I'm a front-end engineer at Catalyst, building AI trading agents for everyday investors. I joined in May 2026 as one of the first engineers, while the company was still in stealth, and helped take the product to its public launch in October alongside a Sequoia-backed $30M raise. Before that I spent four years at Airfoil as its first engineering hire.
Most of my day is still production React and Next.js: the shared components, the motion, the performance budget, the parts that tend to go unowned unless someone picks them up. I care about how a product feels in the hand, so the small things get attention too: the easing on a reveal, the layout shift nobody filed, the tick that lands on a round number. At an early-stage company that also means whatever the launch needs, from sign-in and trading flows to the landing page, and reviewing most of what the rest of the team ships.
I came to this from product design, so I can read a Figma file and I have opinions about it. I don't lead with that. The design side is context for the engineering, not a second job title.
What I've been thinking about more recently is the loop between designers, engineers and the agents both now work through. AI has raised the rate at which explorations show up, and the usual process wasn't built with that in mind. A few things I've been trying in response:
- A second PR lane for prototypes, with lighter CI, so exploring stays cheap
- A
/storiessurface where anyone on the team can put a component or an idea in front of others as code - Component docs generated from the source, so an agent reads one file instead of opening a hundred
There's a longer write-up of the reasoning in the articles. The rest is still being worked out on a real product, with real reviewers, and some of it will probably change.
The other half of a senior role is the less visible part: scoping with leadership, helping decide what the front-end standard is and writing it down, reviewing in a way that nudges fixes into the shared component instead of the call site, and trying new tooling early so the team has a data point before committing. On that last one I've worked through Cursor's Composer, Claude Code and Codex as they arrived, usually a few of them at once across parallel worktrees.

