The Idea
In late 2024, Spencer Mateega and Carlos Georgescu had 48 hours left to file a Y Combinator application and nothing to put in it. No product. No idea. Just a destination: San Francisco. The two had met as teenagers at a Google-run computer science summer programme, stayed close through college, and interned together at Meta before splitting off into finance and engineering. Mateega was finishing at Wharton. Georgescu was studying CS at the University of British Columbia. They got in anyway.
Their first plan was AI agents for finance. It broke almost immediately. The models could write code and summarise filings, but drop them into a nuanced white-collar workflow and they fell apart. Not an architecture problem. Nobody had ever taught them how a professional actually reasons through a hard call. The internet can explain what a cardiologist does. It cannot show you how one thinks.
So they stopped building on top of the models and started building the thing the models were missing. This was the beginning of AfterQuery.
The Execution
- January 2025: Mateega and Georgescu founded AfterQuery with Danny Tang, Mateega's high school co-founder and Wharton roommate, and got into YC's W25 batch off that rushed application. Both founders were still in college, funded by a $500K pre-seed.
- Spring 2025: The pivot. The finance agent idea was scrapped and they started selling labs what the models lacked: expert-generated judgement and step-by-step reasoning across finance, law, medicine and software engineering.
- Through 2025: They assembled a network of nearly 100,000 verified professionals and built custom validation software to screen every dataset into a "Goldilocks" band: hard enough to challenge a state-of-the-art model, not hard enough to stump it.
- 9 April 2026: A $30M Series A at a $300M valuation, led by Altos Ventures with The Raine Group, Y Combinator, BoxGroup, Latitude Capital and angels from OpenAI, Anthropic, DeepMind, Meta and Microsoft. Announced alongside a $100M annual run rate, 14 months after the batch.
- Mid 2026: AfterQuery's work landed in the training stack for Nvidia's Nemotron models, with Mira Murati's Thinking Machines Lab, legal AI startup Legora and Korean lab Motif Technologies also on the customer list.
- July 2026: Mateega said on X that recurring revenue had reached the "hundreds of millions", up from $100M three months earlier.
- August 2026: Unlike most of its US peers, AfterQuery was also training Chinese labs, putting it in the middle of a widening political fight over who gets to buy American reasoning data.
- 1 September 2026: Forbes reported a Series B at a $3.2B valuation, more than 10x the April price in five months. YC partner Gustaf Alströmer called it the fastest run from inception to unicorn in the accelerator's history. The company is profitable and has a lead investor lined up.
- Today: Mateega is 23, Georgescu is 22, and the company they invented after arriving in San Francisco with nothing now serves every frontier lab. Total disclosed funding before this round: roughly $30.5M.
The lesson?
Mateega and Georgescu applied to YC with no product and no idea, then watched the first idea they did have break within weeks. What saved them wasn't a better agent. It was noticing precisely why the agent failed and selling that instead. Every founder building on top of frontier models could see the same gap. Only these two turned the gap into the product. The best AI businesses right now aren't built on what the models can do. They're built on what the models still can't.