A day at the OpenAI office

OpenAI and Cerebral Valley ran the GPT-6 Astra hackathon in New York yesterday, September 10, at the OpenAI office on Lafayette. I was there from doors at nine until the reception wound down.

I’ll be honest about the first thing that hit me: I didn’t expect OpenAI to have an office in New York. For some reason my brain filed them as a purely San Francisco thing. But there it was, on a normal street in SoHo, and the space was sick. Not corporate-sick, actually nice. I walked out with a hat, which is the correct swag.

The hack

The short version: my team left me.

Not dramatically. We spent the morning debating ideas and kept running into the same wall: every idea we liked was something Astra could just do itself. That’s a real problem with a model this capable. The app has to add something the model can’t. We never got past it, people drifted, and by around 2:30 I was solo with three hours left.

So I built Board Studio: you describe a board game to Astra, it drafts one, and the draft shows up as an editable 3D board in the browser. Pick a component, ask for a revision, apply it, then hit Play and actually play it against a local computer opponent. There’s a print kit and STL export at the end. Three hours from nothing to the one-minute demo I submitted at 5:30.

Fair warning before you open the repo: it was rushed, last-minute, built alone in an afternoon, and mainly for fun. It’s an experiment with Three.js and Astra more than a product. I genuinely didn’t know what Three.js was before yesterday. Now I know it’s the library behind most of the 3D you see in a browser, and I know it well enough to have shipped something with it under a deadline. That’s the hackathon tax paying out.

It wasn’t a finalist. I’m fine with that. Submitting something alone, on the day, against a hard deadline, was the rep I needed. And it got me back into Codex, which I hadn’t touched in a while. More on that below.

The finalist demos at seven were genuinely cool to watch. Five teams, three minutes each, VCs from Thrive, Kearney Jackson, and Khosla asking questions on stage. Astra HQ took first for NYC. The gap between what I shipped and what won was obvious and instructive. The winners weren’t just using Astra as a chat box. They were wiring it into real tools and letting it drive.

The people

This is the part that actually mattered.

Alex Biba, who I connected with early and who I just liked. He’s the kind of person I’d have been friends with in high school, except he’s further along than me. He’s worked at Bloomberg, which is where I want to end up. Really cool guy, and one of the reasons the room felt smaller in the good way.

Adam Mehdi, same. Hackathons are strange because you’re competing and also everyone wants everyone else to ship, and both of them were that.

Andrew, who’s at Meta and did Brown CS ‘23. Instant common ground. We’re following each other on Instagram now, which feels like the right level of commitment for someone you met at a hackathon.

Frank Yu, who was easily the most senior person I talked to all day. He’s spent a long time in games, I think on the Xbox side, and I’m going to follow up with him properly instead of letting it evaporate into a LinkedIn connection.

And Chin. I really vibed with him. He felt like a high school friend I’d already had, someone I could skip the small talk with. He’s in a similar spot to me: early, hungry, wants to lock in and actually achieve something, not just talk about it. Those conversations are rare and I want to keep that one going.

The YC post had this same lesson. The talks and the demos are the reason you go. The people are the reason it was worth going.

Codex and Astra

I hadn’t used Codex seriously in months. Claude Code is where I live. But the judging criteria explicitly rewarded using Astra in development, so I leaned in for the day.

It’s good. Astra with a million tokens of context and near-perfect recall changes what “give the agent the whole repo” means. Codex feels more capable than the last time I tried it, and Astra itself seems really promising, especially for the agentic stuff where you hand it a browser and get out of the way. I’m not switching. But I’m also not going to pretend the gap is what it was.

The unhobbling thing from Boris at Startup School applies here too: new model, strip the crutches, see what it can actually do. I did some of that yesterday. I should do more.

Two weeks in New York

I’ve been here about two weeks now and I want to write this down before it becomes normal.

I feel roughly three times more productive than I did at home. I don’t fully know why. I’m staying with a relative in Flushing for now, with Manhattan as the plan once a job lands, so there’s a built-in reason to get out every day. And leaving is always worth it. I’ve met more people in two weeks than in the previous two months. Living in Flushing also means the food situation is unfair. I’m working through the neighborhood one meal at a time.

Yesterday is a good example of the thing I moved here for. I woke up, took the train to an OpenAI office, built something alone, watched people smarter than me demo, and talked to strangers who turned into contacts and maybe friends. That doesn’t happen from Rhode Island.

I love it here. Not in the honeymoon way, or maybe exactly in the honeymoon way, I can’t tell yet. What I know is that I get to experience stuff like this and be part of my peers and a community, in person, on a random Thursday. That’s the whole reason. Savings are ticking down and I’m still job hunting, and I’d make the same call again.

I’m definitely going to try to be back for next year’s.