Prompt engineering leading AI software design innovations

Ask John Montgomery, Microsoft corporate VP of product, how he thinks AI is affecting his work, and he’ll tell you that he’s bullish on the technology—particularly GitHub’s programming helper Copilot.

“You can have GitHub Copilot from me when you pry it from my cold dead hands,” Montgomery told us during a recent interview. “It is the most amazing productivity booster for writing code that I have experienced.”

While that praise should be taken with a grain of salt—GitHub is a subsidiary of Microsoft—there’s no denying the impact that AI is having on programming and software development.

“The technology is amazing at its ability to kind of understand natural language and understand very complex intent, and to be able to reduce that to something that is immediately actionable,” Montgomery said. “It is a game changer.”

Universal. At the GitHub Universe conference in November, GitHub CEO Thomas Dohmke told reporters that the use case of the new developer is that it will open the door to more innovation.

Programming is experiencing a revolution that’s due in part to how AI can erase a level of repetitive labor and allow skilled and amateur developers alike to turn abstract language into real code. Prompt engineering—using AI to generate code based on natural language requests—is making room for new coding possibilities, Dohmke said.

“Natural language gives humans a much better way of expression than any kind of invented programming language, which always will have a much shorter vocabulary and much more defined way of doing things,” Dohmke said. “And so we believe that natural language will unlock a new wave of software developers and a new wave of connectivity for existing developers.”

Natural language. The levels of abstraction in Copilot allow users to continually adjust their code through prompt engineering, GitHub VP of product Mario Rodriguez told us, translating natural language into procedures and refining the process in a dialogue with AI rather than line by line.

“You’re not changing the code, you’re just changing the plan,” Rodriguez said.

By using natural language prompts to generate code, engineers are now working in an environment that’s going to present opportunities for software development.

“There is significant amount of data science and prompt engineering that goes into it; a significant amount of work, getting the right context for those prompts to them work, there’s a significant amount of work also on our AI—which we call responsible AI—which is tuning models and filters to do the right thing,” Rodriguez said. “There’s a significant amount of work and enterprise controls.”

Build up. It took a long time for prompt engineering to get here. Mike Hanley, Cisco CSO, told IT Brew that the technology is finally living up to the promise of the prototype he saw two years ago.

“I remember the first time I saw Copilot for code completion, in May of 2021,” Hanley said, adding, “It was one of those moments where you say, ‘I want to see that again; I want to make sure that was actually what I thought was happening.’”