Impact of Github Copilot on Project Delivery

Burned through a perfectly good night of sleep doing a proper deep dive into GitHub Copilot Spaces—and surprisingly, it didn’t end in rage-quitting or writing a passive-aggressive sticky note to Future Me. Turns out, Copilot Spaces might actually be a low-key revolution for development teams. Instead of Copilot tossing out its usual half-helpful guesses like a code fortune cookie (“Your bug will fix itself in three commits”), Spaces allows for curated context—actual code, docs, specs, and custom instructions wrapped into a RAG (or whatever magick it is doing behind the scenes) that make Copilot feel less like a an AI intern and more like that one senior dev who "just knows" (and somehow never breaks staging). The learned context in Spaces stays synced with the repo automatically, which is a minor miracle based on my past cumbersome experience with multiple Copilot sessions. No more dragging and dropping half-broken code snippets into prompts like some kind of desperate AI caregiver trying to keep the old bot lucid. And—perhaps most importantly—no more deciphering documentation written in a now-extinct dialect of Techlish, last spoken by a lead dev wizard who vanished during the Great Rewrite of ‘22. The collaborative angle is where things get really interesting. Copilot Spaces can encapsulate the hard-earned knowledge embedded in a repo—bugs tamed, APIs deciphered, architectural decisions made at 1 AM—and make that wisdom somehow shareable, without added BO or coffee breadth. Suddenly, onboarding someone new into the project doesn't require deciphering cryptic commit messages or performing dark rituals over a README file that hasn’t been updated since the Before Times. It’s like capturing institutional memory in a jar, minus the weird ethical questions. GitHub Copilot Spaces (github.com/copilot/spaces) feels like the missing link—bridging the gap between "Silly AI that helps sometimes" and "digital agent who actually read the manual and isn’t afraid of the legacy folder." Which is honestly a little unsettling. But also extremely useful. Here is an announcement for more tidbits:

I spend a lot of my time now speaking to companies about AI strategy. It's exciting but I find it challenging sometimes because while discussing the amazing potential we can't ignore the societal risks. I'll start a dialog with you today with one example involving the state of software coding: A recent research paper (https://lnkd.in/emyyQsZw) examined how AI tools can enhance developer productivity, focusing on the use of GitHub Copilot at ANZ Bank, a large organization employing over 5000 engineers. The study found that GitHub Copilot led to a significant increase in developer productivity and job satisfaction, helping engineers code up to 55% faster on average. Additionally: - 46% of code is now being written with the help of GitHub Copilot across all programming languages, and up to 61% for Java code specifically. - 90% of developers reported completing tasks faster with GitHub Copilot. - 73% said it allowed them to better stay in flow and conserve mental energy. - Up to 75% of developers felt more fulfilled and able to focus on satisfying work. The authors conclude that AI will likely transform software engineering practices and the developer experience in the coming years. This raises the question, will AI continue to be primarily an effective assistant, or will more advanced tools begin to change the nature of what it means to be a software engineer? An example of a more ambitious AI coding tool is Devin from Cognition Labs (https://lnkd.in/ewAgg-We), described as an engineering "buddy" that can build alongside developers or independently complete tasks for review. While still early, this six-month-old company has generated significant interest and is valued at $2 billion dollars. We can also see open-source projects exploring similar ideas, such as the combination of Wasp and Aider (https://lnkd.in/ehz3UkdZ), which aims to provide an AI-driven development workflow. As AI continues to advance, it's interesting to consider how the role of these tools may evolve in software development. Could we see a progression from AI "buddies" to "mentors" or even "managers"? While the trajectory from narrow AI to more general or "Super AI" is still largely theoretical, it's a fascinating area of speculation. Personally, I find these developments both exciting and thought-provoking. The potential for AI to augment and enhance human capabilities in software development is significant. However, it's also important to consider the potential risks and disruptions these advancements could bring. What about you? Are you more apprehensive or excited about the future of AI in software development? What potential benefits or concerns come to mind? #AI #SoftwareEngineering #DeveloperProductivity #GitHubCopilot #Devin #CognitionLabs #WaspAider #NarrowAI #GeneralAI #SuperAI

"We seek to build upon research on #AI and productivity to better understand how #GenAI changes how people do work. We look at how the release of a GenAI coding tool (GitHub Copilot) changed how developers allocate their efforts to different types of tasks. We find that GenAI leads workers to spend more time on core work activities and less time on managerial tasks. We show two mechanisms drive this effect - workers with GenAI allocate more of their work efforts to things they can do by themselves (and less to collaborative work) and also do more exploration (new projects, new languages, etc.) and less exploitation (existing projects). Further we find the effects are greater for workers with lower ability. Finally, we do a back-of-the-envelope calculation and show that using GenAI allows developers to start coding in languages that have higher wages, leading to a labor market value impact of nearly $500 million (this would likely diminish in the long run). Though our empirical setting is open source software #OSS, we argue, and find evidence, that the results generalize to private work settings as well." Great work from Manuel Hoffmann, Sam Boysel, Frank Nagle, Sida Peng and Kevin Xu. Thanks to Frank for coming to present it at a recent MIT FutureTech lab meeting where I learned about it.

New study out of Massachusetts Institute of Technology on how GitHub Copilot improved developer productivity at Microsoft, Accenture, and "an anonymous Fortune 100 electronics manufacturing company." They randomly assigned 4,867 developers to the control group (business as usual) or a group that got access to Copilot and monitored the productivity of both groups over the course of a year. "We find that usage of a generative AI code suggestion tool increases software developer productivity by 26.08%. This estimate is based on observing, partly over years, the output of almost five thousand software developers at three different companies as part of their regular job, which strongly supports its external validity." Furthermore, "We find that Copilot significantly raises task completion for more recent hires and those in more junior positions but not for developers with longer tenure and in more senior positions." The test goes back to 2022 using a version of Copilot that we would now find quite antiquated! https://lnkd.in/gfbRHgyt