A Letter to Family and Friends: Starting at AfterQuery

Dear family and friends,

I am going to start a new job soon and I would like to tell you about it. For some of you, I am far away, and my work in Silicon Valley can probably feel like another world. I want to bring some of my experiences back to you and bring you along on my journey.

What is changing?

I am starting a new job as an Applied Scientist at AfterQuery, an applied research lab and one of the fastest-growing startups in history. I will be working on their research team.

The biggest change for me is that I will be moving from an engineering role to a research role. This means that the goal of my job is changing. Previously, my goal was to create products or services for customers to use. Now, my job will be to run machine-learning experiments with AfterQuery’s product and find out whether it actually makes models better. This is exciting to me because I have wanted to work closer to the actual training and improvement of models.

AfterQuery

AfterQuery makes training data and reinforcement-learning environments for big AI labs—think OpenAI, Anthropic, and xAI—to train their models on. The company says that it works with every frontier AI research lab. AfterQuery is a new company: it has around 80 people and crossed an astronomical $100+ million annualized revenue run rate about 14 months after it began. Meaning it went from earning $0 to earning at a $100+ million/year pace in about 14 months. 🤯

How does a company experience that much success so soon, and what do they make? To answer that, I need to explain a little about how AI models work using an everyday analogy.

Imagine a perfect tutor for your kid. They would figure out where your kid needs help, create lessons and problems for them to practice, then give them a test to see whether they actually learned. That is roughly what AfterQuery does for AI models. The models are the students, and the big AI labs are the parents paying for the tutor.

AI labs want models to get better at things like coding, using computers, and reasoning. AfterQuery creates training data and environments where models can practice those skills, plus tests to measure whether the practice worked. If a model struggles with coding, for example, it can practice coding tasks and learn from feedback. My job will be to test whether those lessons actually make the model better. It is not enough to make a great course; we need to show that the student learned.

That gets harder as models get more capable. The problems need to meet the model where it is: if they are too easy or too hard, or the test measures the wrong thing, the model may not improve. Figuring out what to teach next—and whether it worked—is the challenge I am excited to work on.

I know this is a lot, and I do not expect everybody to become an AI researcher with me. I mostly wanted to share why this move feels exciting to me. I will be learning a great deal, and I hope to bring some of what I learn back to you along the way.

If you have some questions let me know, I would love to chat about it.