Pfizer agrees to integrate the Kosmos AI assistant into drug research and development

Pfizer agrees to integrate the Kosmos AI assistant into drug research and development

The system will compare scientific literature with Pfizer’s proprietary data in oncology, immunology and other fields. On 29 October 2026, its developer, Edison Scientific, announced the agreement: findings from previous drug development programs are intended to help scientists choose their next experiments.

Edison’s head, Sam Rodriques, says:

Pfizer has spent decades collecting data on what works and what does not in human biology.

In 2025, Pfizer invested 10,4 billion dollars in internal research and development. Drug development generates laboratory records and findings from experiments and human trials. In Edison’s assessment, these data remain scattered across teams and computer systems.

The developers identify the work on the Robin AI system as one of the foundations of Kosmos: Robin proposed experiments and analyzed the results, while scientists conducted the experiments. According to the developers’ description, a scientist gives Kosmos a research question and data. The system searches papers and patents, then writes and runs code to analyze the data. Using the literature and experimental results, Kosmos formulates a hypothesis and predicts what the next experiment will show.

Scientists can follow the calculations, check the conclusions and adjust how Kosmos proceeds. After a new experiment, they can compare the result with the prediction and revise the hypothesis. The system retains its calculations and conclusions between interactions, allowing researchers to add data and continue the same investigation.

The agreement covers Pfizer’s research and development as a whole. Edison’s scientists and engineers will work with Pfizer teams to integrate the system into the company’s workflows. The developer intends for the findings from one program to serve as input for another: information about a successful or unsuccessful experiment would help the next team choose a hypothesis and an experiment.