shrinking.

AI Training & Licensing Concerns

  • GitLab’s statement that private code isn’t used for training leads to inferences that public code is.
  • Many object that this ignores licenses and attribution requirements; others argue it’s analogous to humans learning from public code and likely fair use.
  • There is discussion of anti‑AI licenses, their legal effectiveness, and whether open‑source licensing still fits the LLM era.

Product Direction, UX, and Trust

  • Several complain GitLab is bloated, with too many half‑finished features while long‑standing requests stagnate.
  • Animations and unclear AI-data disclosures are seen as deliberate marketing choices.
  • Some frame GitLab’s strategy as classic enterprise lock‑in: be the default, not the best, then ratchet prices.