Last spring I watched a friend fly to three AI conferences in five weeks. By the third one — a big-name summit in a convention center that smelled like recycled air and cold brew — he fell asleep during a keynote about agents. Not rude-asleep. Exhausted-asleep. His badge lanyard had a little collection of pins from the previous two events, and he'd stopped being able to tell me anything he'd actually learned. He'd absorbed vibes. That's it.

I've been going to these things for about a decade now, and I've come to a slightly heretical conclusion: most conferences are a terrible way to stay current in AI. Not useless — terrible at the specific job people think they're doing. Let me explain, because the fix isn't "stop going." It's "go on purpose."

The thing conferences are actually good at

Here's the counterintuitive part. When I survey my own memory of what I've gotten out of AI conferences over the years, almost none of it is technical knowledge. The talks I could've watched on YouTube two weeks later, usually at 1.5x speed, usually with the boring middle skipped. What I couldn't replicate was the hallway.

A rough estimate from my own calendar: I'd say something like 80% of the genuinely useful outcomes from any given event — a job lead, a collaborator, a "wait, you hit that bug too?" moment, an honest opinion about a tool that the vendor booth would never give me — happened between sessions, not during them. The stage is the least valuable real estate at a conference. The coffee line is the most valuable.

Once you internalize that, your whole strategy changes. You stop treating a conference like a firehose of information to drink from and start treating it like a place to find your people. You go to two talks, not twelve. You spend the rest of the day talking to humans. You leave less tired and remember more.

Why the firehose approach quietly fails

The staying-current problem is real. The field moves fast enough that a model or technique you learned about in a keynote can be half-obsolete by the time your flight lands. So people overcorrect. They try to attend everything, subscribe to everything, and end up in a state I think of as informed paralysis — vaguely aware of forty things, actually competent at none of them.

The dirty secret is that novelty and usefulness are not the same axis. The talk about the brand-new research paper feels current. But you will almost certainly never use it. Meanwhile the unglamorous skill — writing a decent eval, debugging a retrieval pipeline, knowing when not to reach for a large model — is what actually moves your work forward, and nobody's giving a hype-filled keynote about it.

So conference burnout isn't just physical exhaustion. It's the slow erosion of your ability to tell signal from noise, because you've trained yourself to value the shiniest thing in the room.

The lighter system I actually use

What replaced the firehose, for me, is a small routine that costs a few hours a week and one or two events a quarter. Here's the shape of it:

  • Pick one narrow lane and go deep. I don't try to follow "AI." I follow the two or three subareas that touch my actual work. Everything else I let wash past me. Depth in a narrow lane beats a thin coat of everything.
  • Build in public, even badly. Nothing forces you to stay current like shipping a small thing that uses the current tools. I learn more from one weekend project than from ten talks, because the project doesn't let me nod along without understanding.
  • Attend fewer events, but attend them present. I now favor small meetups and hackathons over big summits. A twenty-person meetup where I actually talk to people beats a five-thousand-person conference where I'm a face in a dark room. When I'm choosing what to go to, I browse [Droppa](https://droppa.link) and filter hard — I'd rather find one great local hackathon than pile up flights.
  • Read the source, skip the summary. Secondhand hot takes are where nuance goes to die. When something matters to my lane, I go read the actual docs, the actual paper, the actual repo.

The part people resist most is going to fewer things. It feels like falling behind. It isn't. Presence compounds; attendance doesn't. Six months of showing up to one small meetup and actually knowing the regulars will teach you more about where the field is really going than a dozen conference badges collecting dust in a drawer.

What to do this week

If you're feeling the burnout — or the low-grade anxiety that you're missing something — try this. Cancel one thing you were going to attend out of FOMO. Take the time you'd have spent and build one tiny project with a tool you keep meaning to learn. Then find a single small local event, something you can drive to, and go with the explicit goal of talking to three people you don't know. You can [find events on Droppa](https://droppa.link) and sort for the small stuff near you.

Staying current in AI was never about consuming the most. It's about staying close to real work and real people, and being rested enough to notice what matters. My friend with the pin collection figured this out eventually. Last I heard, he'd skipped the big summer summit entirely, shipped a side project he was actually proud of, and sounded, for the first time in a while, like he was enjoying the field again.