I went to a 500-person AI builders event on a Saturday. I came back with zero new contacts. The event itself was excellent. The format was simply wrong for what I needed, and I should have been able to see that before I went.
This is a small post about a heuristic I now use to pick which events are worth a Saturday. N=1 — I'm reporting one case and the rule I extracted from it. Your mileage will vary.
What Happened
I'd been told by my own weekly strategy reports for several weeks that "external reach" was the bottleneck. The agent ecosystem I work on every day runs well enough. What it does not do is put me in front of any human who could become a collaborator, a customer, or a peer. The Compass had been telling me: show up somewhere in person.
So I picked a high-profile AI/ML builders gathering in Seoul and went. It was a full day of strong talks. The speakers were sharp, the slides were dense, the content was directly relevant to the systems I'm building. I learned things I will use.
And by the end of the day, I had spoken to no one. Not because the people were unfriendly — because the room was structured so that nobody had to.
Why It Failed (And It Wasn't the Event's Fault)
Looking back, the failure was completely predictable from the format. Two structural reasons:
| Format feature | What it produces | What it doesn't |
|---|---|---|
| Lecture-driven schedule (back-to-back talks, short breaks) | High information density. Solid takeaways. A sense of being "in the field." | Time and space to actually talk to a stranger. Breaks are spent in line for coffee or the bathroom. |
| Large attendance (hundreds of people, anonymous audience) | Visibility of the field. Ambient signal that "this matters." | The repeated-encounter effect that lets two strangers recognize each other and try a second conversation. |
Put differently: lecture format optimizes for knowledge transfer from a few to many. It does not optimize for connection formation among the many. Those are different events, and conflating them is on me, not the organizers.
Surface Area vs. Contact Depth
The mistake I want to name is a measurement error: I was tracking surface area ("did I show up to a thing?") as if it were a proxy for contact depth ("did I actually talk to someone for long enough that we'd recognize each other next month?"). They are not the same metric, and one does not produce the other.
From here on I am separating those two columns in my own tracking. Showing up is cheap to count and easy to feel good about. It is not the same as moving the thing it was supposed to move.
The Heuristic I Now Use
Before committing to an event, I now ask three questions. If two answers point the wrong way, I don't go.
- Format — is the schedule lecture-driven or workshop-driven? Workshops force interaction. Lectures allow you to sit anonymously for eight hours and leave.
- Size — is attendance counted in dozens, or hundreds? Smaller rooms make second encounters likely. Large rooms produce a crowd, which is a different thing from a network.
- Structured interaction time — is there an explicit interactive block (small-group exercise, hands-on session, hosted dinner)? If the only "networking time" is the coffee break, the answer is no.
The pattern is straightforward: format determines outcome more than topic does. A 500-person lecture event on the topic I care about will produce less contact depth than a 30-person workshop on a topic I care about somewhat less. The cost-per-real-connection is enormously different.
A Cultural Note
One more thing I had to be honest with myself about. Even in the right format, I'm not the kind of person who hands out business cards or QR codes to strangers and considers that "networking." A previous version of my strategy report had suggested a "method card with QR" approach — a small leave-behind to seed inbound contacts. I tried imagining myself doing it and immediately knew I wouldn't. It's not how trust forms in the contexts I move in.
What works better, for me, is the opposite order: have an actual person-to-person conversation first, find some genuine common ground, and only then offer the artifact — a link, a doc, a tool. The artifact lands differently when it follows a real exchange. Pre-emptive distribution feels like cold outreach in physical form, and cold outreach is exactly the thing I'm trying to avoid by going in person.
So the heuristic has a second layer for me specifically: events where I can credibly have one good conversation are worth more than events where I could theoretically reach a hundred people.
What I'm Doing Next
- Filtering future event invites through the three questions above.
- Looking explicitly for small, recurring rooms — meetups in the 20-50 range, study groups, hosted dinners — rather than headline conferences.
- Tracking contact depth (names I'd recognize next month, follow-ups scheduled) instead of surface area (events attended).
- Holding the leave-behind artifact in reserve, to follow a conversation rather than precede one.
What I Don't Yet Know
I don't know whether the small-room hypothesis will actually produce contacts either. It might also fail, for entirely different reasons (the people I need to meet may simply not attend those rooms). What I know is that the large-lecture format failed predictably, and the failure mode is structural, not bad luck.
This post will get a second observation appended to its Evolution Log the next time I run the experiment — either confirming the heuristic or, more usefully, telling me where it breaks.
Evolution Log
- 2026-05-12 — Initial observation after one 500-person lecture-format event. Surface-area-vs-depth distinction extracted. Heuristic (format / size / structured interaction time) drafted but unproven.