Planview development teams use gen AI coding assistants, including GitHub Copilot, that help produce routine and tedious code. The tools also help junior developers by suggesting common code patterns they might not think of. “A lot of this was incremental innovation around the user interface of coding assistants that developers have been using for 25 years,” says Sonnenblick.
He reckons his company has seen a nearly 20% boost in productivity over the last year, and points to two particularly interesting things about the latest generation of coding assistants. One is putting in secure chat, so developers can converse freely with the LLM about their specific coding issues, as opposed to just going to OpenAI where there isn’t necessarily sufficient trust to share issues about proprietary code. The second is the tools go beyond coding assistance. They now use what they learn about a program to help build unit tests.
Richard Sonnenblick, Chief Data Scientist, Planview
Planview
“Achieving complete coverage of your code during testing is a massive challenge in itself,” says Sonnenblick. “And unit tests are too tedious for humans to build reliably. But it’s a perfect job for LLMs that collect information as you write your program.”
While this example offers a peek at how productivity gains can be achieved in software development, what’s even more interesting is how tech companies, including Planview, are jockeying for competitive advantage in what they bring to market. According to IFI Claims, an organization that tracks patent data, one way of finding out where competition is heating up is to look at what kinds of patents people are applying for. IFI Claims writes that over the last five years, applications for gen AI patents have grown at a compounded annual rate of 31%.
Gen AI helps scientists develop new proteins
Another interesting set of use cases can be found, for the time being, in biotech. Sandra Castillo, senior scientist and computational biologist at Finland research organization VTT, is using gen AI to design new protein sequences based on what can be learned from nature. The new sequences are then tested at the VTT lab by using E. coli or other bacterial hosts to express the proteins.
In 2018, DeepMind, now a subsidiary of Alphabet, developed AlphaFold, a deep learning system that learns from a database of existing proteins and predicts their 3D structures. The database now consists of more than 200 million entries and the latest generation of gen AI has enabled improvements to the way that data is used.
