Detailed Analysis
The article surfaces a practical workflow gap that has become increasingly common as knowledge workers adopt both Claude.ai and ChatGPT web interfaces for parallel or comparative use: neither platform offers native, editable, syncable project file storage. The author describes a familiar pain point among power users of "projects" features (Anthropic's Projects and OpenAI's equivalent) — while both allow file attachments to give persistent context to a chat, neither allows in-place editing of those files. Any update to a planning document, spec, or reference file requires downloading it, manually reconciling changes, and re-uploading to both platforms separately. This friction compounds when working across two ecosystems simultaneously, since keeping parity between Claude's and ChatGPT's project context means duplicating the same manual process twice. The author notes the predictable failure mode: context silently drifts out of date because the friction of updating discourages frequent syncing, leading to both assistants working from stale, months-old information without any error or warning.
This matters because it exposes a structural limitation in how consumer-facing AI chat products handle persistent knowledge versus how developer-facing tools have evolved. Tools like Claude Code and Codex (explicitly excluded from the author's complaint) already solve this problem by reading directly from a repository, meaning any file changes are automatically visible in the next session with no manual re-upload. The web chat products, by contrast, treat "project files" as static, uploaded blobs disconnected from any live source of truth. This gap reflects a broader design philosophy tension in AI products: the browser-based consumer chat interface prioritizes simplicity and sandboxing, while developer tools prioritize live, bidirectional integration with existing file systems and version control. For users doing conceptual or planning work — not code — in these chat UIs, this leaves them stuck with the more primitive workflow despite doing intellectually similar context-management work.
The workaround described — using a git repository as a shared, versioned source of truth that both AI assistants clone into their working environment at the start of a session — is a clever repurposing of software engineering infrastructure for general knowledge work. Git provides exactly the properties missing from native project file features: version history, diffing, a canonical source of truth, and a mechanism for both platforms to independently "pull" the latest state without the user manually coordinating updates between two separate silos. By having the AI assistants themselves handle the "git add, commit, push" cycle after briefing the user on changes, the author has effectively built an ad hoc synchronization layer on top of two products that don't sync with each other or with any external system natively.
This anecdote reflects a broader trend in the AI assistant space: as these tools move beyond one-off Q&A into sustained, project-based collaboration, the absence of robust context management infrastructure becomes a major friction point for serious users. Anthropic's introduction of Projects, OpenAI's similar feature, and the growing ecosystem of memory and RAG (retrieval-augmented generation) features are all attempts to solve persistent context, but they remain siloed per-platform and often static rather than live. The interest in this kind of git-based workaround signals demand for features like live file-system or repo integration, bidirectional editing, and cross-platform context portability — capabilities that would treat "context" as a first-class, synchronized resource rather than a one-time upload. As multi-model workflows (using both Claude and ChatGPT together) become more normalized, pressure will likely grow on both companies to either open up their project file systems to programmatic, syncable access (via API or connectors) or risk power users continuing to route around them with external infrastructure like git, cloud drives, or third-party sync tools.
Read original article →