Omakase: proactive assistance with actionable suggestions for evolving scientific research projects
The AI research assistant monitors project documents to infer timely queries and distill exhaustive reports into actionable suggestions.
A team from the Allen Institute for AI (AI2) and the University of Washington has developed Omakase, a novel AI research assistant designed to solve a critical bottleneck in scientific workflows. Current AI agents lack the context to proactively assist with long, evolving projects, forcing researchers to manually compress their rich project history into short queries for deep research systems. These systems then produce exhaustive reports that are difficult to parse for concrete next steps. Omakase addresses this by continuously monitoring a user's project documents—like notes, drafts, and data—to automatically infer timely and relevant queries, bridging the context gap that plagues standard AI tools.
The system then takes the lengthy reports generated by a backend research engine and distills them into concise, contextualized suggestions tied directly to the user's ongoing work. In their research, detailed in a paper on arXiv, the team conducted studies with 42 participants using a technology probe and iterative prototypes. The results were promising: participants found the AI-generated queries to be both useful and well-timed. Crucially, they rated Omakase's distilled suggestions as significantly more actionable than the raw, unfiltered reports from the deep research system. This indicates a major step toward AI that can truly collaborate on complex knowledge work over time, rather than just responding to isolated prompts.
By automating the query formulation and report synthesis process, Omakase shifts the paradigm from reactive search tools to proactive research partners. It demonstrates how AI can be integrated into the messy, nonlinear process of discovery, helping scientists identify relevant literature, methodologies, or analyses they might have otherwise missed. The work highlights the importance of building AI systems that understand project evolution, a key challenge as we move beyond single-turn chatbots toward persistent, context-aware agents.
- Proactively infers queries by monitoring project documents, solving the AI context problem for long-term research.
- Distills exhaustive research reports into contextualized, actionable suggestions rated significantly more useful by users in studies.
- Developed and tested through iterative prototyping with 42 participants, showing strong user validation for the proactive assistant model.
Why It Matters
It transforms AI from a reactive search tool into a proactive collaborator for complex, long-term scientific and R&D projects.