Today
I run Frontward, an applied AI agency. We take AI agents from pilot to production and help engineering teams build with them. On the side I build products, including Firefox Agent Bridge, Leaguemate, Squawk, Catcher, and Flashies. The rest is below: evals, benchmarks, and apps.
Previously, I worked on Horizon OS at Meta, developer tooling at Qualcomm, Ford’s ML platform, and Airstream’s connected vehicle platform.
Projects
Frontward
An applied AI agency. Agents from pilot to production, and engineers who build with them.
Leaguemate
Dynasty fantasy football strategy and trade recommendations.
Firefox Agent Bridge
A browser agent for Firefox. Claude Code, Codex, and other MCP clients drive background tabs with trusted clicks and typing.
code-bench
A coding benchmark built from my own commits, for picking a Claude model and effort level.
Text-to-image eval
A side-by-side benchmark of text-to-image models.
Squawk
Local dictation for macOS, built for talking to Claude Code.
Catcher
Minimal, local-first markdown notes in the terminal.
Menu bar apps
Claude usage limits, a calendar, painting wallpapers, and fantasy scores in the macOS menu bar. Rust and GPUI.
Flashies
Flash cards to help you memorize anything.
Writing
Oct 1, 2026
I benchmarked Claude on my own Rust commits to pick a daily driver
Opus 5.5 at high effort passed 5 of 8 runs on four hard Rust tasks from my own commits. Sonnet 5.5 at high passed none, and all 32 failed runs said they were done.
Sep 30, 2026
My Firefox browser agent is 1.7x faster than Claude in Chrome
Claude in Chrome only works in Chrome, so I built the Firefox version. Then I benchmarked it until it won all 16 tasks.
Sep 3, 2026
I take notes in the terminal now
I wanted Obsidian's notes without leaving the terminal. I did not want to learn Neovim to get them.
May 5, 2026
Hillclimbing a parity agent with evals
Doing evals the right way takes a lot of human grunt work.