Human+AI Pair Programming Group

This is the homepage of the Human+AI Pair Programming research group at Freie Universität Berlin. We are part of the Software Engineering Group in the Informatics department.

For more information, please get in touch with Linus Ververs or Lutz Prechelt.

1. Goals and Approach

1. Our goal as software engineering researchers is to understand AI-assisted development in such a way that we can advise practitioners how to use it most efficiently.
2. We propose that the only way to obtain such understanding is to understand the mechanisms at work in the diversity of actual programming processes.
3. This understanding must first be gained in qualitative form before we can start quantifying. Therefore, we perform such investigation based on Grounded Theory Methodology (GTM), working from rich sets of real-life data (full-length programmer video and screen video of pair programming sessions).
4. We have done so since 2004 for Human+Human pair programming. This research has produced insights that will serve as the basis for understanding the Human+AI pair programming process much more quickly.

2. Call for Recordings

Our group has collected what is (as far as we know) the world's largest research collection of real-life industrial pair programming session recordings. Based on these, our research group is the only one worldwide that is deciphering how pair programming really works and how to do it well (that is, efficiently).

For the new topic of human+AI collaboration, however, our recordings collection is so far much smaller. Also, the technology is changing by leaps and bounds, so adding recordings representing recent technology and development styles is always of interest.

If you are using an AI assistant for professional software development, please donate one or more session recordings to our research!
It does not matter which specific AI you are using, how you use it, which types of software you are developing, how large your codebase or your team are, etc.

2.1 How it works

• We either come to you with an external drive to do the recording for you (we use OBS in portable mode, a stable and trustworthy Open Source software) or you download and run the recording appliance yourself with a few clicks under our guidance during a short video conference. We ask you a few questions about your background, the code base, and the task. Then you start working as usual:
• You work on whichever real development problem(s) is or are on your list that day. You do it in preferably your usual workplace on your usual development machine. Session length can be anywhere from 30 minutes to several hours, as fits your work.
• Your recording is kept confidential; only members of our research group have access to it.
  Our interest in the recording contents is not on the level of the software (of which we get to see glimpses that appear on the screen), only at the level of programmer behavior.
  Should we describe an episode from your session in a research publication, it would be anonymized such that neither you nor your software can be recognized.

2.2 What's in for you?

These recordings are invaluable for realistic and relevant research results on human+AI collaboration; they are obviously useful for us. But is such a recording useful for you as well? It is, indeed!

• If you are interested, we come back the next day to give some feedback on easy-to-spot characteristics of your work style we have seen. In our past pair programming research, pairs have regularly found this feedback very helpful; so much so that they often encouraged colleagues to get recorded as well.
• Once we have research results, we offer to come back and present them to your team(s). Our research perspective is highly practice-oriented: We search for behavior patterns that appear to be particularly useful or particularly problematic so that learning about them allows you to become a more effective user of AI development assistants.

SWTIDSR