The launch of Gemini Code Assist, Google's new developer assistant, represents a bold move by the tech giant to unlock value for developers and get ahead of competitors when it comes to AI productivity offerings.
Powered by its powerful new Gemini 1.5 Pro Large Language Model (LLM), Gemini Code Assist was one of the standouts. ads in Google Cloud Next 2024.
There's a real sense that Google Cloud has found a great application here, with the model's million-token context window allowing it to generate results based on a company's entire codebase for context.
Google Cloud has used the event to announce several services that compete with Microsoft's Copilot approach to AI assistants. Among them stand out their Vertex AI Agentsnatural language assistants that companies can use and train with their own data to implement them in specific tasks.
But Gemini Code Assist stands out as a truly unique offering on the market.
While the Gemini 1.5 Pro's accuracy will come down to benchmarking, its gigantic context window far surpasses competitors like GitHub Copilot. OpenAI GPT-4 Turbo, which underpins Microsoft's strategy Pro Copilot offering, has a context window that maxes out at a comparatively small 128,000 tokens.
This gives Google Cloud a competitive advantage, with the Coding Wizard able to generate and translate code into a simple one-time query output.
Gemini Cloud Assist supports more than 20 programming languagesincluding Java, C++, SQL, Pitonand PHP. In addition to code generation, it is capable of large-scale code translation, and the context window also allows companies to convert swaths of source code or even entire code bases to another programming language.
In conversation with Darren Mowry, General Manager of AI Startups at Google Cloud, ITPro discussed the inherent business value of reducing coding time and simplifying the code translation process within AI peer programmers.
“Most founders are technologists,” Mowry began, pointing out the clear market Google Cloud has among startups: About 60% of generative AI startups are Google Cloud customers, compared to 50% of the last year.
“I remember having a big discussion with a founder years ago, who made a comment to me of 'I can always raise more money. I can always hire more people. I can't invent more time'.
“So anything you can do to reduce time, allow me to develop a product faster to differentiate myself more quickly and expand my product set… If you can help me do those things, I'll spend time with you. If not you're going to help me do those things. I have no time”.
Gemini Code Assist has already delivered tangible benefits
Google has already used Gemini Code Assist internally to increase the productivity of its own developers and reduce work, while customers given early access to the tool also reported significant gains.
For example, Turing, the AI-powered technology services company, is already using Code Assist for internal development, providing its workers with coding suggestions based on their own code. and as a result it has measured a 30% improvement in productivity.
“Developers were able to configure their environments 55% faster than before, there was a more than 48% increase in unit test coverage for code, and 60% of developers reported that they could now focus on more satisfying work.
In a masterful demo, Paige Bailey, product manager for generative AI at Google Cloud, migrates a customer-facing web feature based on a brief from the design team.
Bailey said it would typically take a developer days to begin familiarizing themselves with their company's code base, which can consist of more than 100,000 lines of code, and even longer to complete the task of migrating web functionality and verifying their configurations. associated microservices.
“This is because Gemini code transformations with full knowledge of the code base allow us to easily reason across our entire code base,” Paige told the audience.
“By comparison, other models can't handle anything beyond 12,000 to 15,000 lines of code.”
“We're also finding that for a lot of developers, it's really difficult if they try to afford pair programming, which is a good practice,” said Caroline Yap, MD, Global AI Business at Google Cloud.
“But now, when you can have the AI also as a programmer with you and you can also suggest, 'oh, this is how you would write a unit test' or it takes away the work of being a developer so you can focus on building the experiences that you are trying to build as a developer compared to all other things.”
Brad Calder, vice president and general manager of Google Cloud, highlighted that Gemini Code Assist can work with code bases hosted anywhere.
“On premises, GitLab, GitHub, GitBucket, and even across multiple repositories. We don’t require your code to be only on GitHub.”
Calder also noted that “the other leading provider only supports repositories in one country, which may be inappropriate for companies that have strict data residency requirements.”
This is another way in which Gemini Code Assist acts as a broad competitor to other peer programmers on the market.
Potential in the legacy codebase space
With the ability to process and even translate thousands of lines of code in one go, Gemini Code Assist has great potential for companies looking to migrate and update legacy and obsolete code bases.
When ITPro Raising this topic at a Google Cloud Next roundtable, Yap and Nenshad Bardoliwalla, Director of Product Management for Vertex AI at Google Cloud, agreed that legacy code translation is a topic of great interest across the industry.
“We have a very interesting situation in the industry,” Bardoliwalla said.
“People who know how to operate these systems are dying. They literally won't be here anymore and there is no one who can help you understand business rules, data semantics, etc.
“Try finding a COBOL engineer. It's almost impossible”.
But while both recognized the potential of tools like Gemini Code Assist for translating code bases, Bardoliwalla cautioned against the idea that legacy code translation can be achieved solely through LLM.
“Language models help in several use cases… some of the people working on our mainframe monetization programs at Google Cloud have found LLMs useful, but there are other technologies that allow us to migrate in a very simple way.” elegant”.
Bardoliwalla emphasized the need for data processing routines and semantic analysis of the code alongside raw translation, while confirming that customers are very committed to the idea of mass code translation.
It's clear that Gemini Code Assist has an important role to play in addressing the pressing legacy problem, along with the increases in developer productivity it can enable.
