Google’s AI Programmer

Google wants to take the grunt work out of coding.

Sign up to uncover the latest in emerging technology.

In its fight to remain relevant in AI, Google may be adding a multi-tool to its arsenal.

The company wants to patent a system for automating “certain aspects of computer programming” using an AI model. Essentially, this tool watches a programmer write code and makes suggestions for edits or tools that will help them in real time, based on what the programmer is intending to do.

Using machine learning, Google’s system tracks the changes a programmer is making to source code. If the programmer makes certain changes repeatedly, such as changing one variable or function over and over again, the system will determine that they intend to make those changes throughout. The system then takes into account repositories of source code transformations and tools made by other programmers to find the best fit for the job. 

If you’re not a coder, think of it like this: Imagine you’re building a table and aren’t sure which tool from a massive toolkit is best fit to tighten a bunch of tiny screws throughout it. A tool like Google’s might suggest that you use a power drill, instead of a screw driver, to expedite the process. 

Google said in its filing that this tech can automate the tedious, repetitive tasks that often come with coding. While macros can often be individually created to automate repetitive tasks, creating those tools themselves “can be cumbersome and/or require considerable expertise/resources,” the company noted. 

Plus, Google noted, “many such tools may not scale outside of a particular context.” Its tool fills this gap, acting as a jack of all trades. 

Photo via the U.S. Patent and Trademark Office.

Google adding another piece to its hoard of AI tech is about as surprising as a cactus growing in the desert. The company has been touting new AI tools on the horizon for several months, including AI integrations throughout the Google Workspace suite, enhancing its search engine with AI, and merged its two major AI divisions, DeepMind and the Brain Team, into a single supergroup. And with practically the entire internet at its disposal for training data (at least according to its privacy policy update), the company shows no signs of stopping.

This patent in particular adds to previous filings showing Google’s interest in AI-assisted development. The company recently sought to patent a machine learning model that can create a user interface, and an AI tool that can develop and publish a “viable running app,” both from natural language descriptions. Though the tech in the latest patent would likely be put to use by programmers, while the others would be for a no-code crowd, these filings all together could hint at a suite of AI-based development tools in the works. 

But Google faces some steep competition, said Kevin Gordon, co-founder of AI consulting and development firm Velora Labs. AI-assisted coding products already exist in the market, and more are likely to come, especially from competitors like Microsoft, Meta and OpenAI, Gordon said. Microsoft, for example, owns GitHub, which operates an AI coding assistant called GitHub Copilot. Meanwhile, OpenAI launched Code Interpreter earlier this month. 

Gordon wouldn’t be surprised if this tech and others like it end up being fought out in court, he said. “The legal side is probably where you’ll see a lot of the actual battles taking place. Patents like this will be important to defend the play.” 

But if Google’s tech does reach developers hands at some point, Gordon can “definitely see it being a big deal,” he said. A large language model that can keep track of the minutiae across software repositories will likely do a better job at it than even an incredibly skilled programmer, he said. Adding a tool like this as a co-pilot for programming teams could allow for faster software development with less of a headache.  

“There’s the people who have that 10,000 foot view of a repository, and then there’s the people who are really detailed,” said Gordon. “Here, you have a (large language model) that can kind of work at both levels.” 

Have any comments, tips or suggestions? Drop us a line! Email at admin@patentdrop.xyz or shoot us a DM on Twitter @patentdrop. If you want to get Patent Drop in your inbox, click here to subscribe.