Hold-Open Research Notebook
A research and writing assistant for journalists, analysts and nonfiction writers that holds open threads alive across weeks and refuses to draw a conclusion until the writer has actually tested it — flagging questions that were prematurely closed.
Long-form journalists, analysts and non-fiction writers working on multi-week investigations
- Open-question ledger tracking every unresolved claim in the piece
- Source-gap detector that flags claims backed by only one source
- Auto-pushback when the writer drafts a conclusion the notebook has not yet earned
- Export-ready audit trail to show editors which open questions were considered
The curiosity paper's central insight — that an agent's inquiry policy can be tuned to value keeping questions open over collapsing them — is exactly what long-form writers say they wish their AI tools did.
Long-form journalist + analyst workflows are a real, recurring search category ('best AI tools for journalists' articles list 6+ tools including NotebookLM, Elicit, Scite), but the specific 'refuse to close questions prematurely' methodology is a niche pain most users don't articulate as a need.AI for Journalists: We tested the best 6 tools - Saner.AI ↗Best AI Tools for Journalists (2026): Research, Writing & Data ↗
Crowded field: NotebookLM, Perplexity, ChatGPT/Claude, Elicit, Scite, Semantic Scholar all positioned for investigative research, plus the unrelated 'Curiosity Workspace' agent product; differentiated position on epistemic discipline is thin and easily replicated as a system prompt by incumbents.Best AI Tools for Journalists (June 2026) - aipedia.wiki ↗Curiosity - AI Agents ↗
Proven willingness to pay ~$8-20/mo for research AI (NotebookLM Plus $7.99, Pro $19.99), but target buyers are individual reporters and small newsrooms with tight budgets, capping ARPU and limiting enterprise expansion.NotebookLM Pricing 2026: Free vs Plus vs Pro vs Ultra ↗
Deep research for journalists is a durable need, but the 'hold-open threads' mechanic is essentially a prompt-level policy easily folded into frontier assistants (ChatGPT Projects, Claude Projects, NotebookLM), creating constant feature-parity risk.NotebookLM Pricing 2026: Free vs Plus vs Pro vs Ultra ↗
Buildable today as an LLM wrapper with persistent thread storage and a 'hold-open' agent policy; main engineering lift is the cross-week state model and source provenance, both tractable rather than research problems.