A chat model turns a prompt into an answer. A research agent (using the same OpenAI model) can find primary and hard to access documents, download files, write and run code on them, pass outputs between programs, and save the data and calculations for later questions. As an example, it turned six official Netflix viewing reports into 98,586 rows across 29,785 titles, giving us far more information with which to better criticize our current explanations about Netflix's viewership nature. The core capability is that research questions produce a working research folder containing the source files, cleaned data, code and calculations, ready to inspect, update and rerun when new information arrives or fit automatically fit into one's workflows (such as defining a process and have it done automatically across different investment cases). I use this everyday, if you'd like to have someone from the investment field help you with your workflows by properly using agents message me. This is not an investment soliciation, I do not have an investment license. I do, however, want to help you make use of 'AI agents'.