Claude Projects vs Sup AI

“The primitive here is a named, persistent system-prompt-plus-document-store scoped to a workspace — which is genuinely the thing developers have been duct-taping together with system prompt files committed to git and copy-pasted on every new chat. The DX bet is 'make the right thing the default thing': instead of building a wrapper that injects context programmatically, Anthropic just made the UI do it natively. The gap is API parity — if Projects context doesn't flow through the API with the same scoping, developers will still be hand-rolling this, and that's the specific thing I'd want confirmed before calling this a full ship.”

“The HLE claim needs independent verification, but the underlying ensemble approach is architecturally sound for factual Q&A tasks. Running 339 models is expensive — pricing will be the gating factor for production use. The $10 free credit is a fair trial.”

“The direct competitor is ChatGPT's Custom Instructions plus Memory, which has had persistent context for over a year — so Anthropic is catching up, not leading. The scenario where this breaks is team use at scale: shared document libraries with no versioning, no access controls beyond plan-level sharing, and no audit trail mean the first time a team's shared prompt gets silently edited and causes a bad output, trust collapses. What kills this in 12 months isn't a competitor — it's Anthropic itself shipping a proper API-native version that makes the UI feature redundant for the power users who care most about it.”

“Extraordinary claims require extraordinary evidence. A 7.41 point jump on HLE via ensembling — without publishing methodology — smells like benchmark gaming. The latency of running 339 models in parallel is also a real concern for anything other than async research tasks.”

“The job-to-be-done is sharp and singular: stop re-explaining yourself to Claude every time you start a new conversation. Onboarding is as fast as it gets — create a project, paste your instructions, upload a doc, done, under two minutes to value. The product opinion baked in here is correct: most users don't need a memory graph or semantic search over past conversations, they need a stable persona and a document library, and Claude Projects makes exactly that bet without over-engineering it. The gap between shipped and needed is team permission controls — right now it's blunt-instrument sharing, and that will matter the moment any organization with more than five people tries to use this seriously.”

“The thesis this bets on: within two years, AI assistants aren't used as one-off query tools but as persistent collaborators with institutional memory, and whoever owns the persistent context layer owns the workflow. The dependency that has to hold is that Claude remains the preferred model for knowledge-work tasks — if GPT-5 or Gemini Ultra pulls far enough ahead on capability, users don't move their Projects, they just stop opening the tab. The second-order effect nobody is talking about: shared Projects make Claude's system prompt a team artifact, which means prompt engineering starts being treated like documentation — owned, versioned, and argued about in PRs. That's a genuine shift in how organizations relate to AI, and Anthropic is positioning itself as the place where that institutional knowledge lives.”

“Model ensembling is an underexplored direction in the race to reduce hallucination. If Sup AI's approach scales, it could be more durable than fine-tuning individual models — you get the wisdom of the crowd across model families, training data, and architectures simultaneously.”

“For creative work, ensemble outputs tend to regress toward the mean — you get the most-agreed-upon version of something, which is usually the least interesting version. This is a tool for factual accuracy, not creativity. I'd stick with a single strong model for writing.”