Co-founderBehind your login
The one the model thinks you'd start something with.
…
A Tale of Two Conferences
We ran AI on the Lot in two versions: one on the physical Amazon lot, and one as a virtual simulation. This is what we found.
First, the reality
2,494guests registered
1,631 simulated before the doors opened
The reality
Who attended
Filmmakers, media execs, founders, and engineers carried the room, with a long tail of partnerships, press, and investors filling out the edges.
By role
By company type
The reality
What you came to do
Most picked several, but it's overwhelmingly about meeting people and learning.
The reality
Comfort with AI tools
This year the crowd grew more advanced.
The reality
Where you came from
Los Angeles was the center of gravity, with the Bay Area and New York also making a showing.
On the lot
The voices in the room.
A few of the standouts who took the stage — directors, founders, and the people building what comes next.
Now, the simulation
We turned each registration into an agentic sim: cast it as an archetype, gave it an avatar, and ran it through the matchmaker to predict who'd click.
1,631Guests cast as sims
The simulation
The idea: Use AI to bring people together
A private social network, aided by simulation data, that would show you attendees, foster discovery, and play matchmaker.
The simulation
Scanning the physical world reveals the simulation
By scanning badges or venue signs, attendees could learn about what happened in the Simulation.
The badges drew prompts from a hopper of questions based on which archetypes were talking to each other.
- Investor × Media ExecWhat's a deal you've each been close to that you still think about?
- Filmmaker × VFX / PostWhat AI-ified craft makes you the most sad that it might get lost?
- Filmmaker × Media ExecWhat's a thing the other side of the table keeps telling you that you suspect is wrong?
- Technologist × InvestorWhat did you each see recently that made you think 'that's the new thing'?
- Founder × Media ExecWhat metric is getting too much respect right now?
- Founder × InvestorWhat are your founder green flags?
- Researcher × InvestorWhat's a thesis in this space you're each waiting to see proof of?
The simulation
How you styled your sims
We built a pipeline of prompts that translated each guest into a persona. Most of you adjusted your picks.
One note re: gender, since the model was informed by everyone’s archetypes, and the archetypes skewed heavily Director/Founder, this led to more masculine clothing styles...a sign of bias within the model, and also our selves.
Archetypes
- The Director29%
- The Founder23%
- The Engineer15%
- The Designer1%
Critters
- arctic fox10%
- panther10%
- tiger9%
- octopus4%
Props
- sticker-covered open laptop8%
- third-wave coffee cup8%
- creased paperback book8%
- luxury retail shopping bags5%
Settings
- anonymous airbnb living room9%
- malibu beach-house7%
- sun-bleached coffee shop patio7%
- in a car stuck in traffic5%
Moods
- sly smile10%
- cooler than you10%
- plotting10%
- mid-thought1%
Gender
- Woman35%
- Man33%
- Neutral32%
Randomly assigned per generation.
The simulation
The conference before the conference
We worked with The Garden in the Machine to build a virtual conference in emergentic.ai, populated by attendees' digital twins. Over a two-day simulation, they moved, talked, planned their days, and attended talks as the schedule played out.
Per attendee, across the two-day simulation
- 12.9spoken messages
- 12.3received messages
- 621.8heard / context messages
- 8.6direct connections
- 140.7people crossed paths with
- 84.5recurring connections
The simulation
A lesson in wifi and load-bearing
The truth is we built this system in 6 weeks. The simulation worked, but the app's Day 1 launch was plagued by bad wifi, onboarding confusion, and general instability.
Avatar runsProfile likesIn-app activity
Two ways to connect
- Profile like (in-app)88%
- Badge scan (in person)12%
People you should meet
The simulation has six people in mind for you.
A co-founder. A mentor. An investor. A doppelganger. A rival. A new bestie. The model drew each one from the room — who they are is waiting behind your login.
Version 1: Post Mortem
What we learned and where we’re headed
Six weeks from conception to execution is not enough time to build a perfectly polished app. The simulation worked; the app’s first day did not. We’re naming the parts that broke and the parts that didn’t, because we want to improve upon them, with help from our community.
- The app lacked stability: slow loads, an imprecise chatbot, a wobbly scan. More lead time would let us onboard, load-test, and shake out the bugs that only surface at scale.
- The raw simulation, while well-executed, was running on limited input information from guests. Better inputs make better outputs. Next time, we ask deeper questions.
- The model skewed heavily toward Director and Founder archetypes, which led to clothing styles that read as masculine across the board. A sign of bias in the model, and in the team that built it. Also, perhaps we ask people before turning them into animals.
- Better wi-fi.
- It’s one thing to simulate a group of people. It’s another to put that simulation to use. The connective tissue between the digital sim and the physical room needs to be richer: more touchpoints, more utility, more my-future-self-is-guiding-me.
- Ideas we want to try next: meeting-spot recommendations on the lot, more gamified discovery, easter eggs around campus, the social network rendered live on the volumetric. (Reach out if any of these are your thing.)
