A New Generation Of Artificial Intelligence Is Helping Developers Increase Their Productivity

Software developers are experimenting with artificial intelligence to create humanlike content, from writing to illustrations.

‍

There has been a lot of progress made in generative AI since OpenAI pioneered the technology in the ChatGPT writing system and the Dall-E visual illustration generator, both developed by the lab behind OpenAI. The development of these tools, with their impact on search and other core tasks, has sparked growing interest since late last fall, with a wide range of potential applications. 

‍

Software development could be upended by technology in a particularly profound way. ChatGPT is being integrated into Microsoft's Bing search engine, and the OpenAI technology is being integrated into Microsoft's Copilot, an AI system for writing code. Microsoft invested $1 billion in OpenAI in 2019 and pledged as much as $10 billion more in January. 

‍

This tool, along with other tools produced by Amazon.com Inc., International Business Machines Corp., and startups like Tabnine Inc. and Magic AI Inc., has quickly found a receptive audience among developers seeking to boost their productivity.

‍

In contrast to ChatGPT, which some technology leaders have questioned its usefulness in the enterprise, AI coding tools like Copilot, which were first launched for public use last summer by Microsoft-owned coding-collaboration platform GitHub Inc., seem to have the ability to solve some real-world business problems right away, when compared to ChatGPT, which has been questioned by some technology leaders. This AI model was trained on data gathered from GitHub, one of the world's most popular open-source communities where developers contribute and share their code, and serves as the basis for Copilot's AI model.

‍

These tools are not designed to replace developers, according to executives at GitHub and other companies. Instead, they are meant to help developers improve their productivity, much like spell-checking and auto-completion of phrases can help people write documents more effectively. They assist developers by suggesting new code snippets and tests, as well as providing technical recommendations within the code-writing programs that they already use to write code. 

‍

Hyland Software Inc.'s chief innovation officer, Sam Babic, does not expect Copilot and similar tools to actually create enterprise software on their own. “They serve as a second set of eyes,” he said.

‍

Using Copilot has allowed Jonathan Burket, a senior engineering manager at language learning app maker Duolingo Inc., to save 25% of his time by eliminating the need to write code documentation and test cases. “The software is actually capable of understanding the algorithms that need to be run, what checks you are missing, and what things you want to test,” according to Burket. 

‍

Among the appeals of this process is the promise of a faster, less manual software development process, which can, in turn, help companies relieve the pressure caused by a lack of engineering talent, according to some industry analysts and technologists.

‍

"We've been begging for talent for a really long time now," said Chris Taylor, a senior enterprise architect with United Parcel Service Inc.-owned Coyote Logistics LLC, a transportation and shipping company. The company's engineers have been able to focus on higher-level questions like thinking about a system and how that applies at a larger scale without having to worry about having extra developers around,” according to Taylor. “Throughout the company's nearly 350 engineers, Copilot licenses have been purchased for all of them.”

‍

It is still true that some technology leaders are approaching broader adoption with an eye toward possible cybersecurity and intellectual property risks associated with it. Despite the business benefits, some say they'll need to address a range of potential problems, including data sharing compliances and writing industry-specific code.

‍

Last week, GitHub released a new version of Copilot for Business that includes the ability to manage users and run on virtual private networks.

‍

According to Mr. Dohmke, another feature of the tool is that it does not suggest code to engineers that might be under license from another company, helping to mitigate the legal risks associated with the development process. According to a proposed class-action lawsuit filed by GitHub in November, Copilot doesn't give credit to the authors whose code is used to generate its results, according to the complaint.

‍

Aside from a stricter blocking of code suggestions that might include known cybersecurity vulnerabilities, Copilot's business version also includes stricter blocking of code suggestions, Mr. Dohmke added, an issue that CIOs are becoming increasingly aware of as they become more aware of the importance of integrating security into enterprise application development.

‍

As part of Amazon's AI coding tool CodeWhisperer, Amazon has also announced plans to develop a version for enterprises, with security and access controls that are tailored specifically to them, according to Swami Sivasubramanian, Amazon Web Services vice president for database, analytics, and machine learning. The companies also announced on Tuesday that AWS had expanded its partnership with Hugging Face, the maker of machine-learning tools. The companies say the partnership will make it easier for developers to create generative AI applications such as Copilot or ChatGPT.

‍

Among the leaders at Hyland Software, a group product manager named Tiago Cardoso said that the company is trepidatious about deploying Copilot to its engineers without more customizations in place.

‍

As a result of the tightening economy, Hyland Software has decided to hold back on spending on any new tools or technologies for the foreseeable future.

‍

Mr. Cardoso said Copilot and other AI coding tools are still too new, “so we are reviewing every tool, and subscription and going for the essentials.”

‍