How we engineer—Titanium Birch Labs

This section describes how engineering work gets done at Titanium Birch—the repositories, habits, and review loops behind our software—not the investment judgements that software informs. For applications and workflows tied to the investment process, see Investment processes. For how this Labs site is published from private repositories, see About this Labs site.

We build financial tools: data pipelines, portfolio analytics, and workflow automation. Most code lives in Trunk, our monorepo; we also keep standalone repositories where that suits the work better—this Labs site is one example.

Who does what, and where tests-first delivery sits in that picture, is written up once in The operating model. The rest of the articles here each pick up a narrower thread—tooling, testing, skills, delivery mechanics, and how dependable the setup feels in practice.

The articles below are listed in a suggested order, but you can start anywhere—they cross-reference one another.

Articles in this section

The operating model —How a five-person firm runs engineering with AI agents as the primary contributors.

The tech stack —Why an agent-driven engineering team chose each technology—and the trade-offs behind those choices.

How testing works —Sharp requirements, fast suites, and where CI sits—how testing works as an edit loop, not a separate department.

Where we sit on the agentic maturity curve —Dependability signals, rough edges, and what a mature loop looks like when checks are green but the output still misses the brief.

How code gets reviewed —The code review process—what machines handle, what the reviewer agent focuses on, and how comments are usually tagged by priority.

Skills and conventions —How institutional knowledge is encoded so agents can start with shared context, and how that guidance keeps evolving as the team learns.

How a feature ships —The six-phase delivery pipeline—from vague idea to production, run by agents.