Lead Product Manager, AI Agents & Emerging Products at System1
Founding team at a Y Combinator company in 2021, then three years building AI products full-time. Today I lead AI agents and emerging products at System1. On my own projects I write the code, working with a team of coding agents.
I like building the whole thing: the app, the services behind it, the models, and the data that tells me whether any of it works. I spend most days with coding agents (270B+ Codex tokens in 2026), and the projects below are the ones I'm proudest of, including what broke along the way.
Each one tells the story: the problem, what went wrong along the way, and how I fixed it.
Live on the App Store
A sports research app I architected and run in production: a native iPhone app, live voice, and an assistant that has to find its evidence before it answers, on 168 cloud services.
SwiftUI · StoreKit 2 · Python · TypeScript · Cloud Run
Read the storyWorking system, broadcasting 24/7
Point it at a music library and it plans a set, mixes it the way a DJ would, and grades every blend from the audio before anything ships. It renders in the cloud and broadcasts around the clock.
Python · NumPy and SciPy · PyTorch · beat_this · Essentia
Read the storyOpen source
I wanted a coding agent to handle my App Store releases without being able to break one. It checks a build against 13 rules and only submits when every guard holds.
Python · MCP · Codex plugin · App Store Connect API · ES256
Read the storyChatGPT app and Codex plugin
Most codebases call a bigger OpenAI model than the job needs, and models retire on a schedule. Tierwise works out what each call actually needs, shows its reasoning, and opens a pull request once a person approves.
TypeScript · MCP · OpenAI API · Codex plugin · Cloud Run
Read the storyDeepChamp, the business side
How I know, every day, what is limiting DeepChamp's growth: a warehouse that says when it can't be trusted, a weekly constraint engine, and an agent that can ship a fix but not move money.
BigQuery · Dataform · TypeScript · Python · Cloud Run
Read the storyRelease Guard is open source on GitHub. The others are private, and I'm happy to walk through the code.
Longer essays, and short notes on things that broke in production and how I fixed them.
An AI tool can pass every schema check and still fail the person using it. What I've learned about journeys, tool design, latency, errors and evals from putting AI tools in front of real users.
Four incidents from a production AI app: a provider outage, signed webhooks we rejected ourselves, an enforcement flag that nearly locked out paying customers, and a latency fix hiding a timeout. Each ends with what I do now.
When a model acts on someone's account, identity decisions become product decisions. Least privilege, token handling, account linking, logging and consent, from systems that handle real subscriptions.
How a batched model call through a failing gateway blanked a seven-stage decision panel, and the per-stage fallback and snapshot caching that fixed it.
Subscription access drifted from the App Store. The fix was to treat Apple's signed history as the source of truth and reconcile against it every day.
A flag to end voice access on a lapsed App Store period looked safe. A dry run against production data showed it was not, and traced the problem to an old repair job.
Starting the assistant's turn inside the webhook cut seconds of latency, until the wait itself matched the platform's timeout. The fix: nested timeouts and exactly-once delivery.
Lead Product Manager, AI Agents & Emerging Products
AI Product Engineer & Enterprise Transformation Consultant
Founding team, AI product management
WME, SYB Creative (founder), YMU
Questions about anything here, or something you're building. Email is fastest.
