I build AI products and companies. Currently based in NYC.
Some things about me:
- I’m an AI Product Manager at Microsoft working on Copilot Studio (agentic workflows, orchestration, enterprise scale).
- Previously, I founded Dishie (Computer vision tech for dishwashing) and Arbr (software engineering agents). I like the 0-to-1 phase, but I love the 1-to-1B phase where data comes into play. I enjoy problems the most when they can only be solved optimally by expanding the solution space and combining business and technology decisions.
- I studied engineering at Carnegie Mellon ('17) and got my MBA at Yale ('24).
- I love to think in systems, and I advise early-stage teams on AI architecture, agentic systems, and technical org design.
- In my spare time, I'm finishing writing a speculative fiction novel, and hopefully I'll have more to say about that soon.
- When not at a keyboard, you can probably find me playing tennis, reading tech S-1 filings for fun, or catching a live set from the desert to the coast.
Some things I believe:
- Anything that can be measured can be solved.
- Intelligence in every human capacity is becoming a commodity. The areas where people believe AI will never be capable are a symptom of hubris.
- Companies need to be built around AI, rather than AI being retrofitted to work with how companies are currently structured.
- Quality communication is dense. Most AI-generated documents add noise to information rather than efficiency. Every line should either add or remove information.
- Code does not matter; only outcomes matter. Examples beat rules.
- Social simulation with agents will become as important as foundational models. The two spaces will converge because social simulation is inherently a form of eval.
- System 1 models are brilliant (TypeSafe's Jev and Cloudflare's Clairic are two of my favorite releases). Fast, sub-100ms probabilistic routing unlocks product surfaces that giant reasoning models can’t touch.
- Data will always be a moat for niche industries. There is a permanent compute-cost advantage for companies that build and fine-tune smaller models on their own proprietary datasets.
Things I’m thinking about:
- Inverted data & attention markets: Turning products people previously paid for into things they are paid to use in exchange for behavioral data (backed by eyes-off auditing and collective bargaining pipes).
- The post-code bottleneck: Now that agentic coding makes building software trivial, the bottleneck shifts to dynamic storyboarding, automated A/B preference classifiers, and synthetic public-perception testing before customer discovery.
- Prediction markets as information engines: Using collective intelligence and RAG-backed networks to arbitrage sentiment and price economic or clinical outcomes.
- AI-native traditional businesses: Taking high-margin, labor-heavy legacy sectors and rebuilding them from the ground up where organizational knowledge is structured specifically for agents.