Detailed Analysis
A user on the Anthropic subreddit has surfaced a notable usability gap in Claude's Projects feature, specifically on the mobile app. The user reports entering substantial text into a "What are you trying to achieve?" onboarding prompt when creating a new Project, only to later discover that this text is inaccessible for editing or retrieval — neither by the user nor by Claude itself. The user's intent was to repurpose that content as formal Project Instructions, but the inability to access the field has left the text effectively stranded within the interface. The complaint highlights both a discoverability problem and a potential data accessibility issue within Claude's Projects architecture on mobile.
The Projects feature, introduced by Anthropic to allow users to organize conversations with persistent context and custom instructions, relies heavily on upfront configuration to tailor Claude's behavior to specific workflows. The distinction between an onboarding prompt like "What are you trying to achieve?" and the more structured "Project Instructions" field appears to be poorly communicated to users during setup. If the former is used to generate an initial context but is not subsequently editable or even readable, users who invest significant effort into that field may find themselves unable to iterate on or migrate that content — a frustrating outcome for a feature designed to enhance productivity.
The mobile-specific nature of the complaint is significant. Mobile interfaces for complex AI tools frequently lag behind their desktop counterparts in terms of feature parity and editing affordances. If the "What are you trying to achieve?" field feeds into Claude's system-level context but is rendered read-only or invisible after project creation, this represents a meaningful UX regression that could erode trust in the Projects feature among mobile-first users. The inability for Claude itself to surface or quote back that text compounds the problem, suggesting the field may populate a non-conversational layer of the system prompt that is architecturally opaque to the model during chat.
More broadly, this issue reflects a recurring tension in conversational AI product design: the challenge of making powerful, configurable backend systems legible and editable to end users. As Anthropic and its competitors invest heavily in agentic and project-based workflows, the ease with which users can inspect, revise, and migrate their configuration data becomes a meaningful differentiator. Features that silently consume user-generated content without providing retrieval or editing pathways risk undermining the transparency and controllability that AI safety-conscious developers like Anthropic publicly champion. Addressing these friction points — particularly on mobile — will be essential as Projects matures from an early-access feature into a core part of the Claude product experience.
Read original article →