Google showcased 10 research projects. Access ranges from a public tool to a waitlist

At its Research@ event in Mountain View, Google summarized 10 research projects spanning geospatial data, scientific agents, technical document review, health, education and collaborative prototyping. They are not 10 equally mature products with the same access model.

Ten demonstrations put AI inside expert workflows

Earth AI is expanding early access with analysis in Google Earth, a planetary prediction engine and an experimental agent-to-agent API. It integrates Google Dataset Search, which covers more than 100 million public datasets, and TimesFM. Google cites more than 70 geospatial tools and is also committing $10 million to organizations working on crisis resilience and public health.

Other projects sit directly in professional work. A multi-agent orchestrator reproduces prior methods and applies them to new scientific problems. Deep Review checks logical consistency and factual grounding in complex documents. Google says an early version has been used to review more than 16,000 research papers, and access is now expanding to the public.

Google is building the portfolio around verification, data and domain context

The common thread is not a universal chatbot. Google places models between an expert and specific working material: satellite images, tables, scientific literature, clinical records or a technical document. The value therefore comes from connections to data, specialist tools and a verification step, rather than merely a longer model response.

The figures reinforce that point. Google says MedGemma has passed 10 million downloads by more than 65,000 unique users. In an AMIE study, 98 patients safely consulted the system before urgent care visits, and clinicians said its summaries helped them prepare in 75% of cases. Groundsource produced a public dataset with more than 2.6 million flood events for a model that forecasts urban flash-flood likelihood up to 24 hours ahead.

One showcase mixes deployed services, studies and promises

Readiness varies sharply. Deep Review is opening to the public, while TimesFM and TabFM are available to Cloud BigQuery users and their Gemini for business integration is still forthcoming. Earth AI remains in early access with experimental features. The classroom simulations have a pilot program, and Play with Putty has only opened a Google Labs waitlist.

Health results need the right scale too. The AMIE study involved 98 patients at one medical center, and Google describes the work as early. The 75% figure records clinicians' assessment of preparation support, not a clinical outcome or broad deployment.

Progress depends on repeatable results outside Google's pilots

Each project now needs a different signal. For Deep Review, it is the quality of errors found beyond computer science conferences. For Earth AI, it is the number of partners putting the experimental API into real decisions. For AMIE, it is the nationwide randomized study. For learning simulations, classroom behavior matters more than the number of generated screens.

Google presented a strong research map, but a map is not a transport network. The durable advantage will emerge where the company provides access, states limitations and enables independent verification.

Lilith's verdict

Google placed ten polished prototypes on stage, but each comes with a different admission ticket. Research leadership starts when an expert gets through the door, uses the tool and someone else reproduces the result.