Free desk pack · anonymized

Process only. No account size, no venue, no positions, no P&L. Not a performance claim. Download the markdown.

Case study: anatomy of a 10-seat AI agent trading desk

A free desk pack from The Agentic Trader. Anonymized: no accounts, sizes, venues or positions are disclosed. Educational only, not financial advice. Automated trading can lose money; paper-trade first and set hard limits before any agent touches capital.

The setup in one paragraph An active discretionary trader, with no staff, built a desk of ten AI agent "seats" around a single human chair. Each seat has one job, one scorecard and one routine. Agents research, map, check and propose. Execution is gated by an autonomy ladder: a seat only moves from proposing, to paper trading, to live execution with hard limits after it has earned it. The human keeps the chair and the kill switch.

The 10 seats 1. Head trader (orchestrator). Collects every seat's output, resolves conflicts, writes the daily plan and says "no" to anything that fails a check. Scorecard: plan followed vs. deviations; trades rejected that would have broken rules. 2. Research head. Builds and maintains the idea list, catalysts and theses with clear invalidation levels. Scorecard: share of catalysts that were real and correctly dated; thesis hit rate tracked against stated invalidation. 3. Price map. Marks key levels, ranges and structure on the watchlist each session. Scorecard: how often marked levels were respected or cleanly broken vs. noise. 4. On-chain. Watches flows, funding and positioning data for crypto exposure. Scorecard: alerts that preceded a meaningful move vs. false alarms. 5. Expert voice. Summarizes a curated set of outside analysts and commentators into a weekly consensus/contrarian note, with attributions kept internal. Scorecard: signal-to-noise as rated by the human each week. 6. Macro. Calendar, rates, liquidity and cross-asset regime notes. Scorecard: regime calls reviewed monthly; were event risks flagged before they hit? 7. Trader (execution). Turns approved plans into order tickets. Starts at "propose", graduates to paper, then to live within limits. Scorecard: slippage vs. plan, rule breaches (target: zero). 8. Risk. Checks every proposed trade against max loss, position limits, correlation and concentration before it reaches the chair. Can veto. Scorecard: breaches caught; breaches missed (target: zero). 9. Options. Structures, Greeks, IV context and assignment/expiry risk for any options idea. Scorecard: structures that behaved as described across the move. 10. Sim lab. Paper-trades any new rule, seat or automation step before it goes live, and keeps the log. Scorecard: number of changes promoted only after a full sim period.

The flow 1. Overnight / pre-market: macro, on-chain, expert voice and research head publish notes to a shared folder. 2. Price map marks levels for the watchlist. 3. Head trader merges everything into a one-page plan: ideas, levels, invalidation, event risk. 4. Options structures any options ideas; Risk checks every line of the plan. 5. Human chair approves, edits or kills each line. 6. Trader executes at its current autonomy level (propose, paper or live within limits). 7. Post-session: every seat files its scorecard line; head trader writes the review; sim lab logs any proposed change.

Routines - Daily: pre-market plan, intraday risk checks on any new proposal, post-session review and journal. - Weekly: scorecard review per seat; expert-voice consensus note; one improvement chosen and sent to sim lab. - Monthly: promote or demote seats on the autonomy ladder; retire seats whose scorecards don't justify them.

Guardrails - Hard limits (max loss per trade and per day, position size, instruments allowed) live outside the agents and cannot be edited by them. - A kill switch the human controls, tested regularly. - Any limit breach automatically demotes the offending seat back to "propose". - Full logging of every proposal, approval, order and change. - No agent holds withdrawal or transfer permissions. Crypto agents never sign transactions or move funds. - New automation runs in sim lab first, always.

Lessons 1. Start with P&L measurement and plumbing. The desk was only useful once performance was measured honestly and data flowed cleanly between seats. Before that, extra seats just produced more unscored opinions. Build the ledger and the pipes first. 2. Fewer seats, sharper scorecards. Seats with vague jobs drifted. Seats with one job and a number attached got better. If you can't score a seat, merge it or cut it. 3. The orchestrator's job is "no". The head-trader seat added most value by rejecting plans that didn't pass risk or had no invalidation. 4. Automation is earned. Every step up the ladder came after a sim period and a clean scorecard, never because it felt ready.

Start your own (minimum viable desk) Begin with three seats: Risk, Research head and Head trader. Add a scorecard to each. Run two weeks human-in-the-loop. Add a seat only when an existing scorecard tells you where the gap is. Load the free skill packs at theagentictrader.com/skills to give each seat working procedures.

Educational only. Not financial advice. The Agentic Trader teaches how to organize AI agents around your own research and review process. Nothing here is a recommendation to buy, sell, or hold any security or asset. We do not endorse any broker, platform, or asset. AI output can be wrong, so verify everything and make your own decisions. Trading and investing involve risk of loss. Full disclaimer.