Plaid unlocked banking. Stripe unlocked payments. JoyWorksAI unlocks the real world for AI agents.
We ship the structured record of human action - who shows up, where they go, what they actually do. Beta at Yale today; ground-truth for the agentic AI of tomorrow.
Whoever ships the action graph owns the layer that every agentic AI will need. We ship it.
Five Social Opportunities, composed and executed by the system. Each generates a structured outcome record. The agent learns.
Your dinner plans fell through. Four other people within a ten-minute walk are also hungry. You tap yes. The agents pick the restaurant, book the table for five, drop the address in a fresh group chat, and put it on everyone's calendar before you've put your shoes on.
Class is done. You open JoyPulse and see five things within a five-minute walk: a coffee shop with three open seats, a record store listening party starting at four, a pickup volleyball game on the lawn, an open studio at the art building, a brewery tour leaving at five. You tap the brewery. The agents save your spot, drop the meeting point on your phone, and the tour is underway at 5:08.
The JoyTrail your roommate started Monday morning - a coffee, a class she'd been meaning to take, a hike she dragged you on - is at thirty-one stops by now. Eleven people added to it. Tonight is the closing dinner. The agents booked the table at six, sent the directions at seven, and the eight of you whose Sundays overlap are sitting down right now.
You don't know anyone yet. Neither do five other sophomores who said yes to brunch. By Sunday morning the agents have picked the spot, booked the table, sent the invite, and started the group chat. You show up. The rest is already handled.
The hotel left a vinyl-night invitation open - two other guests, a local who knows the scene, tonight only. You tap yes. The agents take the deal, book the cab, hold the table. When you ask what's playing that you've never heard, they check. It's not on Spotify. The record is on your doorstep by the time you're home.
SOURCE · GOOGLE TRENDS · RELATIVE SEARCH INTEREST, WORLDWIDE
The fastest-growing discretionary category is entertainment and travel - up 25.5% year over year and accelerating.
2x in under a year. AI is now daily for ~10% of humanity.
want to convert digital connections into in-person experiences.
The counter-wave is investable, inevitable, and starting now.
The problem is multi-party constraint satisfaction in the real world. Four components solve it.
Always-on ingestion of event APIs, venue data, university calendars, and social signals. Normalize, deduplicate, classify. The unstructured physical world becomes queryable inventory.
Predicting shared enjoyment from behavioral signal and live context. Production-grade ranking at scale. Architecture details under NDA.
Multi-party constraint satisfaction across 2–N users: schedules, preferences, budgets, locations, social graph. Ranked plan synthesis with fallback routing when group dynamics shift.
Booking, coordination, and attendance verification feed back as labeled training data. Every completed event makes the model measurably better at predicting real-world social outcomes.
Filed before agentic AI had a name. Issued before the first consumer concierge shipped. Everything we build runs inside it.
Covers location-based commerce conducted inside conversational interfaces - the exact mechanic every agentic AI concierge is converging toward. Filed two years before the agentic AI wave had a name. Every hospitality transaction we route, and every agent-powered booking we'll ship, lives inside it. A second U.S. patent - the Social Opportunity method itself - is pending.
Every recommendation model trained on web data optimizes for what got clicked. We optimize for what actually happened.
We capture the four signals that matter - the ones the open web has never seen.
Web-trained models see clicks, scrolls, and impressions. Those are weak signals for predicting whether a real-world interaction will work. We verify attendance, measure follow-through, capture peer validation, and track return rate - then feed all of it back as labeled training data.
Because every surface shares one model, campus attendance data improves hospitality recommendations, and hospitality transactions enrich consumer matching. One dataset, compounding across both.
The same coordination model proven across consumer and hospitality. Currently scaling on campus while the hospitality AI is being validated for travel use cases.
"I've spent twenty-five years watching the same pattern: people want to coordinate in person, and every tool we build pulls them further apart. This is the company that fixes it."
JoyWorks matches students for real-world activities: dinners, events, games, hangouts. Attendance tracking, social reputation scoring, real-time intent matching.
AI concierge for lifestyle hotels. Matches guests to in-destination dining, shopping, and entertainment; captures experience revenue hotels currently lose to third parties. Built on our patented conversational-commerce engine.
The AI is only as valuable as the data it trains on. The data is only as good as the users generating it. These are the moves that turn a thesis into a flywheel.
Campus deployments compound into closed social graphs with high interaction density. Hospitality partnerships scale at the brand-group level once the AI is validated for travel; one signed brand then unlocks the portfolio. The wedges feed one dataset. Specifics - playbooks, leverage ratios, account targets - live in the deck.
We're building the action graph for the physical world - the data layer agentic AI will need to act outside the browser. We believe the next great AI company won't be the one that replaces human presence. It'll be the one that restores it.
Do Your Joy isn't our tagline. It's how we work.Each of us is doing our actual joy - the thing we'd still do on a Sunday, that we happened to also be great at. The company is what happens when those practices meet.
25+ years travel & hospitality tech. Patented inventor (US 11,146,913 B2). Previous company served Four Seasons and Hyatt across 100 cities in 40 countries. Yale Alumni Fund Board.
Former Managing Director, AI Prediction & Strategy at Ant Group. Led teams behind TimeLLM and TimeMixer. Founded Bloomberg Labs' AI branch and its GPU computation infrastructure.
10+ years building ML infrastructure at scale. At Meta, orchestrated 3,000+ ranking models at 10,000+ QPS. Prior cloud infrastructure at Amazon. PhD Physics, MS Statistics, UMass Amherst.
Global marketing executive across top-tier agencies. Gen Z brand strategy specialist. Culturally relevant 360° campaigns. Press-worthy creative leadership.
Built PepsiCo's first in-house content studio. Prior: Ogilvy, GREY, FCB, DDB. Portfolio includes Pepsi, Lays, and Claire's. Deep expertise in Gen Z cultural relevance.
That's what every line of code, every campus we onboard, every hospitality pilot serves.
What we believe. Every valuable surface online has been made effortless - search, payments, transit, commerce. The most valuable surface, the one where humans actually live, has been left untouched. We think that's worth a company.
How we hire. We screen for one thing first: do you actually want to build this? Not a similar version. This one. If "training agents on what actually happened" is the most interesting phrase you've read this month, we should talk. The rest we figure out together.
Specifically. If you've shipped ranking, matching, or agentic systems in production and want to work on a problem where the training data is the moat - that's us. What you'd work on: agentic orchestration of real-world social outcomes. Multi-party constraint satisfaction. Training models on outcomes the web has never seen.
And one more thing. Spenser is our Chief Joy Officer. While the rest of us build the platform, he reminds us why. Do Your Joy - preferably outside.
A new kind of social object.
Patent Pending · The Social Opportunity MethodWe call it a Social Opportunity.
For thirty years, social plans required one person to propose and everyone else to respond. That loop is brittle. Group chats collapse. Calendars don't align. Eighty percent of social intent leaks before anything happens.
Our agents observe convergence - what you like, who you know, where you are, what your week holds, what you've done before, who you did it with, whether you came back - and compose the plan. All four dimensions, bound as one. It arrives in your feed fully formed.
You tap yes. It activates.
Patent pending on the Social Opportunity method itself. Filed under: The Social Opportunity Method. Substance available to qualified partners under NDA.
The Social Opportunity
Method