/Patent Tracker
GooglePatent applicationAI & softwareInterest 6/10

Google patents AI system that tests multiple prompts to fix code errors

This filing describes a system where one AI model creates several different instructions, or prompts, for a second AI model to update old computer code. The system then runs the code migration using each prompt and picks the version that predicts the fewest errors before sending the result to the user.

Why it mattersDevelopers who use AI tools to update large amounts of code might notice this approach, as it aims to reduce mistakes by comparing multiple AI-generated solutions.

In the filing’s own words

A system receives a user query to perform at least one code migration operation on source code and provides a prompt associated with the user query as input to a first AI model trained to generate prompts for a second AI model performing code migration operations. The system obtains prompts for the second AI model from the first AI model, migrated versions of the source code for each of the first AI model’s prompts from the second AI model, and an indication of a number of predicted validation errors for the code migration operation performed by the second AI model based on the prompts. The system selects an optimized migrated version associated with a prompt upon determining that a migrated version has a lowest number of predicted validation errors. The system provides the optimized migrated version of the source code to a client device according to the user query.

What exactly they claim (claim 1)

1. A system comprising: a memory; and a set of one or more processing devices coupled to the memory, wherein the set of one or more processing devices is to perform operations comprising: receiving, from a client device associated with a user, a user query to perform at least one code migration operation pertaining to source code associated with an application; providing a prompt associated with the user query as an input to a first artificial intelligence (AI) model, wherein the first AI model is trained to generate optimized prompts for a second AI model to perform code migration operations; obtaining one or more outputs of the first AI model, wherein the one or more outputs comprise a plurality of prompts for the second AI model; obtaining, for each respective prompt of the plurality of prompts of the one or more outputs of the first AI model, a respective output from a plurality of outputs of the second AI model, wherein the respective output of the second AI model comprises a respective migrated version of the source code from a plurality of migrated versions of the source code and, for each respective migrated version of the source code, an indication of a number of predicted validation errors for the at least one code migration operation performed by the second AI model based on the respective prompt, selecting, from the plurality of migrated versions of the source code, the respective migrated version of the source code as an optimized migrated version of the source code associated with the respective prompt as an optimized prompt based on a determination that the respective migrated version has a lowest number of predicted validation errors based on the optimized prompt; and providing the optimized migrated version of the source code to the client device in accordance with the user query.

Facts

Company
Google (Google LLC)
Official title
OPTIMIZING PROMPTS FOR LARGE-SCALE CODE MIGRATION USING ARTIFICIAL INTELLIGENCE
Publication number
US20260267636A1
Application number
19071173
Filed
Mar 5, 2025
Published
Sep 10, 2026
Inventors
7
Claims
11

Classified under

G06F 8/71 Physics & computingG06F 8/33 Physics & computing
Read the full filing (official PDF) ↗ Open in the Patent Tracker

More from Google in AI & software

Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.