Inconsistent feature parity and feedback channels across Claude surfaces (Code, Cowork, claude.ai)
Summary
There are significant, seemingly arbitrary differences in capabilities, UI, and feedback mechanisms across Anthropic's Claude surfaces (Claude Code, Cowork, claude.ai). The same user doing the same work hits different affordances depending on which surface they're in, with no clear rationale. This is a recurring source of friction.
Specific instances observed (one user, one project, one day)
- No trackable bug submission from Cowork. In Claude Code, issues can be filed to GitHub via
ghand tracked by number. In Cowork, the only feedback path is a thumbs-down + freeform comment, which produces no trackable artifact — the user cannot reference, follow, or verify it was received. (This issue itself had to be filed from Code, not Cowork.) - Usage check is command-driven in some surfaces, 3-click menu in Cowork. See #71940.
- Memory is not shared across surfaces. See #71939.
- Model default behaves differently / is overridden. See #71934.
The underlying problem
These read as independent symptoms of one root cause: the surfaces are developed without a shared spec for core cross-cutting concerns — feedback/bug reporting, usage visibility, memory/preferences, model selection. Users experience this as "why does the same thing work differently here?"
Expected behavior
- A consistent, trackable feedback/bug-reporting channel available from every surface (thumbs-down should generate a referenceable ticket the user can follow, or surfaces should expose a direct issue-filing path).
- Core cross-cutting concerns (usage, memory, model selection, feedback) should have parity across surfaces unless there's a documented, intentional reason for divergence.
Why this matters
Users build mental models and workflows. Arbitrary cross-surface divergence breaks those models, forces relearning, and makes the product feel like several disconnected tools rather than one coherent assistant.
Environment
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗