Artificial intelligence (AI) coding tools now account for half the code running in production environments but, more troubling for platform engineers, the size of the average pull request has now doubled, according to a forthcoming report from DX, a provider of an engineering intelligence platform.
Speaking this week at a PlatformCon New York event, Justin Reock, deputy CTO for DX, also told conference attendees that a preview of the report that DX is sharing also confirms that there is now much more volatility than ever as change failure rates continue to rise.
Software engineering teams at the same time are not seeing the 10x gains in productivity that business leaders have come to expect as the hype surrounding AI remains high, noted Reock. The main reason for this is that coding only accounts for 14 to 16% of the time application development teams spend building software, he said.
Many software engineering teams are simply not as AI ready as they need to be because the workflows that software engineering teams rely on to deploy code have not yet evolved. In fact, too many of those teams are overly focused on activity levels and the speed at which code is being written versus the actual impact AI coding is having on the rate at which higher quality applications are being deployed, said Reock.
Less clear is to what degree the rise of AI might ultimately drive more organizations to embrace platform engineering. Originally defined as a set of best practices for managing DevOps workflows at scale, many organizations have been trying to strike a balance between the need for operational stability and the application development environment preferences of individual developers. In an era, however, where more code is being written by AI tools rather than by a developer, there may be an increased need to centralize the management of tooling.
At the same time, the definition of who is considered a developer is expanding. Many end users are now expressing an intent that generates a prompt, which in turn creates code that automates the original intent. Before too long, a much larger percentage of that code will be moving through DevOps pipelines.
Regardless of who or what wrote the code, each software engineering team will need to revisit the metrics used to track productivity. While many of the traditional core metrics remain relevant, there is now an additional set of agentic AI metrics, such as utilization, impact, and cost, that need to be tracked, noted Reock.
Of course, the degree to which software engineering teams have historically tracked metrics has varied widely. Many software engineering teams are also now too overwhelmed by the amount of code being created to review it all, much less understand what impact that code is having on the organization after it has been deployed. Eventually, however, as AI becomes more deeply embedded within DevOps workflows, the level of granular insight into exactly how each line of code impacted not just the application environment, but also the business, will become much more visible than it’s ever been before.