A real systematic trading desk — the kind that runs $5–50M a year inside a hedge fund — takes eight to twenty people.

Portfolio manager. Quant researchers. Quant developers. Execution traders. Risk. Ops. Performance analysts. Market data engineers. Infra. Compliance.

Ten functional roles. Minimum.

I'm doing all ten.

That's what this is. One Person Trade Desk (OPTD) — the build log of one operator filling every seat on a real desk, with Claude Code as the workbench, deep methodology as the moat, and live capital as the forcing function.

The trading space currently splits two ways.

On one side, retail trading content. Signals, alerts, screenshots, "I made $X this week." Almost none of it has methodology behind it. Most of it is selling the dream of trading rather than the work of it.

On the other side, professional quant. Closed, opaque, expensive, irrelevant if you're not inside it.

There's a third thing nobody is showing publicly: what happens when one operator brings real desk methodology to bear on their own capital, with AI as the bench. Not signals. Not vibes. Not biography. The actual work.

The methodology has always been the moat. AI just changed what one person can build with it.

What costs $5M/year at a hedge fund, built for under $500/month in tools.

I wrote the full breakdown and detailed what each of the ten roles does at a real desk, what fills it at OPTD, why this is possible now in 2026 over on the blog:

→ What is the One Person Trade Desk?

If you only read the table in that post, you'll have the model.

Weekly-ish. Single topic per issue, tied to whatever I shipped, broke, or learned that week.

- Build notes. What I wired up, what failed, what the methodology said next. Real artifacts — commits, charts, log lines, fills.

- The book of biases. Twenty-plus ways backtests lie. One entry at a time. By the end, you'll never trust a clean equity curve again.

- The workbench. How Claude Code and agent workflows actually change R&D. Specific prompts, specific failure modes, what I'd never let an agent do.

- Methodology pieces. Regime, risk, execution, validation. The boring stuff that decides whether anything else works.

- Signals, alerts, picks, or anything that looks like one

- Backtest screenshots without methodology

- "I made $X this week" porn

- AI hype dressed up as edge

- Day-to-day notes: @drewautomates

- Build footage when it ships: youtube.com/@onepersontradedesk

Thanks for being here on day one.

— Drew

P.S. A small private community for builders doing the same thing opens later this year. Subscribers here get first access, no separate list to join.