The dry run idea is really smart. We've done something similar, where we had Niv Dayan[1] lead a session on Diva[2] (before it won Best Paper at VLDB 2025!). I had worked with him in the past and thought it would be cool to have him present to the group. Having the author in the room completely changed the quality of the discussion. Most of our sessions right now aren't presenter-led, but I'd like to do more of that.
The group is open to anyone at Microsoft and I don't gatekeep. The papers themselves act as a natural filter. If someone finds the material interesting, they attend and keep coming back. If it's not their thing, they self-select out. Over time, it's led to a core group of regular attendees as well as many who will join ad-hoc.
We usually start with quick overall impressions, then go around with a few prompts: "what's something new you learned?", "what didn't you like?", and "what didn't you fully understand?" (every paper has something, whether it's the evaluation methodology or some algorithm detail). That last question tends to drive most of the discussion because people chime in and build on each other's answers. Sometimes you get lucky with domain expertise in the room. For example, when we read "What Every Programmer Should Know About Memory"[1], one of the attendees was a former Intel engineer who spent their career in memory systems. They answered questions the rest of us wouldn't have even known to ask.
It depends on the theme. If we're picking something in a space the group already knows well, like databases, I'll look at "Best Papers" from recent VLDB/ICDE/SIGMOD conferences. If we're exploring a topic most people are unfamiliar with, we'll go with something more foundational instead. For example, we're starting an arc on datacenters (servers, racks, networking, load balancing, power, cooling, failures, etc.), and most attendees don't have deep background there, so I found a book on the topic that we're going to read through[1].
Thank you! I think the biggest factor for us was that most attendees already had some technical baseline. That makes it way easier to pick papers and have productive discussions. A cross-experience group sounds much harder. We occasionally have non-technical people who attend (e.g., designers), but they usually are very eager to learn. The guided series format might have helped in your case, where you pick the topic and sequence upfront so there's less debate about direction each meeting. Honestly, just getting people to show up is hard at first, so the fact that you got it off the ground at all says something.
There are other groups within Microsoft, but they usually follow a presentation format rather than a collaborative discussion. Off the top of my head, Phil Bernstein[1] and Hanuma Kodavalla[2] run great database seminars for invited speakers. I regularly attend and have presented in both forums; Phil's crowd is mostly researchers, while Hanuma's is mostly full of SQL engineers. Different from a small reading group, but still great.
Appreciate the paper link! We like going back to the basics sometimes, so I'll definitely take a look.
Good question. Most people read the paper on their own time, and we meet over lunch. The meetings themselves are just an hour, so it's not a massive time block. I've found that the people who show up are the ones who are genuinely curious and would be reading this stuff anyway (and sometimes just need a commitment/accountability to do it). Having a group gives them a reason to do it on a schedule.
Hi HN, I've been organizing a systems reading group at Microsoft for five years now. I wrote down some takeaways on what worked (and what didn't). I'd love to hear if anyone else has successfully kept an engineering reading group alive at their company, or if you have any favorite systems papers we should add to our list!