Microsoft patents AI tool that writes and runs software tests from text

1. A method comprising: receiving, via a natural language interface, a first user input including a first natural language description defining an expected functionality of a software feature and a second natural language description defining a process for using the software feature; generating, based on the first natural language description and the second natural language description, a first probabilistic model query including a first contextual input defining an expected test case output format; providing the first probabilistic model query to a probabilistic model; receiving, from the probabilistic model, a first model output comprising a plurality of test cases for the software feature in accordance with the first contextual input defining the expected test case output format, an individual test case defining a sequence of natural language steps for evaluating an aspect of the expected functionality of the software feature; causing the plurality of test cases, each defining a sequence of natural language steps, to be displayed via the natural language interface for review and selection; receiving, via the natural language interface, a second user input selecting a test case from the plurality of test cases displayed via the natural language interface; generating, based on the second user input selecting the test case, a second probabilistic model query and a second contextual input defining an expected software implementation output format; providing the second probabilistic model query to the probabilistic model; receiving, from the probabilistic model, a second model output comprising a software implementation of the test case; causing the software implementation of the test case to be displayed via the natural language interface with a request to confirm the software implementation; receiving, via the natural language interface, a third user input confirming the software implementation; executing, based on the third user input confirming the software implementation, the software implementation to perform the sequence of natural language steps, defined by the test case, for evaluating the aspect of the expected functionality of the software feature; extracting log data defining an outcome of the execution of the software implementation; comparing the log data against a success condition defining an expected outcome of the sequence of natural language steps, defined by the test case, for evaluating the aspect of the expected functionality of the software feature; and classifying the execution of the software implementation as passed or not passed based on the comparison of the log data against the success condition.