I'm a software engineer at Ranger, working on how we test the work of coding agents and build confidence in their results.
One example is a CLI that verifies your coding agent's results in the browser while it works. Along the way, I stumbled upon (and wrote about) the fact that agents can accomplish (some!) browser tasks over twice as quickly with the Playwright MCP as the Playwright CLI.
My interest specifically lies in how to integrate agents into robust, reliable applications. The probabilistic output of LLMs allows them to solve a huge range of problems that traditional software engineering doesn't have an answer for, but it also adds the risk of both subpar (the boring failure) and dangerous (the exciting failure) outcomes every time it's used.
We know that sandboxes, guardrails, evals, observability, focused tools, and a carefully crafted context are best practices, but they're often skipped in favor of shipping faster and retaining more flexibility. I want to build systems that help developers trend closer to best practices without losing velocity or flexibility via platforms that make it easy, and a better understanding of agent tool ergonomics.
Until late 2025, I was at AWS, where I worked on migrations for the open source OpenSearch project. In the years before Amazon, I was at a couple different medical/health startups: Alan Bainbridge Health and Butterfly Network.
I've lived in New York, Salt Lake City, Philadelphia, and Paris, but home is now Denver, Colorado.
Try email: [anything]@mikayla.codes
